Publications

@article{2025_5J_SafeBalance,
  title={Safe and Balanced: A Framework for Constrained Multi-Objective Reinforcement Learning},
  author={Gu*, Shangding and Sel*, Bilgehan and Ding*, Yuhao and Wang, Lu and Lin, Qingwei and Knoll, Alois and Jin, Ming},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2025},
  url_arXiv={https://arxiv.org/abs/2405.16390}
}

@inproceedings{2025_5C_Antifragility,
  title={Position: AI Safety Must Embrace an Antifragile Perspective},
  author={Jin, Ming and Lee, Hyunin },
  booktitle={The International Conference on Machine Learning (ICML)},
  year={2025},
  url_pdf={AntifragileAISafety.pdf}
}

@inproceedings{2025_4C_LLMRL,
  title={LLMs Can Plan Faster Only If We Let Them},
  author={Sel, Bilgehan and Ramakrishnan, Naren and Huang, Lifu and Jia, Ruoxi and  Jin, Ming },
  booktitle={The International Conference on Machine Learning (ICML)},
  year={2025},
  url_pdf={LLM-RL-TokenLength-ICML.pdf}
}

@inproceedings{2025_4C_LLMOverRefuse,
  title={Just Enough Shifts: Mitigating Over-Refusal in Aligned Language Models with Targeted Representation Fine-Tuning},
  author={Dabas, Mahavir and Chen, Si and Fleming, Charles and Jin,  Ming and Jia, Ruoxi },
  booktitle={The International Conference on Machine Learning (ICML)},
  year={2025}
}

@inproceedings{2025_4C_HypoBlack,
  title={A Hypothesis on Black Swan in Unchanging Environments},
  author={Lee, Hyunin and Park, Chanwoo and Abel, David and Jin, Ming },
  booktitle={The Thirteenth International Conference on Learning Representations (ICLR)},
  year={2025},
  url_arXiv={https://arxiv.org/abs/2407.18422}
}

@inproceedings{2025_3C_RobustGym,
  title={Robust Gymnasium: A Unified Modular Benchmark for Robust Reinforcement Learning},
  author={Gu, Shangding and Shi, Laixi and Wen, Muning and Jin, Ming and Mazumdar, Eric and Chi, Yuejie and Wierman, Adam and  Spanos, Costas },
  booktitle={The Thirteenth International Conference on Learning Representations (ICLR)},
  year={2025}
}

@inproceedings{2025_3C_LLMScience,
  title={LLMs Tackle Meta-Analysis: Automating Scientific Hypothesis Generation with Statistical Rigor},
  author={Lin, Tung-Wei and Yang, Runing and ul Abdeen, Zain and Sangiovanni-Vincentelli, Alberto and Huang, Haibo and Jin, Ming},
  booktitle={2nd AI4Research Workshop: Towards a Knowledge-grounded Scientific Research Lifecycle},
  year={2025}
}

@inproceedings{2025_3C_DIPT,
  title={DiPT: Enhancing LLM Reasoning through Diversified Perspective-Taking},
  author={Just, Hoang Anh and Dabas, Mahavir and Huang, Lifu and Jin, Ming and Jia, Ruoxi},
  booktitle={The 2025 Annual Conference of the Nations of the Americas Chapter of the ACL (NAACL)},
  year={2025}
}

@inproceedings{2025_2C_IPFL,
  title={IP-FL: Incentive-driven Personalization in Federated Learning},
  author={Khan, Ahmad and Wang, Xinran and Le, Qi and ul Abdeen, Zain and Khan, Azal andAli, Haider and Jin, Ming and Ding, Jie and Butt, Ali and Anwar, Ali},
  booktitle={39th IEEE International Parallel and Distributed Processing Symposium (IPDPS)},
  year={2025}
}

@inproceedings{2025_2C_DERMS,
  title={Defense against Joint Poisoning and Evasion Attacks: A Case Study of DERMS},
  author={ul Abdeen, Zain and Tawaha, Ahmad and Roy, Padmaksha and Jia, Ruoxi and Freeman,  Laura and Beling, Peter and Liu, Chen-Ching and Sangiovanni-Vincentelli, Alberto and Jin, Ming},
  booktitle={The AAAI-25 Workshop on Artificial Intelligence for Cyber Security (AICS)},
  note = {<font style="color:#FF0000">(oral presentation)</font>},
  year={2025}
}

@inproceedings{2025_1C_LLMPlan,
  title={LLMs Can Plan Only If We Tell Them},
  author={Sel, Bilgehan and Jia, Ruoxi and  Jin, Ming },
  booktitle={The Thirteenth International Conference on Learning Representations (ICLR)},
  year={2025}
}

@inproceedings{2024_5C_skig,
title={Skin-in-the-Game: Decision Making via Multi-Stakeholder Alignment in LLMs},
author={Bilgehan Sel* and Priya Shanmugasundaram* and Mohammad Kachuee and Kun Zhou and Ruoxi Jia and Ming Jin},
booktitle={The 62nd Annual Meeting of the Association for Computational Linguistics (ACL)},
pages={},
year={2024},
  keywords = {LLMs, Alignment},
    url_pdf = {SKIG_2024.pdf}
}

@inproceedings{2024_5C_ToA,
  title={Algorithm of Thoughts: Enhancing Exploration of Ideas
in Large Language Models},
  author={Bilgehan Sel and Ahmad Al-Tawaha and Vanshaj Khattar and Ruoxi Jia  and Ming Jin},
  booktitle={International Conference on Machine Learning (ICML)},
  pages={},
  year={2024},
  url_pdf = {AoT_ICML24.pdf},
  url_arXiv = {https://arxiv.org/pdf/2308.10379.pdf},
  keywords = {Machine Learning, Large Language Model}
}

@inproceedings{2024_5C_PRL,
  title={Pausing Policy Learning in Real-world Reinforcement Learning},
  author={Hyunin Lee and Ming Jin and Javad Lavaei and Somayeh Sojoudi},
  booktitle={International Conference on Machine Learning (ICML)},
  pages={},
  year={2024},
  url_pdf = {Pausing_RL_ICML24.pdf},
  keywords = {Reinforcement Learning},
  note = {<font style="color:#FF0000">(oral presentation)</font>}
}

@article{2024_4J_Teams,
  title={TeaMs-RL: Teaching LLMs to Generate Better Instruction Datasets via Reinforcement Learning},
  author={Gu, Shangding and Knoll, Alois and Jin, Ming},
  journal ={Transactions on Machine Learning Research},
  year={2024},
url_arXiv = {https://arxiv.org/abs/2403.08694}
}

@inproceedings{2024_4C_UQLLM,
  title={Can We Trust the Performance Evaluation of Uncertainty Estimation Methods in Text Summarization?},
  author={He, Jianfeng and Yang, Runing and Yu, Linlin and Li, Changbin and Jia, Ruoxi and Chen, Feng and Jin, Ming and Lu, Chang-Tien},
  booktitle={Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing},
  pages={16514--16575},
  year={2024}
}

@inproceedings{2024_4C_SafeSample,
title={Enhancing Efficiency of Safe Reinforcement Learning via Sample Manipulation},
author={Shangding Gu and Laixi Shi and Yuhao Ding and Alois Knoll and Costas Spanos and Adam Wierman and Ming Jin},
booktitle={The Thirty-eighth Annual Conference on Neural Information Processing Systems
 (NeurIPS)},
pages={},
year={2024},
  keywords = {Safe RL},
    url_arXiv = {https://arxiv.org/pdf/2405.20860}
}

@inproceedings{2024_4C_PCRPO,
  title={Balance Reward and Safety Optimization for Safe Reinforcement Learning: A
Perspective of Gradient Manipulation},
  author={Shangding Gu and Bilgehan Sel and Yuhao Ding and Lu Wang and Qingwei Lin and Ming Jin* and Alois Knoll*},
  booktitle={AAAI Conference on Artificial Intelligence (AAAI)},
  note = {<font style="color:#FF0000">(oral presentation)</font>},
  pages={},
  year={2024},
  keywords = {Machine Learning, Reinforcement Learning}
}

@inproceedings{2024_4C_FairNash,
title={Fairness-Aware Meta-Learning via Nash Bargaining},
author={Yi Zeng and Xuelin Yang and Li Chen and Cristian Canton Ferrer and Ming Jin and Michael Jordan and Ruoxi Jia},
booktitle={The Thirty-eighth Annual Conference on Neural Information Processing Systems
 (NeurIPS)},
pages={},
year={2024},
  keywords = {Meta learning, Game theory},
    url_arXiv = {https://arxiv.org/abs/2406.07029}
}

@inproceedings{2024_4C_Alignment,
title={Boosting Alignment for Post-Unlearning Text-to-Image Generative Models},
author={Myeongseob Ko and Henry Li and Zhun Wang and Jonathan Patsenker and Jiachen T. Wang and Qinbin Li and Ming Jin and Dawn Song and Ruoxi Jia},
booktitle={The Thirty-eighth Annual Conference on Neural Information Processing Systems
 (NeurIPS)},
pages={},
year={2024},
  keywords = {Alignment},
  url_pdf = {unlearning.pdf}
}

@article{2024_3J_MLSER,
author = {Srivastava, Sonali and Wang, Wei and Zhou, Wei and Jin, Ming and Vikesland, Peter J.},
title = {Machine Learning-Assisted Surface-Enhanced Raman Spectroscopy Detection for Environmental Applications: A Review},
journal = {Environmental Science \& Technology},
volume = {0},
number = {0},
pages = {null},
year = {2024},
doi = {10.1021/acs.est.4c06737},
    note ={PMID: 39537382},
eprint = { 
        https://doi.org/10.1021/acs.est.4c06737
},
url_pdf = {MLSER-Review.pdf}
}

@inproceedings{2024_3C_mirror,
  title={The Mirrored Influence Hypothesis: Efficient Data Influence Estimation by Harnessing Forward Passes},
  author={Myeongseob Ko and Feiyang Kang and Weiyan Shi and Ming Jin and Zhou Yu and Ruoxi Jia},
  booktitle={Conference on Computer Vision and Pattern Recognition (CVPR)},
  pages={},
  year={2024},
  url_arXiv = {https://arxiv.org/abs/2402.08922v1},
  keywords = {Cybersecurity, Data Valuation, Machine Learning}
}

@inproceedings{2024_3C_metaCLR,
title={Enhancing Distribution System Resilience: A First-Order Meta-RL algorithm for Critical Load Restoration},
author={Zain ul Abdeen and Xiangyu Zhang and Waris Gil and Ming Jin},
booktitle={IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm)},
pages={},
year={2024},
  keywords = {Power systems, AI},
    url_pdf = {MetaCLR-2024.pdf}
}

@inproceedings{2024_3C_causalprompt,
title={CausalPrompt: Enhancing LLMs with weakly supervised causal reasoning for robust performance in non-language tasks},
author={Tung-Wei Lin* and Vanshaj Khattar* and Yuxuan Huang* and Junho Hong and Ruoxi Jia and Chen-Ching Liu and Alberto Sangiovanni-Vincentelli and Ming Jin},
booktitle={ICLR Workshop: Tackling Climate Change with Machine Learning},
pages={},
year={2024},
  keywords = {Power and energy systems, Machine Learning},
    url_pdf = {CausalPrompt2024.pdf}
}

@inproceedings{2024_3C_antifrgility,
  title={Preparing for Black Swans: The Antifragility Imperative for Machine Learning},
  author={Jin, Ming },
  year={2024},
  url_arXiv={https://arxiv.org/html/2405.11397v1}
}

@inproceedings{2024_3C_ZDA,
title={Zero-day Attack Detection in Digital Substations using In-Context Learning},
author={Faizan Manzoor and Vanshaj Khattar and Chen-Ching Liu and Ming Jin},
booktitle={IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm)},
pages={},
year={2024},
  keywords = {Power systems, AI},
    url_pdf = {ZDA-detect-2024.pdf}
}

@inproceedings{2024_3C_UQinternal,
title={InternalInspector I2: Robust Confidence Estimation in LLMs through Internal States},
author={Mohammad Beigi and Ying Shen and Runing Yang and Zihao Lin and Qifan Wang and Ankith Mohan and Jianfeng He and Ming Jin and Chang-Tien Lu and Lifu Huang},
booktitle={The 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
pages={},
year={2024},
  keywords = {LLM, Uncertainty quantification},
    url_arXiv = {https://arxiv.org/pdf/2406.12053}
}

@inproceedings{2024_3C_OfflineIAC,
  title={Optimization Solution Functions as Deterministic Policies for Offline Reinforcement Learning},
  author={Vanshaj Khattar and Ming Jin},
  booktitle={American Control Conference (ACC)},
  pages={},
  year={2024},
  keywords = {Machine Learning, Reinforcement Learning, Optimization}
}

@inproceedings{2024_3C_Chatenergy,
  title={Democratizing Energy Management with LLM-Assisted Optimization
Autoformalism},
  author={Ming Jin and Bilgehan Sel and Fnu Hardeep and Wotao Yin},
  booktitle={IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm)},
  pages={},
  year={2024},
  url_pdf = {ChatEnergy.pdf},
  keywords = {Machine Learning, Energy}
}

@inproceedings{2024_2C_oodcyber,
  title={Latent Space Correlation-Aware Autoencoders for Anomaly Detection in Skewed Data},
  author={Padmaksha Roy and Himanshu Singhal and Tim O'Shea and Ming Jin},
  booktitle={Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD)},
  pages={},
  year={2024},
  keywords = {Cybersecurity, Machine Learning}
}

@inproceedings{2024_2C_Online,
  title={Does online gradient descent (and variants) still work with biased gradient and variance?},
  author={Ahmad Al-Tawaha and Ming Jin},
  booktitle={American Control Conference (ACC)},
  pages={},
  year={2024},
  keywords = {Machine Learning, Online Learning, Optimization}
}

@inproceedings{2024_1C_cybersim,
  title={Simulation and Analysis of Cyber Attacks on Power and Energy Systems},
  author={Zachary A. Ruttle and Baza Somda and Chen-Ching Liu and Ming Jin},
  booktitle={The 2024 Conference on Innovative Smart Grid Technologies, North America (ISGT NA 2024)},
  pages={},
  year={2024},
  keywords = {Cybersecurity, Power systems}
}

@inproceedings{2023_5C_MetaSRL,
  title={A CMDP-within-online framework for Meta-Safe Reinforcement Learning},
  author={Vanshaj Khattar and Yuhao Ding and Bilgehan Sel and Javad Lavaei and  Ming Jin},
  booktitle={International Conference on Learning Representations (ICLR)},
  note = {<font style="color:#FF0000">(spotlight presentation)</font>},
  pages={},
  year={2023},
  url_pdf={MetaSRL.pdf},
  url_openreview = {https://openreview.net/forum?id=mbxz9Cjehr},
  keywords = {Reinforcement learning, Machine Learning},
  abstract={Meta-reinforcement learning has widely been used as a learning-to-learn framework to solve unseen tasks with limited experience. However, the aspect of constraint violations has not been adequately addressed in the existing works, making their application restricted in real-world settings. In this paper, we study the problem of meta-safe reinforcement learning (meta-SRL) through the CMDP-within-online framework. We obtain task-averaged regret guarantees for the reward maximization (optimality gap) and constraint violations using gradient-based meta-learning and show that the task-averaged optimality gap and constraint satisfaction improve with task-similarity in the static environment, or task-relatedness in the changing environment. Several technical challenges arise when making this framework practical while still having strong theoretical guarantees. To address these challenges, we propose a meta-algorithm that performs inexact online learning on the upper bounds of intra-task optimality gap and constraint violations estimated by off-policy stationary distribution corrections. Furthermore, we enable the learning rates to be adapted for every task and extend our approach to settings with the dynamically changing task environments. Finally, experiments are conducted to demonstrate the effectiveness of our approach. The proposed theoretical framework is the first to handle the nonconvexity and stochastic nature of within-task CMDPs, while exploiting inter-task dependency for multi-task safe learning.  },
}

Meta-reinforcement learning has widely been used as a learning-to-learn framework to solve unseen tasks with limited experience. However, the aspect of constraint violations has not been adequately addressed in the existing works, making their application restricted in real-world settings. In this paper, we study the problem of meta-safe reinforcement learning (meta-SRL) through the CMDP-within-online framework. We obtain task-averaged regret guarantees for the reward maximization (optimality gap) and constraint violations using gradient-based meta-learning and show that the task-averaged optimality gap and constraint satisfaction improve with task-similarity in the static environment, or task-relatedness in the changing environment. Several technical challenges arise when making this framework practical while still having strong theoretical guarantees. To address these challenges, we propose a meta-algorithm that performs inexact online learning on the upper bounds of intra-task optimality gap and constraint violations estimated by off-policy stationary distribution corrections. Furthermore, we enable the learning rates to be adapted for every task and extend our approach to settings with the dynamically changing task environments. Finally, experiments are conducted to demonstrate the effectiveness of our approach. The proposed theoretical framework is the first to handle the nonconvexity and stochastic nature of within-task CMDPs, while exploiting inter-task dependency for multi-task safe learning.

@inproceedings{2023_4C_TempoRL,
  title={Tempo Adaption in Non-stationary Reinforcement Learning},
  author={Hyunin Lee and Yuhao Ding and Jongmin Lee and Ming Jin and Javad Lavaei and Somayeh Sojoudi},
  booktitle={Thirty-seventh Conference on Neural Information Processing Systems (NeurIPS)},
  pages={},
  year={2023},
  url_pdf = {TempoRL23.pdf},
  url_arXiv = {https://arxiv.org/abs/2309.14989},
  keywords = {Machine Learning, Reinforcement Learning}
}

@inproceedings{2023_4C_Sol,
  title={On Solution Functions of Optimization: Universal Approximation and Covering Number Bounds},
  author={Ming Jin and Vanshaj Khattar and Harshal Kaushik and Bilgehan Sel and Ruoxi Jia},
  booktitle={AAAI Conference on Artificial Intelligence (AAAI)},
  note = {<font style="color:#FF0000">(oral presentation)</font>},
  pages={},
  year={2023},
  url_arXiv={https://arxiv.org/abs/2212.01314},
  url_pdf={Sol_function_complexity.pdf},
  keywords = {Optimization,  Machine Learning},
  abstract={We study the expressibility and learnability of solution functions of convex optimization and their multi-layer architectural extension. The main results are: (1) the class of solution functions of linear programming (LP) and quadratic programming (QP) is a universal approximant for the $C^k$ smooth model class or some restricted Sobolev space, and we characterize the rate-distortion, (2) the approximation power is investigated through a viewpoint of regression error, where information about the target function is provided in terms of data observations, (3) compositionality in the form of deep architecture with optimization as a layer is shown to reconstruct some basic functions used in numerical analysis without error, which implies that (4) a substantial reduction in rate-distortion can be achieved with a universal network architecture, and (5) we discuss the statistical bounds of empirical covering numbers for LP/QP, as well as a generic optimization problem (possibly nonconvex) by exploiting tame geometry. Our results provide the **first rigorous analysis of the approximation and learning-theoretic properties of solution functions** with implications for algorithmic design and performance guarantees.  },
}

We study the expressibility and learnability of solution functions of convex optimization and their multi-layer architectural extension. The main results are: (1) the class of solution functions of linear programming (LP) and quadratic programming (QP) is a universal approximant for the $C^k$ smooth model class or some restricted Sobolev space, and we characterize the rate-distortion, (2) the approximation power is investigated through a viewpoint of regression error, where information about the target function is provided in terms of data observations, (3) compositionality in the form of deep architecture with optimization as a layer is shown to reconstruct some basic functions used in numerical analysis without error, which implies that (4) a substantial reduction in rate-distortion can be achieved with a universal network architecture, and (5) we discuss the statistical bounds of empirical covering numbers for LP/QP, as well as a generic optimization problem (possibly nonconvex) by exploiting tame geometry. Our results provide the **first rigorous analysis of the approximation and learning-theoretic properties of solution functions** with implications for algorithmic design and performance guarantees.

@inproceedings{2023_4C_MetaSift,
  title={How to Sift Out a Clean Data Subset in the Presence of Data Poisoning?},
  author={Yi Zeng* and Minzhou Pan* and Himanshu Jahagirdar and Ming Jin and Lingjuan Lyu and Ruoxi Jia},
  booktitle={USENIX Security},
  pages={},
  year={2023},
  url_arXiv = {https://arxiv.org/abs/2210.06516},
  keywords = {Machine Learning},
  abstract={Given the volume of data needed to train modern machine learning models, external suppliers are increasingly used. However, incorporating external data poses data poisoning risks, wherein attackers manipulate their data to degrade model utility or integrity. Most poisoning defenses presume access to a set of clean data (or base set). While this assumption has been taken for granted, given the fast-growing research on stealthy poisoning attacks, a question arises: can defenders really identify a clean subset within a contaminated dataset to support defenses? This paper starts by examining the impact of poisoned samples on defenses when they are mistakenly mixed into the base set. We analyze five defenses and find that their performance deteriorates dramatically with less than 1% poisoned points in the base set. These findings suggest that sifting out a base set with high precision is key to these defenses' performance. Motivated by these observations, we study how precise existing automated tools and human inspection are at identifying clean data in the presence of data poisoning. Unfortunately, neither effort achieves the precision needed. Worse yet, many of the outcomes are worse than random selection. In addition to uncovering the challenge, we propose a practical countermeasure, Meta-Sift. Our method is based on the insight that existing attacks' poisoned samples shifts from clean data distributions. Hence, training on the clean portion of a dataset and testing on the corrupted portion will result in high prediction loss. Leveraging the insight, we formulate a bilevel optimization to identify clean data and further introduce a suite of techniques to improve efficiency and precision. Our evaluation shows that Meta-Sift can sift a clean base set with 100% precision under a wide range of poisoning attacks. The selected base set is large enough to give rise to successful defenses.  },
}

Given the volume of data needed to train modern machine learning models, external suppliers are increasingly used. However, incorporating external data poses data poisoning risks, wherein attackers manipulate their data to degrade model utility or integrity. Most poisoning defenses presume access to a set of clean data (or base set). While this assumption has been taken for granted, given the fast-growing research on stealthy poisoning attacks, a question arises: can defenders really identify a clean subset within a contaminated dataset to support defenses? This paper starts by examining the impact of poisoned samples on defenses when they are mistakenly mixed into the base set. We analyze five defenses and find that their performance deteriorates dramatically with less than 1% poisoned points in the base set. These findings suggest that sifting out a base set with high precision is key to these defenses' performance. Motivated by these observations, we study how precise existing automated tools and human inspection are at identifying clean data in the presence of data poisoning. Unfortunately, neither effort achieves the precision needed. Worse yet, many of the outcomes are worse than random selection. In addition to uncovering the challenge, we propose a practical countermeasure, Meta-Sift. Our method is based on the insight that existing attacks' poisoned samples shifts from clean data distributions. Hence, training on the clean portion of a dataset and testing on the corrupted portion will result in high prediction loss. Leveraging the insight, we formulate a bilevel optimization to identify clean data and further introduce a suite of techniques to improve efficiency and precision. Our evaluation shows that Meta-Sift can sift a clean base set with 100% precision under a wide range of poisoning attacks. The selected base set is large enough to give rise to successful defenses.

@inproceedings{2023_4C_LAVA,
  title={LAVA: Data Valuation without Pre-Specified Learning Algorithms},
  author={Hoang Anh Just* and Feiyang Kang* and Tianhao Wang and Yi Zeng and Myeongseob Ko, Ming Jin and Ruoxi Jia},
  booktitle={International Conference on Learning Representations (ICLR)},
  note = {<font style="color:#FF0000">(spotlight presentation)</font>},
  pages={},
  year={2023},
  url_openreview={https://openreview.net/forum?id=JJuP86nBl4q},
  keywords = {Machine Learning},
  abstract={Traditionally, data valuation is posed as a problem of equitably splitting the validation performance of a learning algorithm among the training data. As a result, the calculated data values depend on many design choices of the underlying learning algorithm. However, this dependence is undesirable for many use cases of data valuation, such as setting priorities over different data sources in a data acquisition process and informing pricing mechanisms in a data marketplace. In these scenarios, data needs to be valued before the actual analysis and the choice of the learning algorithm is still undetermined then. Another side-effect of the dependence is that to assess the value of individual points, one needs to re-run the learning algorithm with and without a point, which incurs a large computation burden. This work leapfrogs over the current limits of data valuation methods by introducing a new framework that can value training data in a way that is oblivious to the downstream learning algorithm. Our main results are as follows. We develop a proxy for the validation performance associated with a training set based on a non-conventional between the training and the validation set. We show that the distance characterizes the upper bound of the validation performance for any given model under certain Lipschitz conditions. We develop a novel method to value individual data based on the sensitivity analysis of the Wasserstein distance. Importantly, these values can be directly obtained from the output of off-the-shelf optimization solvers once the Wasserstein distance is computed. We evaluate our new data valuation framework over various use cases related to detecting low-quality data and show that, surprisingly, the learning-agnostic feature of our framework enables a over the state-of-the-art performance while being Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity. },
}

Traditionally, data valuation is posed as a problem of equitably splitting the validation performance of a learning algorithm among the training data. As a result, the calculated data values depend on many design choices of the underlying learning algorithm. However, this dependence is undesirable for many use cases of data valuation, such as setting priorities over different data sources in a data acquisition process and informing pricing mechanisms in a data marketplace. In these scenarios, data needs to be valued before the actual analysis and the choice of the learning algorithm is still undetermined then. Another side-effect of the dependence is that to assess the value of individual points, one needs to re-run the learning algorithm with and without a point, which incurs a large computation burden. This work leapfrogs over the current limits of data valuation methods by introducing a new framework that can value training data in a way that is oblivious to the downstream learning algorithm. Our main results are as follows. We develop a proxy for the validation performance associated with a training set based on a non-conventional between the training and the validation set. We show that the distance characterizes the upper bound of the validation performance for any given model under certain Lipschitz conditions. We develop a novel method to value individual data based on the sensitivity analysis of the Wasserstein distance. Importantly, these values can be directly obtained from the output of off-the-shelf optimization solvers once the Wasserstein distance is computed. We evaluate our new data valuation framework over various use cases related to detecting low-quality data and show that, surprisingly, the learning-agnostic feature of our framework enables a over the state-of-the-art performance while being Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

@inproceedings{2023_4C_CL,
  title={Winning the CityLearn Challenge: Adaptive Optimization with Evolutionary Search under Trajectory-based Guidance},
  author={Vanshaj Khattar and  Ming Jin},
  booktitle={AAAI Conference on Artificial Intelligence (AAAI) AI for Social Impact Track},
  pages={},
  year={2023},
  url_arXiv={https://arxiv.org/abs/2212.01939},
  url_pdf={ESGuidance_ZOiRL.pdf},
  keywords = {Optimization, Power system, Reinforcement learning, Machine Learning},
  abstract={Modern power systems will have to face difficult challenges in the years to come: frequent blackouts in urban areas caused by high peaks of electricity demand, grid instability exacerbated by the intermittency of renewable generation, and climate change on a global scale amplified by increasing carbon emissions. While current practices are growingly inadequate, the pathway of artificial intelligence (AI)-based methods to widespread adoption is hindered by missing aspects of trustworthiness. The CityLearn Challenge is an exemplary opportunity for researchers from multi-disciplinary fields to investigate the potential of AI to tackle these pressing issues within the energy domain, collectively modeled as a reinforcement learning (RL) task. Multiple real-world challenges faced by contemporary RL techniques are embodied in the problem formulation. In this paper, we present a novel method using the solution function of optimization as policies to compute the actions for sequential decision-making, while notably adapting the parameters of the optimization model from online observations. Algorithmically, this is achieved by an evolutionary algorithm under a novel trajectory-based guidance scheme. Formally, the global convergence property is established. Our agent ranked first in the latest 2021 CityLearn Challenge, being able to achieve superior performance in almost all metrics while maintaining some key aspects of interpretability.  },
}

Modern power systems will have to face difficult challenges in the years to come: frequent blackouts in urban areas caused by high peaks of electricity demand, grid instability exacerbated by the intermittency of renewable generation, and climate change on a global scale amplified by increasing carbon emissions. While current practices are growingly inadequate, the pathway of artificial intelligence (AI)-based methods to widespread adoption is hindered by missing aspects of trustworthiness. The CityLearn Challenge is an exemplary opportunity for researchers from multi-disciplinary fields to investigate the potential of AI to tackle these pressing issues within the energy domain, collectively modeled as a reinforcement learning (RL) task. Multiple real-world challenges faced by contemporary RL techniques are embodied in the problem formulation. In this paper, we present a novel method using the solution function of optimization as policies to compute the actions for sequential decision-making, while notably adapting the parameters of the optimization model from online observations. Algorithmically, this is achieved by an evolutionary algorithm under a novel trajectory-based guidance scheme. Formally, the global convergence property is established. Our agent ranked first in the latest 2021 CityLearn Challenge, being able to achieve superior performance in almost all metrics while maintaining some key aspects of interpretability.

@inproceedings{2023_4C_AUP,
  title={Towards Robustness Certification Against Universal Perturbations},
  author={Yi Zeng* and Zhouxing Shi* and Ming Jin and Feiyang Kang and Lingjuan Lyu and Cho-Jui Hsieh and Ruoxi Jia},
  booktitle={International Conference on Learning Representations (ICLR)},
  pages={},
  year={2023},
  url_openreview={https://openreview.net/forum?id=7GEvPKxjtt},
  keywords = {Machine Learning},
  abstract={In this paper, we investigate the problem of certifying neural network robustness against universal perturbations (UPs), which have been widely used in universal adversarial attacks and backdoor attacks. Existing robustness certification methods aim to provide robustness guarantees for each sample with respect to the worst-case perturbations given a neural network. However, those sample-wise bounds will be loose when considering the UP threat model as they overlook the important constraint that the perturbation should be shared across all samples. We propose a method based on a combination of linear relaxation-based perturbation analysis and Mixed Integer Linear Programming to establish the first robust certification method for UP. In addition, we develop a theoretical framework for computing error bounds on the entire population using the certification results from a randomly sampled batch. Aside from an extensive evaluation of the proposed certification, we further show how the certification facilitates efficient comparison of robustness among different models or efficacy among different universal adversarial attack defenses and enables accurate detection of backdoor target classes.  },
}

In this paper, we investigate the problem of certifying neural network robustness against universal perturbations (UPs), which have been widely used in universal adversarial attacks and backdoor attacks. Existing robustness certification methods aim to provide robustness guarantees for each sample with respect to the worst-case perturbations given a neural network. However, those sample-wise bounds will be loose when considering the UP threat model as they overlook the important constraint that the perturbation should be shared across all samples. We propose a method based on a combination of linear relaxation-based perturbation analysis and Mixed Integer Linear Programming to establish the first robust certification method for UP. In addition, we develop a theoretical framework for computing error bounds on the entire population using the certification results from a randomly sampled batch. Aside from an extensive evaluation of the proposed certification, we further show how the certification facilitates efficient comparison of robustness among different models or efficacy among different universal adversarial attack defenses and enables accurate detection of backdoor target classes.

@inproceedings{2023_3J_MSAS,
  title={Model Residuals as Shields: A Bilevel Formulation
to Defend Smart Grids from Poisoning Attacks},
  author={Tung-Wei Lin and Padmaksha Roy and Yi Zeng and Ming Jin and Ruoxi Jia and Chen-Ching Liu and Alberto Sangiovanni-Vinecentelli},
  booktitle={Preprint},
  pages={},
  year={2023},
  url_pdf = {Power_Cybersecurity_MSAS2023.pdf},
  keywords = {Machine Learning, Power Systems, Cybersecurity}
}

@inproceedings{2023_3C_NRRL,
  title={Non-stationary Risk-sensitive Reinforcement Learning: Near-optimal Dynamic Regret, Adaptive Detection, and Separation Design},
  author={Yuhao Ding and Ming Jin and Javad Lavaei},
  booktitle={AAAI Conference on Artificial Intelligence (AAAI)},
  note = {<font style="color:#FF0000">(oral presentation)</font>},
  pages={},
  year={2023},
  url_pdf={Nonstationary_RL2022.pdf},
  url_arXiv={https://arxiv.org/pdf/2211.10815.pdf},
  keywords = {Optimization, Reinforcement Learning, Machine Learning},
  abstract={We study risk-sensitive reinforcement learning (RL) based on an entropic risk measure in episodic non-stationary Markov decision processes (MDPs). Both the reward functions and the state transition kernels are unknown and allowed to vary arbitrarily over time with a budget on their cumulative variations. When this variation budget is known a prior, we propose two restart-based algorithms, namely Restart-RSMB and Restart-RSQ, and establish their dynamic regrets. Based on these results, we further present a meta-algorithm that does not require any prior knowledge of the variation budget and can adaptively detect the non-stationarity on the exponential value functions. A dynamic regret lower bound is then established for non-stationary risk-sensitive RL to certify the near-optimality of the proposed algorithms. Our results also show that the risk control and the handling of the non-stationarity can be separately designed in the algorithm if the variation budget is known a prior, while the non-stationary detection mechanism in the adaptive algorithm depends on the risk parameter. This work offers the first non-asymptotic theoretical analyses for the non-stationary risk-sensitive RL in the literature. },
}

We study risk-sensitive reinforcement learning (RL) based on an entropic risk measure in episodic non-stationary Markov decision processes (MDPs). Both the reward functions and the state transition kernels are unknown and allowed to vary arbitrarily over time with a budget on their cumulative variations. When this variation budget is known a prior, we propose two restart-based algorithms, namely Restart-RSMB and Restart-RSQ, and establish their dynamic regrets. Based on these results, we further present a meta-algorithm that does not require any prior knowledge of the variation budget and can adaptively detect the non-stationarity on the exponential value functions. A dynamic regret lower bound is then established for non-stationary risk-sensitive RL to certify the near-optimality of the proposed algorithms. Our results also show that the risk control and the handling of the non-stationarity can be separately designed in the algorithm if the variation budget is known a prior, while the non-stationary detection mechanism in the adaptive algorithm depends on the risk parameter. This work offers the first non-asymptotic theoretical analyses for the non-stationary risk-sensitive RL in the literature.

@inproceedings{2023_3C_MetaLMRS,
  title={Learning-to-Learn to Guide Random Search: Derivative-Free Meta Blackbox Optimization on Manifold},
  author={Bilgehan Sel and Ahmad Al-Tawaha and Yuhao Ding and Ruoxi Jia and Bo Ji and Javad Lavaei and  Ming Jin},
  booktitle={Learning for Dynamics & Control Conference (L4DC)},
  note = {<font style="color:#FF0000">(oral presentation)</font>},
  pages={},
  year={2023},
  url_pdf={Meta_LMRS.pdf},
  keywords = {Optimization, Reinforcement learning, Machine Learning},
  abstract={Solving a sequence of high-dimensional, nonconvex, but potentially similar optimization problems poses a computational challenge in engineering applications. We propose the \emph{first} meta-learning framework that leverages the shared structure among sequential tasks to improve the computational efficiency and sample complexity of derivative-free optimization. Based on the observation that most practical high-dimensional functions lie on a latent low-dimensional manifold, which can be further shared among instances, our method jointly learns the meta-initialization of a search point and a meta-manifold.  Theoretically, we establish the benefit of meta-learning in this challenging setting. Empirically, we demonstrate the effectiveness of the proposed algorithm in two high-dimensional reinforcement learning tasks.  },
}

Solving a sequence of high-dimensional, nonconvex, but potentially similar optimization problems poses a computational challenge in engineering applications. We propose the \emphfirst meta-learning framework that leverages the shared structure among sequential tasks to improve the computational efficiency and sample complexity of derivative-free optimization. Based on the observation that most practical high-dimensional functions lie on a latent low-dimensional manifold, which can be further shared among instances, our method jointly learns the meta-initialization of a search point and a meta-manifold. Theoretically, we establish the benefit of meta-learning in this challenging setting. Empirically, we demonstrate the effectiveness of the proposed algorithm in two high-dimensional reinforcement learning tasks.

@inproceedings{2023_3C_MIA,
  title={Practical Membership Inference Attacks Against Large-Scale Multi-Modal Models: A Pilot Study},
  author={Myeongseob Ko and Ming Jin and Chenguang Wang and Ruoxi Jia},
  booktitle={ICCV},
  pages={},
  year={2023},
  keywords = {Machine Learning}
}

@article{2023_3C_BNODE,
  title={Certifiably Robust Neural ODE with Learning-based Barrier Function},
  author={Runing Yang and Ruoxi Jia and Xiangyu Zhang and  Ming Jin},
  journal={IEEE Control Systems Letters (Special Issue on Data-Driven Analysis and Control)},
  pages={},
  year={2023},
  url_pdf={B-NODE22.pdf},
  keywords = {Optimization, Control theory, Machine Learning},
  abstract={Neural Ordinary Differential Equations (ODEs) have gained traction in many applications. While recent studies have focused on empirically increasing the robustness of neural ODEs against natural or adversarial attacks, certified robustness is still lacking. In this work, we propose a framework for training a neural ODE using barrier functions and demonstrate improved robustness for classification problems. We further provide the first generalization guarantee of robustness against adversarial attacks using a wait-and-judge scenario approach.  },
}

Neural Ordinary Differential Equations (ODEs) have gained traction in many applications. While recent studies have focused on empirically increasing the robustness of neural ODEs against natural or adversarial attacks, certified robustness is still lacking. In this work, we propose a framework for training a neural ODE using barrier functions and demonstrate improved robustness for classification problems. We further provide the first generalization guarantee of robustness against adversarial attacks using a wait-and-judge scenario approach.

@inproceedings{2023_2C_TUNEOPT,
  title={TUNEOPT: An Evolutionary Reinforcement Learning HVAC Controller For Energy-Comfort Optimization Tuning},
  author={Mostafa Meimand and Vanshaj Khattar and Zahra Yazdani and Farrokh Jazizadeh and Ming Jin},
  booktitle={ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation (BuildSys)},
  pages={},
  year={2023},
  url_pdf = {TUNEOPT23.pdf},
  keywords = {Machine Learning, Control, Energy}
}

@article{2023_2C_SobolevTrain,
  title={A Theoretical Analysis of Using Gradient Data for Sobolev Training in RKHS},
  author={Zain ul Abdeen and Ruoxi Jia and Vassilis Kekatos and Ming Jin},
  year={2023},
   journal = {IFAC World Congress}, 
  url_pdf={Sobolev_training2022.pdf},
  keywords = {Machine Learning, Optimization},
  abstract={Recent works empirically demonstrated that incorporating target derivatives, in addition to the conventional usage of target values, during the training process improves the accuracy of the predictor and data efficiency. Despite the successful application of gradient data in the learning process, very little is understood theoretically about their performance guarantee. In this paper, our goal is to highlight (i) the limitations of gradient data on their performance guarantees, especially in low-data regimes, and (ii) the extent to which the gradients affect the learning rate. Our result implies that in a low-data regime, if the Lipschitz of the target function is below a threshold, gradient data for Sobolev training outperforms the classical training in terms of sample efficiency. For a target function with a large Lipschitz constant, there is a threshold for training data size beyond which the gradient data perform better than conventional training. The convergence behavior of gradient data for Sobolev training is studied, and the learning rate of order $\mathcal{O}(n^{-\frac{1}{2}+\epsilon})$ is derived. Experiments are conducted to determine the effect of gradient data in the learning process.   }
}

Recent works empirically demonstrated that incorporating target derivatives, in addition to the conventional usage of target values, during the training process improves the accuracy of the predictor and data efficiency. Despite the successful application of gradient data in the learning process, very little is understood theoretically about their performance guarantee. In this paper, our goal is to highlight (i) the limitations of gradient data on their performance guarantees, especially in low-data regimes, and (ii) the extent to which the gradients affect the learning rate. Our result implies that in a low-data regime, if the Lipschitz of the target function is below a threshold, gradient data for Sobolev training outperforms the classical training in terms of sample efficiency. For a target function with a large Lipschitz constant, there is a threshold for training data size beyond which the gradient data perform better than conventional training. The convergence behavior of gradient data for Sobolev training is studied, and the learning rate of order $\mathcal{O}(n^{-\frac{1}{2}+ε})$ is derived. Experiments are conducted to determine the effect of gradient data in the learning process.

@article{2023_2C_CyberPower,
  title={Decision-Focused Learning for Inverse Noncooperative Games: Generalization Bounds and Convergence Analysis},
  author={Ahmad Al-Tawaha and Harshal Kaushik and Bilgehan Sel and Ruoxi Jia and Ming Jin},
  year={2023},
   journal = {IFAC World Congress}, 
  url_pdf={DFL-conf2022.pdf},
  keywords = {Machine Learning, Optimization},
  abstract={Finding the equilibrium strategy of agents is one of the central problems in game theory. Perhaps equally intriguing is the inverse of the above problem: from the available finite set of actions at equilibrium, how can we learn the utilities of players competing against each other and eventually use the learned models to predict their future actions?  Instead of following an estimate-then-predict approach, this work proposes a decision-focused learning (DFL) method that directly learns the utility function to improve prediction accuracy. The game's equilibrium is represented as a layer and integrated into an end-to-end optimization framework. We discuss the statistical bounds of covering numbers for the set of solution functions corresponding to the solution of a generic parametric variational inequality. Also, we establish the generalization bound for the set of solution functions with respect to smooth loss function with an improved rate. Moreover, we proposed an algorithm based on the iterative differentiation strategy to forward and back propagate through the equilibrium layer. The convergence analysis of the proposed algorithm is established. Finally, We numerically validate the proposed framework in the utility learning problem among the agents whose utility functions are approximated by partially input convex neural networks (PICNN).  }
}

Finding the equilibrium strategy of agents is one of the central problems in game theory. Perhaps equally intriguing is the inverse of the above problem: from the available finite set of actions at equilibrium, how can we learn the utilities of players competing against each other and eventually use the learned models to predict their future actions? Instead of following an estimate-then-predict approach, this work proposes a decision-focused learning (DFL) method that directly learns the utility function to improve prediction accuracy. The game's equilibrium is represented as a layer and integrated into an end-to-end optimization framework. We discuss the statistical bounds of covering numbers for the set of solution functions corresponding to the solution of a generic parametric variational inequality. Also, we establish the generalization bound for the set of solution functions with respect to smooth loss function with an improved rate. Moreover, we proposed an algorithm based on the iterative differentiation strategy to forward and back propagate through the equilibrium layer. The convergence analysis of the proposed algorithm is established. Finally, We numerically validate the proposed framework in the utility learning problem among the agents whose utility functions are approximated by partially input convex neural networks (PICNN).

@article{2022_3C_ppg,
  title={Recurrent Neural Network Controllers Synthesis with Stability Guarantees for Partially Observed Systems},
  author={Gu, Fangda and Yin, He and El Ghaoui, Laurent and Arcak, Murat and Seiler, Peter and Jin, Ming},
  year={2022},
   journal = "AAAI Conference on Artificial Intelligence (AAAI)",
  url_pdf={PPG_2021.pdf},
  url_arXiv={https://arxiv.org/abs/2109.03861},
  keywords = {Machine learning, Reinforcement learning},
  abstract={Neural network controllers have become popular in control tasks thanks to their flexibility and expressivity. Stability is a crucial property for safety-critical dynamical systems, while stabilization of partially observed systems, in many cases, requires controllers to retain and process long-term memories of the past. We consider the important class of recurrent neural networks (RNN) as dynamic controllers for nonlinear uncertain partially-observed systems, and derive convex stability conditions based on integral quadratic constraints, S-lemma and sequential convexification. To ensure stability during the learning and control process, we propose a projected policy gradient method that iteratively enforces the stability conditions in the reparametrized space taking advantage of mild additional information on system dynamics. Numerical experiments show that our method learns stabilizing controllers while using fewer samples and achieving higher final performance compared with policy gradient.}
}

Neural network controllers have become popular in control tasks thanks to their flexibility and expressivity. Stability is a crucial property for safety-critical dynamical systems, while stabilization of partially observed systems, in many cases, requires controllers to retain and process long-term memories of the past. We consider the important class of recurrent neural networks (RNN) as dynamic controllers for nonlinear uncertain partially-observed systems, and derive convex stability conditions based on integral quadratic constraints, S-lemma and sequential convexification. To ensure stability during the learning and control process, we propose a projected policy gradient method that iteratively enforces the stability conditions in the reparametrized space taking advantage of mild additional information on system dynamics. Numerical experiments show that our method learns stabilizing controllers while using fewer samples and achieving higher final performance compared with policy gradient.

@article{2022_3C_AdUn,
  title={Adversarial Unlearning of Backdoors via Implicit Hypergradient},
  author={Yi Zeng and Si Chen and Won Park and Z. Morley Mao and Ming Jin and Ruoxi Jia},
  year={2022},
   journal = {International Conference on Learning Representations (ICLR)}, 
  url_pdf={AdversarialUnlearning.pdf},
  url_arXiv={https://arxiv.org/abs/2110.03735},
  keywords = {Machine Learning},
  abstract={We propose a minimax formulation for removing backdoors from a given poisoned model based on a small set of clean data. This formulation encompasses much of prior work on backdoor removal. We propose the Implicit Bacdoor Adversarial Unlearning (I-BAU) algorithm to solve the minimax. Unlike previous work, which breaks down the minimax into separate inner and outer problems, our algorithm utilizes the implicit hypergradient to account for the interdependence between inner and outer optimization. We theoretically analyze its convergence and the generalizability of the robustness gained by solving minimax on clean data to unseen test data. In our evaluation, we compare I-BAU with six state-of-art backdoor defenses on seven backdoor attacks over two datasets and various attack settings, including the common setting where the attacker targets one class as well as important but underexplored settings where multiple classes are targeted. I-BAU's performance is comparable to and most often significantly better than the best baseline. Particularly, its performance is more robust to the variation on triggers, attack settings, poison ratio, and clean data size. Moreover, I-BAU requires less computation to take effect; particularly, it is more than 13× faster than the most efficient baseline in the single-target attack setting. Furthermore, it can remain effective in the extreme case where the defender can only access 100 clean samples -- a setting where all the baselines fail to produce acceptable results.  }
}

We propose a minimax formulation for removing backdoors from a given poisoned model based on a small set of clean data. This formulation encompasses much of prior work on backdoor removal. We propose the Implicit Bacdoor Adversarial Unlearning (I-BAU) algorithm to solve the minimax. Unlike previous work, which breaks down the minimax into separate inner and outer problems, our algorithm utilizes the implicit hypergradient to account for the interdependence between inner and outer optimization. We theoretically analyze its convergence and the generalizability of the robustness gained by solving minimax on clean data to unseen test data. In our evaluation, we compare I-BAU with six state-of-art backdoor defenses on seven backdoor attacks over two datasets and various attack settings, including the common setting where the attacker targets one class as well as important but underexplored settings where multiple classes are targeted. I-BAU's performance is comparable to and most often significantly better than the best baseline. Particularly, its performance is more robust to the variation on triggers, attack settings, poison ratio, and clean data size. Moreover, I-BAU requires less computation to take effect; particularly, it is more than 13× faster than the most efficient baseline in the single-target attack setting. Furthermore, it can remain effective in the extreme case where the defender can only access 100 clean samples – a setting where all the baselines fail to produce acceptable results.

@article{2022_2J_proxy,
  title={Controlling Smart Inverters using Proxies: A Chance-Constrained DNN-based Approach},
  author={Gupta, Sarthak and Kekatos, Vassilis and Jin, Ming},
  year={2022},
  journal = "IEEE Transactions on Smart Grid",
  url_pdf={Control_Proxy_2021.pdf},
  url_arXiv={https://arxiv.org/abs/2105.00429},
  keywords = {Optimization, Machine learning, Power system},
  abstract={Coordinating inverters at scale under uncertainty is the desideratum for integrating renewables in distribution grids. Unless load demands and solar generation are telemetered frequently, controlling inverters given approximate grid conditions or proxies thereof becomes a key specification. Although deep neural networks (DNNs) can learn optimal inverter schedules, guaranteeing feasibility is largely elusive. Rather than training DNNs to imitate already computed optimal power flow (OPF) solutions, this work integrates DNN-based inverter policies into the OPF. The proposed DNNs are trained through two OPF alternatives that confine voltage deviations on the average and as a convex restriction of chance constraints. The trained DNNs can be driven by partial, noisy, or proxy descriptors of the current grid conditions. This is important when OPF has to be solved for an unobservable feeder. DNN weights are trained via back-propagation and upon differentiating the AC power flow equations assuming the network model is known. Otherwise, a gradient-free variant is put forth. The latter is relevant when inverters are controlled by an aggregator having access only to a power flow solver or a digital twin of the feeder. Numerical tests compare the DNN-based inverter control schemes with the optimal inverter setpoints in terms of optimality and feasibility.}}
  

Coordinating inverters at scale under uncertainty is the desideratum for integrating renewables in distribution grids. Unless load demands and solar generation are telemetered frequently, controlling inverters given approximate grid conditions or proxies thereof becomes a key specification. Although deep neural networks (DNNs) can learn optimal inverter schedules, guaranteeing feasibility is largely elusive. Rather than training DNNs to imitate already computed optimal power flow (OPF) solutions, this work integrates DNN-based inverter policies into the OPF. The proposed DNNs are trained through two OPF alternatives that confine voltage deviations on the average and as a convex restriction of chance constraints. The trained DNNs can be driven by partial, noisy, or proxy descriptors of the current grid conditions. This is important when OPF has to be solved for an unobservable feeder. DNN weights are trained via back-propagation and upon differentiating the AC power flow equations assuming the network model is known. Otherwise, a gradient-free variant is put forth. The latter is relevant when inverters are controlled by an aggregator having access only to a power flow solver or a digital twin of the feeder. Numerical tests compare the DNN-based inverter control schemes with the optimal inverter setpoints in terms of optimality and feasibility.

@inproceedings{2022_2C_learning,
  title={Learning Neural Networks under Input-Output Specifications},
  author={Zain ul Abdeen and He Yin and Vassilis Kekatos and Ming Jin},
  booktitle={American Control Conference},
  pages={},
  year={2022},
  url_pdf={LearnNNSpecs.pdf},
  keywords = {Optimization, Machine Learning},
  abstract={In this paper, we examine an important problem of learning neural networks that certifiably meet certain specifications on input-output behaviors.  Our strategy is to find an inner approximation of the set of admissible policy parameters, which is convex in a transformed space. To this end, we address the key technical challenge of convexifying the verification condition for neural networks, which is derived by abstracting the nonlinear specifications and activation functions with quadratic constraints. In particular, we propose a reparametrization scheme of the original neural network based on loop transformation, which leads to a convex condition that can be computationally enforced during learning. The theoretical construction is validated in an experiment that specifies reachable sets for different regions of inputs.  },
  keywords={Machine learning, Optimization}
}

In this paper, we examine an important problem of learning neural networks that certifiably meet certain specifications on input-output behaviors. Our strategy is to find an inner approximation of the set of admissible policy parameters, which is convex in a transformed space. To this end, we address the key technical challenge of convexifying the verification condition for neural networks, which is derived by abstracting the nonlinear specifications and activation functions with quadratic constraints. In particular, we propose a reparametrization scheme of the original neural network based on loop transformation, which leads to a convex condition that can be computationally enforced during learning. The theoretical construction is validated in an experiment that specifies reachable sets for different regions of inputs.

@article{2022_2C_CyberPower,
  title={Defense against Joint Poisoning and Evasion Attacks: A Case Study of DERMS},
  author={Zain ul Abdeen and Padmaksha Roy and Ahmad Al-Tawaha and Rouxi Jia and Laura Freeman and Peter Beling and Chen-Ching Liu and Alberto Sangiovanni-Vincentelli and Ming Jin},
  year={2022},
   journal = {Preprint}, 
  url_pdf={Cybersecurity_PS2022.pdf},
  keywords = {Machine Learning, Cybersecurity, Power system},
  abstract={There is an upward trend of deploying distributed energy resource management systems (DERMS) to control modern power grids. However, DERMS controller communication lines are vulnerable to cyberattacks that could potentially impact operational reliability. While a data-driven intrusion detection system (IDS) can potentially thwart attacks during deployment, also known as  the evasion attack, the training of the detection algorithm may be corrupted by adversarial data injected into the database, also known as the poisoning attack.  In this paper, we propose the \emph{first} framework of IDS that is robust against joint poisoning and evasion attacks. We formulate the defense mechanism as a bilevel optimization, where the inner and outer levels deal with attacks that occur during training time and testing time, respectively. We verify the robustness of our method on the IEEE-13 bus feeder model against a diverse set of poisoning and evasion attack scenarios. The results indicate that our proposed method outperforms the baseline technique in terms of accuracy, precision, and recall for intrusion detection.  }
}

There is an upward trend of deploying distributed energy resource management systems (DERMS) to control modern power grids. However, DERMS controller communication lines are vulnerable to cyberattacks that could potentially impact operational reliability. While a data-driven intrusion detection system (IDS) can potentially thwart attacks during deployment, also known as the evasion attack, the training of the detection algorithm may be corrupted by adversarial data injected into the database, also known as the poisoning attack. In this paper, we propose the \emphfirst framework of IDS that is robust against joint poisoning and evasion attacks. We formulate the defense mechanism as a bilevel optimization, where the inner and outer levels deal with attacks that occur during training time and testing time, respectively. We verify the robustness of our method on the IEEE-13 bus feeder model against a diverse set of poisoning and evasion attack scenarios. The results indicate that our proposed method outperforms the baseline technique in terms of accuracy, precision, and recall for intrusion detection.

@article{2021_2J_ONO,
  title={Dynamic Regret Bounds for Constrained Online Nonconvex Optimization
Based on Polyak-Lojasiewicz Regions},
  author={Julie Mulvaney-Kemp and SangWoo Park and Ming Jin and Javad Lavaei},
  year={2022},
   note={IEEE Transactions on Control of Network Systems}, 
  url_pdf={ONO_dynamic_2022.pdf},
  keywords = {Optimization},
  abstract={Online optimization problems are well-understood in the convex case, where algorithmic performance is typically measured relative to the best fixed decision. In this paper, we shed light on online nonconvex optimization problems in which algorithms are evaluated against the optimal decision at each time using the more useful notion of dynamic regret. The focus is on loss functions which are arbitrarily nonconvex, but have global solutions that are slowly time-varying. We address this problem by first analyzing the region around the global solution at each time to define time-varying target sets, which contain the global solution and exhibit desirable properties under the projected gradient descent algorithm. All points in a target set satisfy the proximal Polyak-Łojasiewicz inequality, among other conditions. Then, we introduce two algorithms and prove that the dynamic regret for each algorithm is bounded by a function of the temporal variation in the optimal decision. The first algorithm assumes that the decision maker has some prior knowledge about the initial objective function and may query the gradient repeatedly at each time. This algorithm ensures that decisions are within the target set at every time. The second algorithm makes no assumption about prior knowledge. It instead relies on random sampling and memory to find and then track the target sets over time. In this case, the landscape of the loss functions determines the likelihood that the dynamic regret will be small. Numerical experiments validate these theoretical results and highlight the impact of a single low-complexity problem early in the sequence.}
}

Online optimization problems are well-understood in the convex case, where algorithmic performance is typically measured relative to the best fixed decision. In this paper, we shed light on online nonconvex optimization problems in which algorithms are evaluated against the optimal decision at each time using the more useful notion of dynamic regret. The focus is on loss functions which are arbitrarily nonconvex, but have global solutions that are slowly time-varying. We address this problem by first analyzing the region around the global solution at each time to define time-varying target sets, which contain the global solution and exhibit desirable properties under the projected gradient descent algorithm. All points in a target set satisfy the proximal Polyak-Łojasiewicz inequality, among other conditions. Then, we introduce two algorithms and prove that the dynamic regret for each algorithm is bounded by a function of the temporal variation in the optimal decision. The first algorithm assumes that the decision maker has some prior knowledge about the initial objective function and may query the gradient repeatedly at each time. This algorithm ensures that decisions are within the target set at every time. The second algorithm makes no assumption about prior knowledge. It instead relies on random sampling and memory to find and then track the target sets over time. In this case, the landscape of the loss functions determines the likelihood that the dynamic regret will be small. Numerical experiments validate these theoretical results and highlight the impact of a single low-complexity problem early in the sequence.

@inproceedings{2021_3C_power,
  title={Power up! Robust graph convolutional network via graph powering},
  author={Jin, Ming and Chang, Heng and Zhu, Wenwu and Sojoudi, Somayeh},
  booktitle={AAAI Conference on Artificial Intelligence (AAAI)},
  url_pdf={Robust_GCN.pdf},
  url_arXiv={https://arxiv.org/abs/1905.10029},
  keywords={Graph theory, Machine learning},
  abstract={Graph convolutional networks (GCNs) are powerful tools for graph-structured data. However, they have been recently shown to be vulnerable to topological attacks. To enhance adversarial robustness, we go beyond spectral graph theory to robust graph theory. By challenging the classical graph Laplacian, we propose a new convolution operator that is provably robust in the spectral domain and is incorporated in the GCN architecture to improve expressivity and interpretability. By extending the original graph to a sequence of graphs, we also propose a robust training paradigm that encourages transferability across graphs that span a range of spatial and spectral characteristics. The proposed approaches are demonstrated in extensive experiments to simultaneously improve performance in both benign and adversarial situations.},
  year={2021}
}

Graph convolutional networks (GCNs) are powerful tools for graph-structured data. However, they have been recently shown to be vulnerable to topological attacks. To enhance adversarial robustness, we go beyond spectral graph theory to robust graph theory. By challenging the classical graph Laplacian, we propose a new convolution operator that is provably robust in the spectral domain and is incorporated in the GCN architecture to improve expressivity and interpretability. By extending the original graph to a sequence of graphs, we also propose a robust training paradigm that encourages transferability across graphs that span a range of spatial and spectral characteristics. The proposed approaches are demonstrated in extensive experiments to simultaneously improve performance in both benign and adversarial situations.

@article{2021_3C_imitation,
  title={Imitation Learning with Stability and Safety Guarantees},
  author={Yin, He and Seiler, Peter and Jin, Ming and Arcak, Murat},
  year={2021},
  note={IEEE Control Systems Letters}, 
  url_pdf={SafeIL2020.pdf},
  url_arXiv={https://arxiv.org/abs/2012.09293},
  keywords = {Control theory, Machine learning},
  abstract={A method is presented to synthesize neural network (NN) controllers with stability and safety guarantees through imitation learning. Convex stability and safety conditions are derived for linear time-invariant plant dynamics with NN controllers. The proposed approach merges Lyapunov theory with local quadratic constraints to bound the nonlinear activation functions in the NN. The safe imitation learning problem is formulated as an optimization problem with the goal of minimizing the imitation learning loss, and maximizing volume of the region of attraction associated with the NN controller, while enforcing the stability and safety conditions. An alternating direction method of multipliers based algorithm is proposed to solve the optimization. The method is illustrated on an inverted pendulum system and aircraft longitudinal dynamics.}}

A method is presented to synthesize neural network (NN) controllers with stability and safety guarantees through imitation learning. Convex stability and safety conditions are derived for linear time-invariant plant dynamics with NN controllers. The proposed approach merges Lyapunov theory with local quadratic constraints to bound the nonlinear activation functions in the NN. The safe imitation learning problem is formulated as an optimization problem with the goal of minimizing the imitation learning loss, and maximizing volume of the region of attraction associated with the NN controller, while enforcing the stability and safety conditions. An alternating direction method of multipliers based algorithm is proposed to solve the optimization. The method is illustrated on an inverted pendulum system and aircraft longitudinal dynamics.

@article{2021_3C_dqm,
  title={A Unified Framework for Task-Driven Data Quality Management},
  author={Wang, Tianhao and Zeng, Yi and Jin, Ming and Jia, Ruoxi},
  year={2021},
   note={Preprint.}, 
  url_pdf={DQM_2021.pdf},
  url_arXiv={https://arxiv.org/abs/2106.05484?context=cs},
  keywords = {Machine learning},
  abstract={High-quality data is critical to train performant Machine Learning (ML) models, highlighting the importance of Data Quality Management (DQM). Existing DQM schemes often cannot satisfactorily improve ML performance because, by design, they are oblivious to downstream ML tasks. Besides, they cannot handle various data quality issues (especially those caused by adversarial attacks) and have limited applications to only certain types of ML models. Recently, data valuation approaches (e.g., based on the Shapley value) have been leveraged to perform DQM; yet, empirical studies have observed that their performance varies considerably based on the underlying data and training process. In this paper, we propose a task-driven, multi-purpose, model-agnostic DQM framework, DataSifter, which is optimized towards a given downstream ML task, capable of effectively removing data points with various defects, and applicable to diverse models. Specifically, we formulate DQM as an optimization problem and devise a scalable algorithm to solve it. Furthermore, we propose a theoretical framework for comparing the worst-case performance of different DQM strategies. Remarkably, our results show that the popular strategy based on the Shapley value may end up choosing the worst data subset in certain practical scenarios. Our evaluation shows that DataSifter achieves and most often significantly improves the state-of-the-art performance over a wide range of DQM tasks, including backdoor, poison, noisy/mislabel data detection, data summarization, and data debiasing.}
}

High-quality data is critical to train performant Machine Learning (ML) models, highlighting the importance of Data Quality Management (DQM). Existing DQM schemes often cannot satisfactorily improve ML performance because, by design, they are oblivious to downstream ML tasks. Besides, they cannot handle various data quality issues (especially those caused by adversarial attacks) and have limited applications to only certain types of ML models. Recently, data valuation approaches (e.g., based on the Shapley value) have been leveraged to perform DQM; yet, empirical studies have observed that their performance varies considerably based on the underlying data and training process. In this paper, we propose a task-driven, multi-purpose, model-agnostic DQM framework, DataSifter, which is optimized towards a given downstream ML task, capable of effectively removing data points with various defects, and applicable to diverse models. Specifically, we formulate DQM as an optimization problem and devise a scalable algorithm to solve it. Furthermore, we propose a theoretical framework for comparing the worst-case performance of different DQM strategies. Remarkably, our results show that the popular strategy based on the Shapley value may end up choosing the worst data subset in certain practical scenarios. Our evaluation shows that DataSifter achieves and most often significantly improves the state-of-the-art performance over a wide range of DQM tasks, including backdoor, poison, noisy/mislabel data detection, data summarization, and data debiasing.

@inproceedings{2021_1C_dim,
  title={Diminishing Regret for Online Nonconvex Optimization},
  author={SangWoo Park and Julie Mulvaney-Kemp and Ming Jin and Javad Lavaei},
  booktitle={American Control Conference},
  pages={},
  year={2021},
  url_pdf={regret_ONO_2020_1.pdf},
  abstract={A single nonconvex optimization is NP-hard in the worst case, and so is a sequence of nonconvex problems viewed separately. For online nonconvex optimization (ONO) problems, widely used local search algorithms are guaranteed to track a sequence of local optima, but offer no promises about global optimality. In this paper, we introduce the concept of nonconvexity regret to measure the performance of a local search method against a global optimization solver for ONO. We define the notion of depth of a global minimum, and show that memory and random explorations drive the nonconvexity regret to zero if the variability of the objective function is low compared to the depth of the global minima. We prove probabilistic guarantees on the regret bound that depend on the evolution of the landscapes of the time-varying objective functions. Then, based on the notions of missing mass and 1-occupancy set, we develop a practical algorithm that works even when there is no such information on the landscapes. The theoretical results imply that the existence of a low-complexity optimization at any arbitrary time instance of ONO can nullify the NP-hardness of the entire ONO problem. The results are verified through numerical simulations.},
  keywords={Optimization}
}

A single nonconvex optimization is NP-hard in the worst case, and so is a sequence of nonconvex problems viewed separately. For online nonconvex optimization (ONO) problems, widely used local search algorithms are guaranteed to track a sequence of local optima, but offer no promises about global optimality. In this paper, we introduce the concept of nonconvexity regret to measure the performance of a local search method against a global optimization solver for ONO. We define the notion of depth of a global minimum, and show that memory and random explorations drive the nonconvexity regret to zero if the variability of the objective function is low compared to the depth of the global minima. We prove probabilistic guarantees on the regret bound that depend on the evolution of the landscapes of the time-varying objective functions. Then, based on the notions of missing mass and 1-occupancy set, we develop a practical algorithm that works even when there is no such information on the landscapes. The theoretical results imply that the existence of a low-complexity optimization at any arbitrary time instance of ONO can nullify the NP-hardness of the entire ONO problem. The results are verified through numerical simulations.

@ARTICLE{2020_3J_boundary,
  author={M. {Jin} and J. {Lavaei} and S. {Sojoudi} and R. {Baldick}},
  journal={IEEE Transactions on Information Forensics and Security}, 
  title={Boundary Defense Against Cyber Threat for Power System State Estimation}, 
  year={2020},
  volume={},
  number={},
  pages={1-16},
  doi={10.1109/TIFS.2020.3043065},
  abstract={The operation of power grids is becoming increasingly data-centric. While the abundance of data could improve system efficiency, it poses major reliability challenges. In particular, state estimation aims to find the operating state of a network from the telemetered data, but an undetected attack on the data could lead to making wrong operational decisions for the system and trigger a large-scale blackout. Nevertheless, understanding the vulnerability of state estimation with regards to cyberattacks, which is a special instance of graph-structured quadratic sensing problem, has been hindered by the lack of tools for studying the topological and data-analytic aspects of networks. Algorithmic robustness is critical in extracting reliable information from abundant but untrusted grid data. For a large-scale power grid, we quantify, analyze, and visualize the regions of the network that are not robust to cyberattacks in the sense that there exists a data manipulation strategy for each of those local regions that misleads the operator at the global scale and yields a wrong estimation of the state of the network at almost all buses. We also propose an optimization-based graphical boundary defense mechanism to identify the border of the geographical area in which data have been manipulated. The proposed method does not allow a local attack to have a global effect on the data analysis of the entire network, which enhances the situational awareness of the grid, especially in the face of adversity. The developed mathematical framework reveals key geometric and algebraic factors that can affect algorithmic robustness and is used to study the vulnerability of the U.S. power grid in this paper.},
  url_pdf =    {TIFS_Boundary_Defense.pdf},
  url_link =     {https://ieeexplore.ieee.org/document/9290080},
  keywords = {Cybersecurity, Power system, Optimization, Graph theory}}
  

The operation of power grids is becoming increasingly data-centric. While the abundance of data could improve system efficiency, it poses major reliability challenges. In particular, state estimation aims to find the operating state of a network from the telemetered data, but an undetected attack on the data could lead to making wrong operational decisions for the system and trigger a large-scale blackout. Nevertheless, understanding the vulnerability of state estimation with regards to cyberattacks, which is a special instance of graph-structured quadratic sensing problem, has been hindered by the lack of tools for studying the topological and data-analytic aspects of networks. Algorithmic robustness is critical in extracting reliable information from abundant but untrusted grid data. For a large-scale power grid, we quantify, analyze, and visualize the regions of the network that are not robust to cyberattacks in the sense that there exists a data manipulation strategy for each of those local regions that misleads the operator at the global scale and yields a wrong estimation of the state of the network at almost all buses. We also propose an optimization-based graphical boundary defense mechanism to identify the border of the geographical area in which data have been manipulated. The proposed method does not allow a local attack to have a global effect on the data analysis of the entire network, which enhances the situational awareness of the grid, especially in the face of adversity. The developed mathematical framework reveals key geometric and algebraic factors that can affect algorithmic robustness and is used to study the vulnerability of the U.S. power grid in this paper.

@inproceedings{2020_3C_dl,
  title={Deep Learning for Reactive Power Control of Smart Inverters under Communication Constraints},
  author={Gupta, Sarthak and Kekatos, Vassilis and Jin, Ming},
  booktitle={IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm)},
  year={2020},
  url_link={https://smartgridcomm.info/day/1},
  url_pdf={DL_SmartInverter_2020.pdf},
  url_arXiv={https://arxiv.org/abs/2007.05868},
  abstract={Aiming for the median solution between cyber-intensive optimal power flow (OPF) solutions and subpar local control, this work advocates deciding inverter injection setpoints using deep neural networks (DNNs). Instead of fitting OPF solutions in a black-box manner, inverter DNNs are naturally integrated with the feeder model and trained to minimize a grid-wide objective subject to inverter and network constraints enforced on the average over uncertain grid conditions. Learning occurs in a quasi-stationary fashion and is posed as a stochastic OPF, handled via stochastic primal-dual updates acting on grid data scenarios. Although trained as a whole, the proposed DNN is operated in a master-slave architecture. Its master part is run at the utility to output a condensed control signal broadcast to all inverters. Its slave parts are implemented by inverters and are driven by the utility signal along with local inverter readings. This novel DNN structure uniquely addresses the small-big data conundrum where utilities collect detailed smart meter readings yet on an hourly basis, while in real time inverters should be driven by local inputs and minimal utility coordination to save on communication. Numerical tests corroborate the efficacy of this physics-aware DNN-based inverter solution over an optimal control policy.},
  keywords={Machine learning, Power system}
}

Aiming for the median solution between cyber-intensive optimal power flow (OPF) solutions and subpar local control, this work advocates deciding inverter injection setpoints using deep neural networks (DNNs). Instead of fitting OPF solutions in a black-box manner, inverter DNNs are naturally integrated with the feeder model and trained to minimize a grid-wide objective subject to inverter and network constraints enforced on the average over uncertain grid conditions. Learning occurs in a quasi-stationary fashion and is posed as a stochastic OPF, handled via stochastic primal-dual updates acting on grid data scenarios. Although trained as a whole, the proposed DNN is operated in a master-slave architecture. Its master part is run at the utility to output a condensed control signal broadcast to all inverters. Its slave parts are implemented by inverters and are driven by the utility signal along with local inverter readings. This novel DNN structure uniquely addresses the small-big data conundrum where utilities collect detailed smart meter readings yet on an hourly basis, while in real time inverters should be driven by local inputs and minimal utility coordination to save on communication. Numerical tests corroborate the efficacy of this physics-aware DNN-based inverter solution over an optimal control policy.

@ARTICLE{2020_2J_stabilityrl,
  author={M. {Jin} and J. {Lavaei}},
  journal={IEEE Access}, 
  title={Stability-Certified Reinforcement Learning: A Control-Theoretic Perspective}, 
  year={2020},
  volume={},
  number={},
  pages={1-1},
  doi={10.1109/ACCESS.2020.3045114},
  abstract={We investigate the important problem of certifying stability of reinforcement learning policies when interconnected with nonlinear dynamical systems. We show that by regulating the partial gradients of policies, strong guarantees of robust stability can be obtained based on a proposed semidefinite programming feasibility problem. The method is able to certify a large set of stabilizing controllers by exploiting problem-specific structures; furthermore, we analyze and establish its (non)conservatism. Empirical evaluations on two decentralized control tasks, namely multi-flight formation and power system frequency regulation, demonstrate that the reinforcement learning agents can have high performance within the stability-certified parameter space and also exhibit stable learning behaviors in the long run.},
  url_pdf =    {SafeRL_2020.pdf},
  url_link =     {https://ieeexplore.ieee.org/document/9296215},
  keywords = {Control theory, Machine learning, Power system}}

We investigate the important problem of certifying stability of reinforcement learning policies when interconnected with nonlinear dynamical systems. We show that by regulating the partial gradients of policies, strong guarantees of robust stability can be obtained based on a proposed semidefinite programming feasibility problem. The method is able to certify a large set of stabilizing controllers by exploiting problem-specific structures; furthermore, we analyze and establish its (non)conservatism. Empirical evaluations on two decentralized control tasks, namely multi-flight formation and power system frequency regulation, demonstrate that the reinforcement learning agents can have high performance within the stability-certified parameter space and also exhibit stable learning behaviors in the long run.

@ARTICLE{2020_2J_scalable,
  author={M. {Jin} and I. {Molybog} and R. {Mohammadi-Ghazi} and J. {Lavaei}},
  journal={IEEE Transactions on Smart Grid}, 
  title={Scalable and Robust State Estimation From Abundant but Untrusted Data}, 
  year={2020},
  volume={11},
  number={3},
  pages={1880-1894},
  doi={10.1109/TSG.2019.2944986},
  abstract={Power system state estimation is an important problem in grid operation that has a long tradition of research since 1960s. Due to the nonconvexity of the problem, existing approaches based on local search methods are susceptible to spurious local minima, which could endanger the reliability of the system. In general, even in the absence of noise, it is challenging to provide a practical condition under which one can uniquely identify the global solution due to its NP-hardness. In this study, we propose a linear basis of representation that succinctly captures the topology of the network and enables an efficient two-stage estimation method in case the amount of measured data is not too low. Based on this framework, we propose an identifiability condition that numerically depicts the boundary where one can warrant efficient recovery of the unique global minimum. Furthermore, we develop a robustness metric called “mutual incoherence,” which underpins theoretical analysis of global recovery condition and statistical error bounds in the presence of both dense noise and bad data. The method demonstrates superior performance over existing methods in terms of both estimation accuracy and bad data robustness in an array of benchmark systems. Above all, it is scalable to large systems with more than 13,000 buses and can achieve accurate estimation within a minute.},
  url_pdf =    {PSSE_Linear_Estimator.pdf},
  url_link =     {https://ieeexplore.ieee.org/document/8855013},
  keywords = {Power system, Optimization, Cybersecurity}}
  

Power system state estimation is an important problem in grid operation that has a long tradition of research since 1960s. Due to the nonconvexity of the problem, existing approaches based on local search methods are susceptible to spurious local minima, which could endanger the reliability of the system. In general, even in the absence of noise, it is challenging to provide a practical condition under which one can uniquely identify the global solution due to its NP-hardness. In this study, we propose a linear basis of representation that succinctly captures the topology of the network and enables an efficient two-stage estimation method in case the amount of measured data is not too low. Based on this framework, we propose an identifiability condition that numerically depicts the boundary where one can warrant efficient recovery of the unique global minimum. Furthermore, we develop a robustness metric called “mutual incoherence,” which underpins theoretical analysis of global recovery condition and statistical error bounds in the presence of both dense noise and bad data. The method demonstrates superior performance over existing methods in terms of both estimation accuracy and bad data robustness in an array of benchmark systems. Above all, it is scalable to large systems with more than 13,000 buses and can achieve accurate estimation within a minute.

@inproceedings{2020_2C_ope,
  title={Towards Off-policy Evaluation as a Prerequisite for Real-world Reinforcement Learning in Building Control},
  author={Chen, Bingqing and Jin, Ming and Wang, Zhe and Hong, Tianzhen and Berg{\'e}s, Mario},
  booktitle={International Workshop on Reinforcement Learning for Energy Management in Buildings \& Cities (RLEM)},
  pages={52--56},
  year={2020},
  url_link={https://dl.acm.org/doi/abs/10.1145/3427773.3427871},
  url_pdf={OPE_RLEM20.pdf},
  abstract={We present an initial study of off-policy evaluation (OPE), a problem prerequisite to real-world reinforcement learning (RL), in the context of building control. OPE is the problem of estimating a policy's performance without running it on the actual system, using historical data from the existing controller. It enables the control engineers to ensure a new, pretrained policy satisfies the performance requirements and safety constraints of a real-world system, prior to interacting with it. While many methods have been developed for OPE, no study has evaluated which ones are suitable for building operational data, which are generated by deterministic policies and have limited coverage of the state-action space. After reviewing existing works and their assumptions, we adopted the approximate model (AM) method. Furthermore, we used bootstrapping to quantify uncertainty and correct for bias. In a simulation study, we evaluated the proposed approach on 10 policies pretrained with imitation learning. On average, the AM method estimated the energy and comfort costs with 1.84% and 14.1% error, respectively.},
  keywords={Machine learning, Energy system, Smart city}
}

We present an initial study of off-policy evaluation (OPE), a problem prerequisite to real-world reinforcement learning (RL), in the context of building control. OPE is the problem of estimating a policy's performance without running it on the actual system, using historical data from the existing controller. It enables the control engineers to ensure a new, pretrained policy satisfies the performance requirements and safety constraints of a real-world system, prior to interacting with it. While many methods have been developed for OPE, no study has evaluated which ones are suitable for building operational data, which are generated by deterministic policies and have limited coverage of the state-action space. After reviewing existing works and their assumptions, we adopted the approximate model (AM) method. Furthermore, we used bootstrapping to quantify uncertainty and correct for bias. In a simulation study, we evaluated the proposed approach on 10 policies pretrained with imitation learning. On average, the AM method estimated the energy and comfort costs with 1.84% and 14.1% error, respectively.

@article{2020_1J_survey,
title = "A Survey on Conic Relaxations of Optimal Power Flow Problem",
journal = "European Journal of Operational Research",
volume = "287",
number = "2",
pages = "391 - 409",
year = "2020",
issn = "0377-2217",
doi = "https://doi.org/10.1016/j.ejor.2020.01.034",
abstract = "Conic optimization has recently emerged as a powerful tool for designing tractable and guaranteed algorithms for power system operation. On the one hand, tractability is crucial due to the large size of modern electricity transmission grids. This is a result of the numerous interconnections that have been built over time. On the other hand, guarantees are needed to ensure reliability and safety for consumers at a time when power systems are growing in complexity. This is in large part due to the high penetration of renewable energy sources and the advent of electric vehicles. The aim of this paper is to review the latest literature in order to demonstrate the success of conic optimization when applied to power systems. The main focus is on how linear programming, second-order cone programming, and semidefinite programming can be used to address a central problem named the optimal power flow problem. We describe how they are used to design convex relaxations of this highly challenging non-convex optimization problem. We also show how sum-of-squares can be used to strengthen these relaxations. Finally, we present advances in first-order methods, interior-point methods, and nonconvex methods for solving conic optimization. Challenges for future research are also discussed.",
url_link = "http://www.sciencedirect.com/science/article/pii/S0377221720300552",
url_pdf = "Review_OPF_2019.pdf",
author = "Fariba Zohrizadeh and Cedric Josz and Ming Jin and Ramtin Madani and Javad Lavaei and Somayeh Sojoudi",
keywords = "Power system, Optimization, Graph theory"
}

Conic optimization has recently emerged as a powerful tool for designing tractable and guaranteed algorithms for power system operation. On the one hand, tractability is crucial due to the large size of modern electricity transmission grids. This is a result of the numerous interconnections that have been built over time. On the other hand, guarantees are needed to ensure reliability and safety for consumers at a time when power systems are growing in complexity. This is in large part due to the high penetration of renewable energy sources and the advent of electric vehicles. The aim of this paper is to review the latest literature in order to demonstrate the success of conic optimization when applied to power systems. The main focus is on how linear programming, second-order cone programming, and semidefinite programming can be used to address a central problem named the optimal power flow problem. We describe how they are used to design convex relaxations of this highly challenging non-convex optimization problem. We also show how sum-of-squares can be used to strengthen these relaxations. Finally, we present advances in first-order methods, interior-point methods, and nonconvex methods for solving conic optimization. Challenges for future research are also discussed.

@article{2020_0J_control,
title = "Control of Superheat of Organic Rankine Cycle under Transient Heat Source based on Deep Reinforcement Learning",
journal = "Applied Energy",
volume = "278",
pages = "115637",
year = "2020",
issn = "0306-2619",
doi = "https://doi.org/10.1016/j.apenergy.2020.115637",
url_link = "http://www.sciencedirect.com/science/article/pii/S0306261920311399",
author = "Xuan Wang and Rui Wang and Ming Jin and Gequn Shu and Hua Tian and Jiaying Pan",
keywords = "Energy system, Machine learning",
abstract = "The organic Rankine cycle (ORC) is a promising technology for engine waste heat recovery. During real-world operation, the engine working condition varies frequently to satisfy the power demand; thus, the transient nature of engine waste heat presents significant control challenges for the ORC. To control the superheat of the ORC precisely under a transient heat source, several optimal control methods have been used such as model predictive control and dynamic programing. However, most of them depend strongly on the accurate prediction of future disturbances. Deep reinforcement learning (DRL) is an artificial-intelligence algorithm that can overcome the aforementioned disadvantage, but the potential of DRL in control of thermodynamic systems has not yet been investigated. Thus, this paper proposes two DRL-based control methods for controlling the superheat of ORC under a transient heat source. One directly uses the DRL agent to learn the control strategy (DRL control), and the other uses the DRL agent to optimize the parameters of the proportional–integral–derivative (PID) controller (DRL-based PID control). Additionally, a switching mechanism between different DRL controllers is proposed for improving the training efficiency and enlarging the operation range of the controller. The results of this study indicate that the DRL agent can satisfactorily perform the control task and optimize the traditional controller under the trained and untrained transient heat source. Specifically, the DRL control can track the reference superheat with an average error of only 0.19 K, whereas that of the traditional PID control is 2.16 K. Furthermore, the proposed switching DRL control exhibits excellent tracking performance with an average error of only 0.21 K and robustness over a wide range of operation conditions. The successful application of DRL demonstrates its considerable potential for the control of thermodynamic systems, providing a useful reference and motivation for the application to other thermodynamic systems."
}

The organic Rankine cycle (ORC) is a promising technology for engine waste heat recovery. During real-world operation, the engine working condition varies frequently to satisfy the power demand; thus, the transient nature of engine waste heat presents significant control challenges for the ORC. To control the superheat of the ORC precisely under a transient heat source, several optimal control methods have been used such as model predictive control and dynamic programing. However, most of them depend strongly on the accurate prediction of future disturbances. Deep reinforcement learning (DRL) is an artificial-intelligence algorithm that can overcome the aforementioned disadvantage, but the potential of DRL in control of thermodynamic systems has not yet been investigated. Thus, this paper proposes two DRL-based control methods for controlling the superheat of ORC under a transient heat source. One directly uses the DRL agent to learn the control strategy (DRL control), and the other uses the DRL agent to optimize the parameters of the proportional–integral–derivative (PID) controller (DRL-based PID control). Additionally, a switching mechanism between different DRL controllers is proposed for improving the training efficiency and enlarging the operation range of the controller. The results of this study indicate that the DRL agent can satisfactorily perform the control task and optimize the traditional controller under the trained and untrained transient heat source. Specifically, the DRL control can track the reference superheat with an average error of only 0.19 K, whereas that of the traditional PID control is 2.16 K. Furthermore, the proposed switching DRL control exhibits excellent tracking performance with an average error of only 0.21 K and robustness over a wide range of operation conditions. The successful application of DRL demonstrates its considerable potential for the control of thermodynamic systems, providing a useful reference and motivation for the application to other thermodynamic systems.

@ARTICLE{2019_3J_powercyber,
  author={M. {Jin} and J. {Lavaei} and K. H. {Johansson}},
  journal={IEEE Transactions on Automatic Control}, 
  title={Power Grid AC-Based State Estimation: Vulnerability Analysis Against Cyber Attacks}, 
  year={2019},
  volume={64},
  number={5},
  pages={1784-1799},
  doi={10.1109/TAC.2018.2852774},
  abstract={To ensure grid efficiency and reliability, power system operators continuously monitor the operational characteristics of the grid through a critical process called state estimation (SE), which performs the task by filtering and fusing various measurements collected from grid sensors. This study analyzes the vulnerability of the key operation module, namely ac-based SE, against potential cyber attacks on data integrity, also known as false data injection attack (FDIA). A general form of FDIA can be formulated as an optimization problem, whose objective is to find a stealthy and sparse data injection vector on the sensor measurements with the aim of making the state estimate spurious and misleading. Due to the nonlinear ac measurement model and the cardinality constraint, the problem includes both continuous and discrete nonlinearities. To solve the FDIA problem efficiently, we propose a novel convexification framework based on semidefinite programming (SDP). By analyzing a globally optimal SDP solution, we delineate the “attackable region” for any given set of measurement types and grid topology, where the spurious state can be falsified by FDIA. Furthermore, we prove that the attack is stealthy and sparse, and derive performance bounds. Simulation results on various IEEE test cases indicate the efficacy of the proposed convexification approach. From the grid protection point of view, the results of this study can be used to design a security metric for the current practice against cyber attacks, redesign the bad data detection scheme, and inform proposals of grid hardening. From a theoretical point of view, the proposed framework can be used for other nonconvex problems in power systems and beyond.},
  url_pdf =    {FDIA_AC_2018.pdf},
  url_link =     {https://ieeexplore.ieee.org/document/8403288},
  keywords = {Power system, Optimization, Graph theory, Cybersecurity}
  }

To ensure grid efficiency and reliability, power system operators continuously monitor the operational characteristics of the grid through a critical process called state estimation (SE), which performs the task by filtering and fusing various measurements collected from grid sensors. This study analyzes the vulnerability of the key operation module, namely ac-based SE, against potential cyber attacks on data integrity, also known as false data injection attack (FDIA). A general form of FDIA can be formulated as an optimization problem, whose objective is to find a stealthy and sparse data injection vector on the sensor measurements with the aim of making the state estimate spurious and misleading. Due to the nonlinear ac measurement model and the cardinality constraint, the problem includes both continuous and discrete nonlinearities. To solve the FDIA problem efficiently, we propose a novel convexification framework based on semidefinite programming (SDP). By analyzing a globally optimal SDP solution, we delineate the “attackable region” for any given set of measurement types and grid topology, where the spurious state can be falsified by FDIA. Furthermore, we prove that the attack is stealthy and sparse, and derive performance bounds. Simulation results on various IEEE test cases indicate the efficacy of the proposed convexification approach. From the grid protection point of view, the results of this study can be used to design a security metric for the current practice against cyber attacks, redesign the bad data detection scheme, and inform proposals of grid hardening. From a theoretical point of view, the proposed framework can be used for other nonconvex problems in power systems and beyond.

@inproceedings{2019_2C_towards,
  title={Towards Robust and Scalable Power System State Estimation},
  author={Jin, Ming and Molybog, Igor and Mohammadi-Ghazi, Reza and Lavaei, Javad},
  booktitle={IEEE Conference on Decision and Control (CDC)},
  pages={3245--3252},
  year={2019},
  abstract={Power system state estimation is an important instance of data-driven decision making in power systems. Yet due to the nonconvexity of the problem, existing approaches based on local search methods are susceptible to spurious local minima. In this study, we propose a linear basis of representation that succinctly captures the topology of the network and enables an efficient two-stage estimation method when the amount of measured data is not too low. Furthermore, we develop a robustness metric called "mutual incoherence," which provides robustness guarantees in the presence of bad data. The proposed method demonstrates superior performance over existing methods in terms of both estimation accuracy and bad data detection for an array of benchmark systems. This technique is shown to be scalable to large systems with more than 13,000 nodes and can achieve an accurate estimation within a minute.},
  url_link={https://ieeexplore.ieee.org/document/9030243},
  keywords={Power system, Optimization},
  url_pdf={linear-SE_2019_2.pdf}
}

 

Power system state estimation is an important instance of data-driven decision making in power systems. Yet due to the nonconvexity of the problem, existing approaches based on local search methods are susceptible to spurious local minima. In this study, we propose a linear basis of representation that succinctly captures the topology of the network and enables an efficient two-stage estimation method when the amount of measured data is not too low. Furthermore, we develop a robustness metric called "mutual incoherence," which provides robustness guarantees in the presence of bad data. The proposed method demonstrates superior performance over existing methods in terms of both estimation accuracy and bad data detection for an array of benchmark systems. This technique is shown to be scalable to large systems with more than 13,000 nodes and can achieve an accurate estimation within a minute.

@INPROCEEDINGS{2019_2C_advanced,
title = "Advanced Building Control via Deep Reinforcement Learning",
booktitle = "International Conference on Applied Energy (ICAE)",
volume = "158",
pages = "6158 - 6163",
year = "2019",
issn = "1876-6102",
doi = "https://doi.org/10.1016/j.egypro.2019.01.494",
url_link = "http://www.sciencedirect.com/science/article/pii/S187661021930517X",
url_pdf={BuildingRL_ICAE_CR.pdf},
author = "Ruoxi Jia and Ming Jin and Kaiyu Sun and Tianzhen Hong and Costas Spanos",
keywords = "Smart city, Machine learning, Energy system",
abstract = "Building control is a challenging task, not least because of complex building dynamics ad multiple control objectives that are often conflicting. To tackle this challenge, we explore an end-to-end deep reinforcement learning paradigm, which learns an optimal control strategy to reduce energy consumption and to enhance occupant comfort from the data of building-controller interactions. Because real-world control policies need to be interpretable and efficient in learning, this work makes the following key contributions: (1) we investigated a systematic approach to encode expert knowledge in reinforcement learning through “experience replay” and/or “expert policy guidance”; (2) we proposed to regulate the smoothness property of the neural network to penalize the erratic behavior, which is found to dramatically stabilize the learning process and lead to interpretable control laws; (3) we established a virtual testbed for building control by combining the state-of-the-art building energy simulator EnergyPlus with a python environment to provide a systematic evaluation and comparison platform, which will not only further our understanding of the strengths and weaknesses of existing building control algorithms, but also suggest directions for future research. We experimentally verified our proposed deep reinforcement learning paradigm on the virtual testbed in case studies, which demonstrated promising results."
}

Building control is a challenging task, not least because of complex building dynamics ad multiple control objectives that are often conflicting. To tackle this challenge, we explore an end-to-end deep reinforcement learning paradigm, which learns an optimal control strategy to reduce energy consumption and to enhance occupant comfort from the data of building-controller interactions. Because real-world control policies need to be interpretable and efficient in learning, this work makes the following key contributions: (1) we investigated a systematic approach to encode expert knowledge in reinforcement learning through “experience replay” and/or “expert policy guidance”; (2) we proposed to regulate the smoothness property of the neural network to penalize the erratic behavior, which is found to dramatically stabilize the learning process and lead to interpretable control laws; (3) we established a virtual testbed for building control by combining the state-of-the-art building energy simulator EnergyPlus with a python environment to provide a systematic evaluation and comparison platform, which will not only further our understanding of the strengths and weaknesses of existing building control algorithms, but also suggest directions for future research. We experimentally verified our proposed deep reinforcement learning paradigm on the virtual testbed in case studies, which demonstrated promising results.

@INPROCEEDINGS{2019_1C_biscuit,
title = "BISCUIT: Building Intelligent System Customer Investment Tools",
booktitle = "International Conference on Applied Energy (ICAE)",
volume = "158",
pages = "6152 - 6157",
year = "2019",
issn = "1876-6102",
doi = "https://doi.org/10.1016/j.egypro.2019.01.495",
url_link = "http://www.sciencedirect.com/science/article/pii/S1876610219305181",
url_pdf = "BISCUIT_ICAE_CR.pdf",
author = "Ming Jin and Ruoxi Jia and Hari Prasanna Das and Wei Feng and Costas Spanos",
keywords = "Smart city, Optimization, Energy system",
abstract = "Smart buildings as human-cyber-physical systems (h-CPSs) are capable of providing intelligent services, such as indoor positioning, personalized lighting, demand-based heating ventilation and air-conditioning, and automatic fault detection and recovery, just to name a few. However, most buildings nowadays lack the basic components and infrastructure to support such services. The investment decision of intelligent system design and retrofit can be a daunting task, because it involves both hardware (sensors, actuators, servers) and software (operating systems, service algorithms), which have issues of compatibility, functionality constraints, and opportunities of co-design of synergy. This work proposes a user-oriented investment decision toolset aimed at handling the complexity of exploration in the large design space and to enhance cost-effectiveness, energy efficiency, and human-centric values. The toolset is demonstrated in a case study to retrofit a medium-sized building, where it is shown to propose a design that significantly lowers the overall investment cost while achieving user specifications."
}

Smart buildings as human-cyber-physical systems (h-CPSs) are capable of providing intelligent services, such as indoor positioning, personalized lighting, demand-based heating ventilation and air-conditioning, and automatic fault detection and recovery, just to name a few. However, most buildings nowadays lack the basic components and infrastructure to support such services. The investment decision of intelligent system design and retrofit can be a daunting task, because it involves both hardware (sensors, actuators, servers) and software (operating systems, service algorithms), which have issues of compatibility, functionality constraints, and opportunities of co-design of synergy. This work proposes a user-oriented investment decision toolset aimed at handling the complexity of exploration in the large design space and to enhance cost-effectiveness, energy efficiency, and human-centric values. The toolset is demonstrated in a case study to retrofit a medium-sized building, where it is shown to propose a design that significantly lowers the overall investment cost while achieving user specifications.

@article{2019_0J_provision,
title = "Provision of secondary frequency regulation by coordinated dispatch of industrial loads and thermal power plants",
journal = "Applied Energy",
volume = "241",
pages = "302 - 312",
year = "2019",
issn = "0306-2619",
doi = "https://doi.org/10.1016/j.apenergy.2019.03.025",
url_link = "http://www.sciencedirect.com/science/article/pii/S0306261919304234",
author = "Yi Bao and Jian Xu and Wei Feng and Yuanzhang Sun and Siyang Liao and Rongxin Yin and Yazhou Jiang and Ming Jin and Chris Marnay",
keywords = "Control theory, Power system",
abstract = "Demand responsive industrial loads with high thermal inertia have potential to provide ancillary service for frequency regulation in the power market. To capture the benefit, this study proposes a new hierarchical framework to coordinate the demand responsive industrial loads with thermal power plants in an industrial park for secondary frequency control. In the proposed framework, demand responsive loads and generating resources are coordinated for optimal dispatch in two-time scales: (1) the regulation reserve of the industrial park is optimally scheduled in a day-ahead manner. The stochastic regulation signal is replaced by the specific extremely trajectories. Furthermore, the extremely trajectories are achieved by the day-ahead predicted regulation mileage. The resulting benefit is to transform the stochastic reserve scheduling problem into a deterministic optimization; (2) a model predictive control strategy is proposed to dispatch the industry park in real time with an objective to maximize the revenue. The proposed technology is tested using a real-world industrial electrolysis power system based upon Pennsylvania, Jersey, and Maryland (PJM) power market. Various scenarios are simulated to study the performance of the proposed approach to enable industry parks to provide ancillary service into the power market. The simulation results indicate that an industrial park with a capacity of 500 MW can provide up to 40 MW ancillary service for participation in the secondary frequency regulation. The proposed strategy is demonstrated to be capable of maintaining the economic and secure operation of the industrial park while satisfying performance requirements from the real world regulation market."
}

Demand responsive industrial loads with high thermal inertia have potential to provide ancillary service for frequency regulation in the power market. To capture the benefit, this study proposes a new hierarchical framework to coordinate the demand responsive industrial loads with thermal power plants in an industrial park for secondary frequency control. In the proposed framework, demand responsive loads and generating resources are coordinated for optimal dispatch in two-time scales: (1) the regulation reserve of the industrial park is optimally scheduled in a day-ahead manner. The stochastic regulation signal is replaced by the specific extremely trajectories. Furthermore, the extremely trajectories are achieved by the day-ahead predicted regulation mileage. The resulting benefit is to transform the stochastic reserve scheduling problem into a deterministic optimization; (2) a model predictive control strategy is proposed to dispatch the industry park in real time with an objective to maximize the revenue. The proposed technology is tested using a real-world industrial electrolysis power system based upon Pennsylvania, Jersey, and Maryland (PJM) power market. Various scenarios are simulated to study the performance of the proposed approach to enable industry parks to provide ancillary service into the power market. The simulation results indicate that an industrial park with a capacity of 500 MW can provide up to 40 MW ancillary service for participation in the secondary frequency regulation. The proposed strategy is demonstrated to be capable of maintaining the economic and secure operation of the industrial park while satisfying performance requirements from the real world regulation market.

@article{2019_0J_personal,
title = "Personal thermal comfort models with wearable sensors",
journal = "Building and Environment",
volume = "162",
pages = "106281",
year = "2019",
issn = "0360-1323",
doi = "https://doi.org/10.1016/j.buildenv.2019.106281",
url_link = "http://www.sciencedirect.com/science/article/pii/S0360132319304913",
author = "Shichao Liu and Stefano Schiavon and Hari Prasanna Das and Ming Jin and Costas J. Spanos",
keywords = "Smart city, Machine learning, Energy system",
abstract = "A personal comfort model is an approach to thermal comfort modeling, for thermal environmental design and control, that predicts an individual's thermal comfort response, instead of the average response of a large population. We developed personal thermal comfort models using lab grade wearable in normal daily activities. We collected physiological signals (e.g., skin temperature, heart rate) of 14 subjects (6 female and 8 male adults) and environmental parameters (e.g., air temperature, relative humidity) for 2–4 weeks (at least 20 h per day). Then we trained 14 models for each subject with different machine-learning algorithms to predict their thermal preference. The results show that the median prediction power could be up to 24%/78%/0.79 (Cohen's kappa/accuracy/AUC) with all features considered. The median prediction power reaches 21%/71%/0.7 after 200 subjective votes. We explored the importance of different features on the prediction performance by considering all subjects in one dataset. When all features included for the entire dataset, personal comfort models can generate the highest performance of 35%/76%/0.80 by the most predictive algorithm. Personal comfort models display the highest prediction power when occupants' thermal sensations is outside thermal neutrality. Skin temperature measured at the ankle is more predictive than measured at the wrist. We suggest that Cohen's kappa or AUC should be employed to assess the performance of personal thermal comfort models for imbalanced datasets due to the capacity to exclude random success."
}

A personal comfort model is an approach to thermal comfort modeling, for thermal environmental design and control, that predicts an individual's thermal comfort response, instead of the average response of a large population. We developed personal thermal comfort models using lab grade wearable in normal daily activities. We collected physiological signals (e.g., skin temperature, heart rate) of 14 subjects (6 female and 8 male adults) and environmental parameters (e.g., air temperature, relative humidity) for 2–4 weeks (at least 20 h per day). Then we trained 14 models for each subject with different machine-learning algorithms to predict their thermal preference. The results show that the median prediction power could be up to 24%/78%/0.79 (Cohen's kappa/accuracy/AUC) with all features considered. The median prediction power reaches 21%/71%/0.7 after 200 subjective votes. We explored the importance of different features on the prediction performance by considering all subjects in one dataset. When all features included for the entire dataset, personal comfort models can generate the highest performance of 35%/76%/0.80 by the most predictive algorithm. Personal comfort models display the highest prediction power when occupants' thermal sensations is outside thermal neutrality. Skin temperature measured at the ankle is more predictive than measured at the wrist. We suggest that Cohen's kappa or AUC should be employed to assess the performance of personal thermal comfort models for imbalanced datasets due to the capacity to exclude random success.

@ARTICLE{2018_3J_occ,
  author={M. {Jin} and N. {Bekiaris-Liberis} and K. {Weekly} and C. J. {Spanos} and A. M. {Bayen}},
  journal={IEEE Transactions on Automation Science and Engineering}, 
  title={Occupancy Detection via Environmental Sensing}, 
  year={2018},
  volume={15},
  number={2},
  pages={443-455},
  doi={10.1109/TASE.2016.2619720},
  abstract={Sensing by proxy (SbP) is proposed in this paper as a sensing paradigm for occupancy detection, where the inference is based on “proxy” measurements such as temperature and CO2 concentrations. The effects of occupants on indoor environments are captured by constitutive models comprising a coupled partial differential equation-ordinary differential equation system that exploits the spatial and physical features. Sensor fusion of multiple environmental parameters is enabled in the proposed framework. We report on experiments conducted under simulated conditions and real-life circumstances, when the variation of occupancy follows a schedule as the ground truth. The inference of the number of occupants in the room based on CO2 concentration at the air return and air supply vents by our approach achieves an overall mean squared error of 0.6044 (fractional person), while the best alternative by Bayes net is 1.2061 (fractional person). Results from the projected ventilation analysis show that SbP can potentially save 55% of total ventilation compared with the traditional fixed schedule ventilation strategy, while at the same time maintain a reasonably comfort profile for the occupants.},
  url_link = "https://ieeexplore.ieee.org/document/7742900",
  url_pdf =    {occ_tase.pdf},
  url_synopsis={SbP_synopsis.pdf},
  url_code = {https://github.com/jinming99/Sensing-by-proxy},
  keywords={Control theory, Smart city, Data mining}}
  

Sensing by proxy (SbP) is proposed in this paper as a sensing paradigm for occupancy detection, where the inference is based on “proxy” measurements such as temperature and CO2 concentrations. The effects of occupants on indoor environments are captured by constitutive models comprising a coupled partial differential equation-ordinary differential equation system that exploits the spatial and physical features. Sensor fusion of multiple environmental parameters is enabled in the proposed framework. We report on experiments conducted under simulated conditions and real-life circumstances, when the variation of occupancy follows a schedule as the ground truth. The inference of the number of occupants in the room based on CO2 concentration at the air return and air supply vents by our approach achieves an overall mean squared error of 0.6044 (fractional person), while the best alternative by Bayes net is 1.2061 (fractional person). Results from the projected ventilation analysis show that SbP can potentially save 55% of total ventilation compared with the traditional fixed schedule ventilation strategy, while at the same time maintain a reasonably comfort profile for the occupants.

@article{2018_2J_microgrid,
title = "Microgrid to enable optimal distributed energy retail and end-user demand response",
journal = "Applied Energy",
volume = "210",
pages = "1321 - 1335",
year = "2018",
issn = "0306-2619",
doi = "https://doi.org/10.1016/j.apenergy.2017.05.103",
url_link = "http://www.sciencedirect.com/science/article/pii/S0306261917306062",
url_pdf =    {mr-pod.pdf},
url_conference ={der_isgt.pdf},
author = "Ming Jin and Wei Feng and Chris Marnay and Costas Spanos",
keywords = "Power system, Energy system",
abstract = "In the face of unprecedented challenges in environmental sustainability and grid resilience, there is an increasingly held consensus regarding the adoption of distributed and renewable energy resources such as microgrids (MGs), and the utilization of flexible electric loads by demand response (DR) to potentially drive a necessary paradigm shift in energy production and consumption patterns. However, the potential value of distributed generation and demand flexibility has not yet been fully realized in the operation of MGs. This study investigates the pricing and operation strategy with DR for a MG retailer in an integrated energy system (IES). Based on co-optimizing retail rates and MG dispatch formulated as a mixed integer quadratic programming (MIQP) problem, our model devises a dynamic pricing scheme that reflects the cost of generation and promotes DR, in tandem with an optimal dispatch plan that exploits spark spread and facilitates the integration of renewables, resulting in improved retailer profits and system stability. Main issues like integrated energy coupling and customer bill reduction are addressed during pricing to ensure rates competitiveness and customer protection. By evaluating on real datasets, the system is demonstrated to optimally coordinate storage, renewables, and combined heat and power (CHP), reduce carbon dioxide emission while maintaining profits, and effectively alleviate the PV curtailment problem. The model can be used by retailers and MG operators to optimize their operations, as well as regulators to design new utility rates in support of the ongoing transformation of energy systems."
}

In the face of unprecedented challenges in environmental sustainability and grid resilience, there is an increasingly held consensus regarding the adoption of distributed and renewable energy resources such as microgrids (MGs), and the utilization of flexible electric loads by demand response (DR) to potentially drive a necessary paradigm shift in energy production and consumption patterns. However, the potential value of distributed generation and demand flexibility has not yet been fully realized in the operation of MGs. This study investigates the pricing and operation strategy with DR for a MG retailer in an integrated energy system (IES). Based on co-optimizing retail rates and MG dispatch formulated as a mixed integer quadratic programming (MIQP) problem, our model devises a dynamic pricing scheme that reflects the cost of generation and promotes DR, in tandem with an optimal dispatch plan that exploits spark spread and facilitates the integration of renewables, resulting in improved retailer profits and system stability. Main issues like integrated energy coupling and customer bill reduction are addressed during pricing to ensure rates competitiveness and customer protection. By evaluating on real datasets, the system is demonstrated to optimally coordinate storage, renewables, and combined heat and power (CHP), reduce carbon dioxide emission while maintaining profits, and effectively alleviate the PV curtailment problem. The model can be used by retailers and MG operators to optimize their operations, as well as regulators to design new utility rates in support of the ongoing transformation of energy systems.

@article{2018_2J_automated,
title = "Automated mobile sensing: Towards high-granularity agile indoor environmental quality monitoring",
journal = "Building and Environment",
volume = "127",
pages = "268 - 276",
year = "2018",
issn = "0360-1323",
doi = "https://doi.org/10.1016/j.buildenv.2017.11.003",
url_link = "http://www.sciencedirect.com/science/article/pii/S0360132317305012",
url_code="https://github.com/jinming99/IEQbot",
author = "Ming Jin and Shichao Liu and Stefano Schiavon and Costas Spanos",
note={<a style="color:#FF0000" href="https://www.journals.elsevier.com/building-and-environment/news/building-and-environment-best-paper-awards-previous-years">Best paper award (3 out of >3000)</a>},
keywords = "Smart city, Energy system, Data mining",
abstract = "Indoor environmental quality (IEQ) is a critical aspect of the built environment to ensure occupant health, comfort, well-being and productivity. Existing IEQ monitoring approaches rely on sensor networks deployed at selected locations to collect environmental measurements, and are limited in scale and adaptability due to infrastructure cost and maintenance requirement. To enable high-granularity IEQ monitoring with agile adaption to the dynamic indoor environment, we propose an “automated mobile sensing” system that dispatches a sensor-rich navigation-capable robot to actively survey the indoor space. Data collected in this fashion is sparse in the joint temporal and spatial domain, and cannot be used directly for IEQ evaluation. To deal with this special characteristics, we developed a spatio-temporal interpolation algorithm to capture the global trend and local variation in order to use the data efficiently to reconstruct the IEQ dynamics. We compared the performance of the automated mobile sensing with a dense sensor network in a laboratory where we measured the air-change effectiveness (ASHRAE standard 129) for four different conditions. Results indicate that automated mobile sensing is able to accurately estimate the parameters with a minimal sensor cost and calibration effort. Potential applications of this system include indoor thermal comfort, lighting, indoor air quality and acoustic monitoring, pollutant source identification, and building commissioning. We shared publicly the source codes for robot control, sensor setup, and interpolation algorithm to encourage comparison study and further development."
}

Indoor environmental quality (IEQ) is a critical aspect of the built environment to ensure occupant health, comfort, well-being and productivity. Existing IEQ monitoring approaches rely on sensor networks deployed at selected locations to collect environmental measurements, and are limited in scale and adaptability due to infrastructure cost and maintenance requirement. To enable high-granularity IEQ monitoring with agile adaption to the dynamic indoor environment, we propose an “automated mobile sensing” system that dispatches a sensor-rich navigation-capable robot to actively survey the indoor space. Data collected in this fashion is sparse in the joint temporal and spatial domain, and cannot be used directly for IEQ evaluation. To deal with this special characteristics, we developed a spatio-temporal interpolation algorithm to capture the global trend and local variation in order to use the data efficiently to reconstruct the IEQ dynamics. We compared the performance of the automated mobile sensing with a dense sensor network in a laboratory where we measured the air-change effectiveness (ASHRAE standard 129) for four different conditions. Results indicate that automated mobile sensing is able to accurately estimate the parameters with a minimal sensor cost and calibration effort. Potential applications of this system include indoor thermal comfort, lighting, indoor air quality and acoustic monitoring, pollutant source identification, and building commissioning. We shared publicly the source codes for robot control, sensor setup, and interpolation algorithm to encourage comparison study and further development.

@INPROCEEDINGS{2018_2C_control,
  author={M. {Jin} and J. {Lavaei}},
  booktitle={IEEE Conference on Decision and Control (CDC)}, 
  title={Control-Theoretic Analysis of Smoothness for Stability-Certified Reinforcement Learning}, 
  year={2018},
  volume={},
  number={},
  pages={6840-6847},
  doi={10.1109/CDC.2018.8618996},
  abstract={It is critical to obtain stability certificate before deploying reinforcement learning in real-world mission-critical systems. This study justifies the intuition that smoothness (i.e., small changes in inputs lead to small changes in outputs) is an important property for stability-certified reinforcement learning from a control-theoretic perspective. The smoothness margin can be obtained by solving a feasibility problem based on semi-definite programming for both linear and nonlinear dynamical systems, and it does not need to access the exact parameters of the learned controllers. Numerical evaluation on nonlinear and decentralized frequency control for large-scale power grids demonstrates that the smoothness margin can certify stability during both exploration and deployment for (deep) neural-network policies, which substantially surpass nominal controllers in performance. The study opens up new opportunities for robust Lipschitz continuous policy learning.},
  url_pdf={smoothRL_cdc_tech.pdf},
  url_link={https://ieeexplore.ieee.org/document/8618996},
  keywords={Control theory, Machine learning}}
  

It is critical to obtain stability certificate before deploying reinforcement learning in real-world mission-critical systems. This study justifies the intuition that smoothness (i.e., small changes in inputs lead to small changes in outputs) is an important property for stability-certified reinforcement learning from a control-theoretic perspective. The smoothness margin can be obtained by solving a feasibility problem based on semi-definite programming for both linear and nonlinear dynamical systems, and it does not need to access the exact parameters of the learned controllers. Numerical evaluation on nonlinear and decentralized frequency control for large-scale power grids demonstrates that the smoothness margin can certify stability during both exploration and deployment for (deep) neural-network policies, which substantially surpass nominal controllers in performance. The study opens up new opportunities for robust Lipschitz continuous policy learning.

@ARTICLE{2018_1J_robust,
  author={I. C. {Konstantakopoulos} and L. J. {Ratliff} and M. {Jin} and S. S. {Sastry} and C. J. {Spanos}},
  journal={IEEE Transactions on Control Systems Technology}, 
  title={A Robust Utility Learning Framework via Inverse Optimization}, 
  year={2018},
  volume={26},
  number={3},
  pages={954-970},
  doi={10.1109/TCST.2017.2699163},
  abstract={In many smart infrastructure applications, flexibility in achieving sustainability goals can be gained by engaging end users. However, these users often have heterogeneous preferences that are unknown to the decision maker tasked with improving operational efficiency. Modeling user interaction as a continuous game between noncooperative players, we propose a robust parametric utility learning framework that employs constrained feasible generalized least squares estimation with heteroskedastic inference. To improve forecasting performance, we extend the robust utility learning scheme by employing bootstrapping with bagging, bumping, and gradient boosting ensemble methods. Moreover, we estimate the noise covariance, which provides approximated correlations between players, which we leverage to develop a novel correlated utility learning framework. We apply the proposed methods both to a toy example arising from Bertrand-Nash competition between two firms and to data from a social game experiment designed to encourage energy efficient behavior among smart building occupants. Using occupant voting data for shared resources such as lighting, we simulate the game defined by the estimated utility functions to demonstrate the performance of the proposed methods.},
  url_pdf = {robust_inverse_learning.pdf},
  keywords={Smart city, Optimization, Machine learning, Game theory},
  url_link = {https://ieeexplore.ieee.org/document/7932982}}
  

In many smart infrastructure applications, flexibility in achieving sustainability goals can be gained by engaging end users. However, these users often have heterogeneous preferences that are unknown to the decision maker tasked with improving operational efficiency. Modeling user interaction as a continuous game between noncooperative players, we propose a robust parametric utility learning framework that employs constrained feasible generalized least squares estimation with heteroskedastic inference. To improve forecasting performance, we extend the robust utility learning scheme by employing bootstrapping with bagging, bumping, and gradient boosting ensemble methods. Moreover, we estimate the noise covariance, which provides approximated correlations between players, which we leverage to develop a novel correlated utility learning framework. We apply the proposed methods both to a toy example arising from Bertrand-Nash competition between two firms and to data from a social game experiment designed to encourage energy efficient behavior among smart building occupants. Using occupant voting data for shared resources such as lighting, we simulate the game defined by the estimated utility functions to demonstrate the performance of the proposed methods.

@article{2018_1J_review,
  title={A Review of Microgrid Development in the United States--A Decade of Progress on Policies, Demonstrations, Controls, and Software Tools},
  author={Feng, Wei and Jin, Ming and Liu, Xu and Bao, Yi and Marnay, Chris and Yao, Cheng and Yu, Jiancheng},
  journal={Applied energy},
  volume={228},
  pages={1656--1668},
  year={2018},
  publisher={Elsevier},
  abstract={Microgrids have become increasingly popular in the United States. Supported by favorable federal and local policies, microgrid projects can provide greater energy stability and resilience within a project site or community. This paper reviews major federal, state, and utility-level policies driving microgrid development in the United States. Representative U.S. demonstration projects are selected and their technical characteristics and non-technical features are introduced. The paper discusses trends in the technology development of microgrid systems as well as microgrid control methods and interactions within the electricity market. Software tools for microgrid design, planning, and performance analysis are illustrated with each tool’s core capability. Finally, the paper summarizes the successes and lessons learned during the recent expansion of the U.S. microgrid industry that may serve as a reference for other countries developing their own microgrid industries.},
  url_pdf =    {MG_Review.pdf},
  url_link =     {https://www.sciencedirect.com/science/article/abs/pii/S0306261918309644},
  keywords = "Power system"
}

Microgrids have become increasingly popular in the United States. Supported by favorable federal and local policies, microgrid projects can provide greater energy stability and resilience within a project site or community. This paper reviews major federal, state, and utility-level policies driving microgrid development in the United States. Representative U.S. demonstration projects are selected and their technical characteristics and non-technical features are introduced. The paper discusses trends in the technology development of microgrid systems as well as microgrid control methods and interactions within the electricity market. Software tools for microgrid design, planning, and performance analysis are illustrated with each tool’s core capability. Finally, the paper summarizes the successes and lessons learned during the recent expansion of the U.S. microgrid industry that may serve as a reference for other countries developing their own microgrid industries.

@ARTICLE{2018_1J_design,
  author={R. {Jia} and B. {Jin} and M. {Jin} and Y. {Zhou} and I. C. {Konstantakopoulos} and H. {Zou} and J. {Kim} and D. {Li} and W. {Gu} and R. {Arghandeh} and P. {Nuzzo} and S. {Schiavon} and A. L. {Sangiovanni-Vincentelli} and C. J. {Spanos}},
  journal={Proceedings of the IEEE}, 
  title={Design Automation for Smart Building Systems}, 
  year={2018},
  volume={106},
  number={9},
  pages={1680-1699},
  doi={10.1109/JPROC.2018.2856932},
  url_link={https://ieeexplore.ieee.org/document/8466990},
  url_pdf={Design_IEEEProc.pdf},
  abstract={Smart buildings today are aimed at providing safe, healthy, comfortable, affordable, and beautiful spaces in a carbon and energy-efficient way. They are emerging as complex cyber-physical systems with humans in the loop. Cost, the need to cope with increasing functional complexity, flexibility, fragmentation of the supply chain, and time-to-market pressure are rendering the traditional heuristic and ad hoc design paradigms inefficient and insufficient for the future. In this paper, we present a platform-based methodology for smart building design. Platform-based design (PBD) promotes the reuse of hardware and software on shared infrastructures, enables rapid prototyping of applications, and involves extensive exploration of the design space to optimize design performance. In this paper, we identify, abstract, and formalize components of smart buildings, and present a design flow that maps high-level specifications of desired building applications to their physical implementations under the PBD framework. A case study on the design of on-demand heating, ventilation, and air conditioning (HVAC) systems is presented to demonstrate the use of PBD.},
  keywords={Cyber-physical system, Optimization, Energy system}}

Smart buildings today are aimed at providing safe, healthy, comfortable, affordable, and beautiful spaces in a carbon and energy-efficient way. They are emerging as complex cyber-physical systems with humans in the loop. Cost, the need to cope with increasing functional complexity, flexibility, fragmentation of the supply chain, and time-to-market pressure are rendering the traditional heuristic and ad hoc design paradigms inefficient and insufficient for the future. In this paper, we present a platform-based methodology for smart building design. Platform-based design (PBD) promotes the reuse of hardware and software on shared infrastructures, enables rapid prototyping of applications, and involves extensive exploration of the design space to optimize design performance. In this paper, we identify, abstract, and formalize components of smart buildings, and present a design flow that maps high-level specifications of desired building applications to their physical implementations under the PBD framework. A case study on the design of on-demand heating, ventilation, and air conditioning (HVAC) systems is presented to demonstrate the use of PBD.

@article{2018_1J_cascade,
title = "Cascade energy optimization for waste heat recovery in distributed energy systems",
journal = "Applied Energy",
volume = "230",
pages = "679 - 695",
year = "2018",
issn = "0306-2619",
doi = "https://doi.org/10.1016/j.apenergy.2018.08.124",
url_link = "http://www.sciencedirect.com/science/article/pii/S0306261918312984",
author = "Xuan Wang and Ming Jin and Wei Feng and Gequn Shu and Hua Tian and Youcai Liang",
keywords = "Optimization, Energy system",
abstract = "The efficiency of distributed energy systems can be significantly increased through waste heat recovery from industry or power generation. The technologies used for this process are typically dependent on the quality and temperature grades of waste heat. To maximize the efficiency of cascade heat utilization, it is important to optimize the choice of waste heat recovery technologies and their operation. In this paper, a detailed mixed integer linear programming optimization model is proposed for waste heat recovery in a district-scale microgrid. The model can distinguish waste heat quality for planning and operation optimization of distributed energy systems. Heat utilization technologies are formulated in this developed model and categorized in different temperature grades. The developed model is validated using four typical cases under different settings of system operation and business models. It is found that the optimization model, by distinguishing waste heat temperature, can increase energy cost savings by around 5%, compared to models that do not consider waste heat temperature grades. Additionally, the results indicate that the developed model can provide more realistic configuration and technologies dispatch."
}

The efficiency of distributed energy systems can be significantly increased through waste heat recovery from industry or power generation. The technologies used for this process are typically dependent on the quality and temperature grades of waste heat. To maximize the efficiency of cascade heat utilization, it is important to optimize the choice of waste heat recovery technologies and their operation. In this paper, a detailed mixed integer linear programming optimization model is proposed for waste heat recovery in a district-scale microgrid. The model can distinguish waste heat quality for planning and operation optimization of distributed energy systems. Heat utilization technologies are formulated in this developed model and categorized in different temperature grades. The developed model is validated using four typical cases under different settings of system operation and business models. It is found that the optimization model, by distinguishing waste heat temperature, can increase energy cost savings by around 5%, compared to models that do not consider waste heat temperature grades. Additionally, the results indicate that the developed model can provide more realistic configuration and technologies dispatch.

@inproceedings{2018_0C_review,
title = "Review of Microgrid Development in the United States and China and Lessons Learned for China",
booktitle = "Applied Energy Symposium and Forum, Renewable Energy Integration with Mini/Microgrids (REM)",
volume = "145",
pages = "217 - 222",
year = "2018",
issn = "1876-6102",
doi = "https://doi.org/10.1016/j.egypro.2018.04.038",
url_link = "http://www.sciencedirect.com/science/article/pii/S1876610218300444",
url_pdf ="REM2017.pdf",
author = "Jiancheng Yu and Chris Marnay and Ming Jin and Cheng Yao and Xu Liu and Wei Feng",
keywords = "Energy system, Power system",
abstract = "The U.S. has emerged as the microgrid development leader with around 40% of worldwide capacity. Over the last decade, demonstrations have been executed by a mix of civilian federal, military, private, and local government entities. While their motivations are mixed, resilience became the focus following Superstorm Sandy in 2012, especially in the highly active northeast states. This paper describes U.S. microgrid demonstrations. Then it shows China’s effort to develop microgrids and compares the difference between U.S. and Chinese projects. Finally, based on U.S. experience, recommendations are provided for Chinese microgrids."
}

The U.S. has emerged as the microgrid development leader with around 40% of worldwide capacity. Over the last decade, demonstrations have been executed by a mix of civilian federal, military, private, and local government entities. While their motivations are mixed, resilience became the focus following Superstorm Sandy in 2012, especially in the highly active northeast states. This paper describes U.S. microgrid demonstrations. Then it shows China’s effort to develop microgrids and compares the difference between U.S. and Chinese projects. Finally, based on U.S. experience, recommendations are provided for Chinese microgrids.

@INPROCEEDINGS{2017_4C_inverse,
  title={Inverse reinforcement learning via deep gaussian process},
  author={Jin, Ming and Damianou, Andreas and Abbeel, Pieter and Spanos, Costas},
  booktitle={Conference on Uncertainty in Artificial Intelligence (UAI)},
  url_pdf={dgpirl_uai.pdf},
  url_supplementary={dgpirl_supp.pdf},
  url_code={https://github.com/jinming99/DGP-IRL},
  url_link={http://auai.org/uai2017/proceedings/papers/48.pdf},
  abstract={We propose a new approach to inverse reinforcement learning (IRL) based on the deep Gaussian process (deep GP) model, which is capable of learning complicated reward structures with few demonstrations. Our model stacks multiple latent GP layers to learn abstract representations of the state feature space, which is linked to the demonstrations through the Maximum Entropy learning framework. Incorporating the IRL engine into the nonlinear latent structure renders existing deep GP inference approaches intractable. To tackle this, we develop a non-standard variational approximation framework which extends previous inference schemes. This allows for approximate Bayesian treatment of the feature space and guards against overfitting. Carrying out representation and inverse reinforcement learning simultaneously within our model outperforms state-of-the-art approaches, as we demonstrate with experiments on standard benchmarks ("object world","highway driving") and a new benchmark ("binary world").},
  keywords={Machine learning, Optimization},
  note={<font style="color:#FF0000">(oral presentation)</font>},
  year={2017}
}

We propose a new approach to inverse reinforcement learning (IRL) based on the deep Gaussian process (deep GP) model, which is capable of learning complicated reward structures with few demonstrations. Our model stacks multiple latent GP layers to learn abstract representations of the state feature space, which is linked to the demonstrations through the Maximum Entropy learning framework. Incorporating the IRL engine into the nonlinear latent structure renders existing deep GP inference approaches intractable. To tackle this, we develop a non-standard variational approximation framework which extends previous inference schemes. This allows for approximate Bayesian treatment of the feature space and guards against overfitting. Carrying out representation and inverse reinforcement learning simultaneously within our model outperforms state-of-the-art approaches, as we demonstrate with experiments on standard benchmarks ("object world","highway driving") and a new benchmark ("binary world").

@ARTICLE{2017_3J_virtual,
  author={M. {Jin} and R. {Jia} and C. J. {Spanos}},
  journal={IEEE Transactions on Mobile Computing}, 
  title={Virtual Occupancy Sensing: Using Smart Meters to Indicate Your Presence}, 
  year={2017},
  volume={16},
  number={11},
  pages={3264-3277},
  doi={10.1109/TMC.2017.2684806},
  abstract={Occupancy detection for buildings is crucial to improving energy efficiency, user comfort, and space utility. However, existing methods require dedicated system setup, continuous calibration, and frequent maintenance. With the instrumentation of electricity meters in millions of homes and offices, however, power measurement presents a unique opportunity for a non-intrusive and cost-effective way to detect occupant presence. This study develops solutions to the problems when no data or limited data is available for training, as motivated by difficulties in ground truth collection. Experimental evaluations on data from both residential and commercial buildings indicate that the proposed methods for binary occupancy detection are nearly as accurate as models learned with sufficient data, with accuracies of approximately 78 to 93 percent for residences and 90 percent for offices. This study shows that power usage contains valuable and sensitive user information, demonstrating a virtual occupancy sensing approach with minimal system calibration and setup.},
  url_pdf = {smart_meter_presence.pdf},
  url_supplementary = {smart_meter_presence_supp.pdf},
  url_link = {https://ieeexplore.ieee.org/document/7882676},
  url_media={https://spectrum.ieee.org/view-from-the-valley/energy/the-smarter-grid/what-does-your-smart-meter-know-about-you},
  keywords = "Smart city, Data mining, Machine learning",
  note={<a style="color:#FF0000" href="https://spectrum.ieee.org/view-from-the-valley/energy/the-smarter-grid/what-does-your-smart-meter-know-about-you">(Featured in 'IEEE Spectrum')</a>}
  }
  
  

Occupancy detection for buildings is crucial to improving energy efficiency, user comfort, and space utility. However, existing methods require dedicated system setup, continuous calibration, and frequent maintenance. With the instrumentation of electricity meters in millions of homes and offices, however, power measurement presents a unique opportunity for a non-intrusive and cost-effective way to detect occupant presence. This study develops solutions to the problems when no data or limited data is available for training, as motivated by difficulties in ground truth collection. Experimental evaluations on data from both residential and commercial buildings indicate that the proposed methods for binary occupancy detection are nearly as accurate as models learned with sufficient data, with accuracies of approximately 78 to 93 percent for residences and 90 percent for offices. This study shows that power usage contains valuable and sensitive user information, demonstrating a virtual occupancy sensing approach with minimal system calibration and setup.

@INPROCEEDINGS{2017_3C_semidefinite,
  author={M. {Jin} and J. {Lavaei} and K. {Johansson}},
  booktitle={Annual Allerton Conference on Communication, Control, and Computing (Allerton)}, 
  title={A semidefinite programming relaxation under false data injection attacks against power grid AC state estimation}, 
  year={2017},
  volume={},
  number={},
  pages={236-243},
  doi={10.1109/ALLERTON.2017.8262743},
  url_link={https://ieeexplore.ieee.org/document/8262743},
  url_pdf={cyberattack_allerton.pdf},
  abstract={The integration of sensing and information technology renders the power grid susceptible to cyber-attacks. To understand how vulnerable the state estimator is, we study its behavior under the worst attacks possible. A general false data injection attack (FDIA) based on the AC model is formulated, where the attacker manipulates sensor measurements to mislead the system operator to make decisions based on a falsified state. To stage such an attack, the optimization problem incorporates constraints of limited resources (allowing only a limited number of measurements to be altered), and stealth operation (ensuring the cyber hack cannot be identified by the bad data detection algorithm). Due to the nonlinear AC power flow model and combinatorial selection of compromised sensors, the problem is nonconvex and cannot be solved in polynomial time; however, it is shown that convexification of the original problem based on a semidefinite programming (SDP) relaxation and a sparsity penalty is able to recover a near-optimal solution. This represents the first study to solve the AC-based FDIA. Simulations on a 30-bus system illustrate that the proposed attack requires only sparse sensor manipulation and remains stealthy from the residual-based bad data detection mechanism. In light of the analysis, this study raises new challenges on grid defense mechanism and attack detection strategy.},
  keywords={Optimization, Cybersecurity, Power system}}

The integration of sensing and information technology renders the power grid susceptible to cyber-attacks. To understand how vulnerable the state estimator is, we study its behavior under the worst attacks possible. A general false data injection attack (FDIA) based on the AC model is formulated, where the attacker manipulates sensor measurements to mislead the system operator to make decisions based on a falsified state. To stage such an attack, the optimization problem incorporates constraints of limited resources (allowing only a limited number of measurements to be altered), and stealth operation (ensuring the cyber hack cannot be identified by the bad data detection algorithm). Due to the nonlinear AC power flow model and combinatorial selection of compromised sensors, the problem is nonconvex and cannot be solved in polynomial time; however, it is shown that convexification of the original problem based on a semidefinite programming (SDP) relaxation and a sparsity penalty is able to recover a near-optimal solution. This represents the first study to solve the AC-based FDIA. Simulations on a 30-bus system illustrate that the proposed attack requires only sparse sensor manipulation and remains stealthy from the residual-based bad data detection mechanism. In light of the analysis, this study raises new challenges on grid defense mechanism and attack detection strategy.

@article{2017_2J_moddr,
title = "MOD-DR: Microgrid optimal dispatch with demand response",
journal = "Applied Energy",
volume = "187",
pages = "758 - 776",
year = "2017",
issn = "0306-2619",
doi = "https://doi.org/10.1016/j.apenergy.2016.11.093",
url_link = "http://www.sciencedirect.com/science/article/pii/S030626191631724X",
author = "Ming Jin and Wei Feng and Ping Liu and Chris Marnay and Costas Spanos",
keywords = "Power system",
abstract = "In the face of unprecedented challenges of upcoming fossil fuel shortage and reliability and security of the grid, there is an increasing interest in adopting distributed, renewable, energy resources, such as microgrids (MGs), and engaging flexible electric loads in power system operations to potentially drive a paradigm shift in energy production and consumption patterns. Prior work on MG dispatch has leveraged decentralized technologies like combined heat and power (CHP) and heat pumps to promote efficiency and economic gains; however, the flexibility of demand has yet to be fully exploited in cooperation with the grid to offer added benefits and ancillary services. The object of the study is to develop microgrid optimal dispatch with demand response (MOD-DR), which fills in the gap by coordinating both the demand and supply sides in a renewable-integrated, storage-augmented, DR-enabled MG to achieve economically viable and system-wide resilient solutions. The key contribution of this paper is the formulation of a multi-objective optimization with prevailing constraints and utility trade-off based on the model of a large-scale MG with flexible loads, which leads to the derivation of strategies that incorporate uncertainty in scheduling. Evaluation using real datasets is conducted to analyze the uncertainty effects and demand response potentials, demonstrating in a campus prototype a 17.5% peak load reduction and 8.8% cost savings for MOD-DR compared to the non-trivial baseline, which is on par with the Oracle for perfect predictions.",
url_pdf = {MOD-DR.pdf}}

In the face of unprecedented challenges of upcoming fossil fuel shortage and reliability and security of the grid, there is an increasing interest in adopting distributed, renewable, energy resources, such as microgrids (MGs), and engaging flexible electric loads in power system operations to potentially drive a paradigm shift in energy production and consumption patterns. Prior work on MG dispatch has leveraged decentralized technologies like combined heat and power (CHP) and heat pumps to promote efficiency and economic gains; however, the flexibility of demand has yet to be fully exploited in cooperation with the grid to offer added benefits and ancillary services. The object of the study is to develop microgrid optimal dispatch with demand response (MOD-DR), which fills in the gap by coordinating both the demand and supply sides in a renewable-integrated, storage-augmented, DR-enabled MG to achieve economically viable and system-wide resilient solutions. The key contribution of this paper is the formulation of a multi-objective optimization with prevailing constraints and utility trade-off based on the model of a large-scale MG with flexible loads, which leads to the derivation of strategies that incorporate uncertainty in scheduling. Evaluation using real datasets is conducted to analyze the uncertainty effects and demand response potentials, demonstrating in a campus prototype a 17.5% peak load reduction and 8.8% cost savings for MOD-DR compared to the non-trivial baseline, which is on par with the Oracle for perfect predictions.

@inproceedings{2017_2C_indoor,
  title={Indoor environmental quality monitoring by autonomous mobile sensing},
  author={Jin, Ming and Liu, Shichao and Tian, Yulun and Lu, Mingjian and Schiavon, Stefano and Spanos, Costas},
  booktitle={ACM International Conference on Systems for Energy-Efficient Built Environments (BuildSys)},
  pages={1--4},
  year={2017},
  url_pdf={ieq_buildsys.pdf},
  abstract={Indoor environmental quality (IEQ) monitoring is a critical task in building operation, maintenance, and diagnosis. Current approach based on static sensor network is not scalable for IEQ assessment that relies on costly sensing instruments. The study proposes to leverage autonomous mobility to reduce sensing infrastructure cost and enable real-time high-granularity monitoring that can be otherwise inhibitively laborious. Unique to the autonomous mobile sensing methodology, the collected IEQ samples are highly sparse in both spatial and temporal domains. The study develops spatiotemporal (ST) interpolation methods based on ST binning, global trend extraction, and local variation estimation, which efficiently use the data to construct accurate depiction of the indoor environment evolution. The method is evaluated by a standard protocol for ventilation assessment, where the estimation is shown to be highly correlated with the ground truth, and reveals the true ventilation conditions.},
  keywords={Data mining, Smart city, Energy system}
}

Indoor environmental quality (IEQ) monitoring is a critical task in building operation, maintenance, and diagnosis. Current approach based on static sensor network is not scalable for IEQ assessment that relies on costly sensing instruments. The study proposes to leverage autonomous mobility to reduce sensing infrastructure cost and enable real-time high-granularity monitoring that can be otherwise inhibitively laborious. Unique to the autonomous mobile sensing methodology, the collected IEQ samples are highly sparse in both spatial and temporal domains. The study develops spatiotemporal (ST) interpolation methods based on ST binning, global trend extraction, and local variation estimation, which efficiently use the data to construct accurate depiction of the indoor environment evolution. The method is evaluated by a standard protocol for ventilation assessment, where the estimation is shown to be highly correlated with the ground truth, and reveals the true ventilation conditions.

@ARTICLE{2017_1J_winips,
  author={H. {Zou} and M. {Jin} and H. {Jiang} and L. {Xie} and C. J. {Spanos}},
  journal={IEEE Transactions on Wireless Communications}, 
  title={WinIPS: WiFi-Based Non-Intrusive Indoor Positioning System With Online Radio Map Construction and Adaptation}, 
  year={2017},
  volume={16},
  number={12},
  pages={8118-8130},
  doi={10.1109/TWC.2017.2757472},
  abstract={WiFi fingerprinting-based indoor positioning system (IPS) has become the most promising solution for indoor localization. However, there are two major drawbacks that hamper its large-scale implementation. First, an offline site survey process is required which is extremely time-consuming and labor-intensive. Second, the RSS fingerprint database built offline is vulnerable to environmental dynamics. To address these issues comprehensively, in this paper, we propose WinIPS, a WiFi-based non-intrusive IPS that enables automatic online radio map construction and adaptation, aiming for calibration-free indoor localization. WinIPS can capture data packets transmitted in existing WiFi traffic and extract the RSS and MAC addresses of both WiFi access points (APs) and mobile devices in a non-intrusive manner. APs can be used as online reference points for radio map construction. A novel Gaussian process regression model is proposed to approximate the non-uniform RSS distribution of an indoor environment. Extensive experiments were conducted, which demonstrated that WinIPS outperforms existing solutions in terms of both RSS estimation accuracy and localization accuracy.},
  url_pdf = {WinIPS.pdf},
  url_link = {https://ieeexplore.ieee.org/document/8057286/versions},
  keywords = "Smart city, Data mining, Machine learning"}

WiFi fingerprinting-based indoor positioning system (IPS) has become the most promising solution for indoor localization. However, there are two major drawbacks that hamper its large-scale implementation. First, an offline site survey process is required which is extremely time-consuming and labor-intensive. Second, the RSS fingerprint database built offline is vulnerable to environmental dynamics. To address these issues comprehensively, in this paper, we propose WinIPS, a WiFi-based non-intrusive IPS that enables automatic online radio map construction and adaptation, aiming for calibration-free indoor localization. WinIPS can capture data packets transmitted in existing WiFi traffic and extract the RSS and MAC addresses of both WiFi access points (APs) and mobile devices in a non-intrusive manner. APs can be used as online reference points for radio map construction. A novel Gaussian process regression model is proposed to approximate the non-uniform RSS distribution of an indoor environment. Extensive experiments were conducted, which demonstrated that WinIPS outperforms existing solutions in terms of both RSS estimation accuracy and localization accuracy.

@INPROCEEDINGS{2017_1C_leveraging,
  author={I. C. {Konstantakopoulos} and L. J. {Ratliff} and M. {Jin} and C. J. {Spanos}},
  booktitle={American Control Conference (ACC)}, 
  title={Leveraging correlations in utility learning}, 
  year={2017},
  volume={},
  number={},
  pages={5249-5256},
  doi={10.23919/ACC.2017.7963770},
  url_link={https://ieeexplore.ieee.org/abstract/document/7963770},
  url_pdf={leverage_correlation.pdf},
  abstract={We present two approaches for leveraging correlations in learning the utilities of non-cooperative agents' competing in a game: correlation and coalition utility learning. In the former, we estimate the correlations between agents using constrained Feasible Generalized Least Squares with noise estimation and then use the estimated correlations to generate a correlation utility function for each agent which is a weighted sum of its own estimated utility function and all the agents' estimated utilities that are highly correlated with them. We then optimize the weights to boost the performance of the estimators. In the latter, we use a small amount of training data to estimate the correlations between players and form coalitions between agents that are positively correlated. We then estimate the parameters of the utility functions for each coalition where agents in a coalition jointly optimize their utilities. The correlation utility learning method outperforms existing schemes while the coalition utility learning method is simple enough to be adapted to an online framework after an initial training phase, yet it matches the performance of much more complex schemes. To demonstrate the efficacy of the estimation schemes, we apply them to data collected from a social game framework for incentivizing more efficient shared resource consumption in smart buildings.},
  keywords={Game theory, Smart city, Optimization, Data mining}}
  

We present two approaches for leveraging correlations in learning the utilities of non-cooperative agents' competing in a game: correlation and coalition utility learning. In the former, we estimate the correlations between agents using constrained Feasible Generalized Least Squares with noise estimation and then use the estimated correlations to generate a correlation utility function for each agent which is a weighted sum of its own estimated utility function and all the agents' estimated utilities that are highly correlated with them. We then optimize the weights to boost the performance of the estimators. In the latter, we use a small amount of training data to estimate the correlations between players and form coalitions between agents that are positively correlated. We then estimate the parameters of the utility functions for each coalition where agents in a coalition jointly optimize their utilities. The correlation utility learning method outperforms existing schemes while the coalition utility learning method is simple enough to be adapted to an online framework after an initial training phase, yet it matches the performance of much more complex schemes. To demonstrate the efficacy of the estimation schemes, we apply them to data collected from a social game framework for incentivizing more efficient shared resource consumption in smart buildings.

@article{2017_0J_transportation,
title = "Measuring fine-grained metro interchange time via smartphones",
journal = "Transportation Research Part C: Emerging Technologies",
volume = "81",
pages = "153 - 171",
year = "2017",
issn = "0968-090X",
doi = "https://doi.org/10.1016/j.trc.2017.05.014",
url_link = "http://www.sciencedirect.com/science/article/pii/S0968090X17301523",
url_pdf = {/metro_smartphone.pdf},
author = "Weixi Gu and Kai Zhang and Zimu Zhou and Ming Jin and Yuxun Zhou and Xi Liu and Costas J. Spanos and Zuo-Jun (Max) Shen and Wei-Hua Lin and Lin Zhang",
keywords = "Data mining, Smart city",
abstract = "High variability interchange times often significantly affect the reliability of metro travels. Fine-grained measurements of interchange times during metro transfers can provide valuable insights on the crowdedness of stations, usage of station facilities and efficiency of metro lines. Measuring interchange times in metro systems is challenging since agent-operated systems like automatic fare collection systems only provide coarse-grained trip information and popular localization services like GPS are often inaccessible underground. In this paper, we propose a smartphone-based interchange time measuring method from the passengers’ perspective. It leverages low-power sensors embedded in modern smartphones to record ambient contextual features, and utilizes a two-tier classifier to infer interchange states during a metro trip, and further distinguishes 10 fine-grained cases during interchanges. Experimental results within 6months across over 14 subway lines in 3 major cities demonstrate that our approach yields an overall interchange state inference F1-measurement of 91.0% and an average time error of less than 2min at an inference interval of 20s, and an average accuracy of 89.3% to distinguish the 10 fine-grained interchange cases. We also conducted a series of case studies using measurements collected from crowdsourced users during 3months, which reveals findings previously unattainable without fine-grained interchange time measurements, such as portions of waiting time during interchange, interchange directions, usage of facilities (stairs/escalators/lifts), and the root causes of long interchange times."
}

High variability interchange times often significantly affect the reliability of metro travels. Fine-grained measurements of interchange times during metro transfers can provide valuable insights on the crowdedness of stations, usage of station facilities and efficiency of metro lines. Measuring interchange times in metro systems is challenging since agent-operated systems like automatic fare collection systems only provide coarse-grained trip information and popular localization services like GPS are often inaccessible underground. In this paper, we propose a smartphone-based interchange time measuring method from the passengers’ perspective. It leverages low-power sensors embedded in modern smartphones to record ambient contextual features, and utilizes a two-tier classifier to infer interchange states during a metro trip, and further distinguishes 10 fine-grained cases during interchanges. Experimental results within 6months across over 14 subway lines in 3 major cities demonstrate that our approach yields an overall interchange state inference F1-measurement of 91.0% and an average time error of less than 2min at an inference interval of 20s, and an average accuracy of 89.3% to distinguish the 10 fine-grained interchange cases. We also conducted a series of case studies using measurements collected from crowdsourced users during 3months, which reveals findings previously unattainable without fine-grained interchange time measurements, such as portions of waiting time during interchange, interchange directions, usage of facilities (stairs/escalators/lifts), and the root causes of long interchange times.

@article{2017_0J_longitudinal,
title = "Longitudinal assessment of thermal and perceived air quality acceptability in relation to temperature, humidity, and CO2 exposure in Singapore",
journal = "Building and Environment",
volume = "115",
pages = "80 - 90",
year = "2017",
issn = "0360-1323",
doi = "https://doi.org/10.1016/j.buildenv.2017.01.014",
url_link = "http://www.sciencedirect.com/science/article/pii/S036013231730015X",
author = "Toby C.T. Cheung and Stefano Schiavon and Elliott T. Gall and Ming Jin and William W Nazaroff",
keywords = "Smart city, Data mining",
abstract = "Thermal acceptability (TA) and perceived air quality acceptability (PAQA) are typically analyzed in climate chambers or cross-sectional field studies. Individual factors, such as expectations and perceived environment history, may influence the acceptability response. Longitudinal studies with multi-day design are absent in the literature. Fifteen Singaporean subjects participated in a 7-day longitudinal experiment in which they carried a portable sensor that continuously recorded personal air temperature, relative humidity and carbon dioxide concentration at 1-min intervals. Instantaneous TA and PAQA were regularly sampled by survey for each subject. High acceptability was found at home, restaurants and workplaces, whereas low acceptability was found for outdoor and transport environments. The participants, from Singapore's modern tropical environment spent an average of 96% of their time indoors. Weak associations were reported between acceptabilities and measured physical parameters taken independently. Clustering data by location, subject's sleeping ventilation habit, air-conditioning operation status and the changes in physical parameters over a designated time period enhanced the understanding of the acceptability results. In general, acceptability was lower for those who slept in air-conditioned environments than for those who slept without air-conditioning. The carbon dioxide mixing ratio was critical for PAQA predictions but not for TA. The Gaussian process (GP) had a better predictive power than a multiple linear regression approach. Using GP, we found that a general predictive model had comparable simulation performance as for individual predictive models. The longitudinal experiment has demonstrated effectiveness for TA and PAQA analysis, which could be beneficial to future studies in personal comfort prediction."
}

Thermal acceptability (TA) and perceived air quality acceptability (PAQA) are typically analyzed in climate chambers or cross-sectional field studies. Individual factors, such as expectations and perceived environment history, may influence the acceptability response. Longitudinal studies with multi-day design are absent in the literature. Fifteen Singaporean subjects participated in a 7-day longitudinal experiment in which they carried a portable sensor that continuously recorded personal air temperature, relative humidity and carbon dioxide concentration at 1-min intervals. Instantaneous TA and PAQA were regularly sampled by survey for each subject. High acceptability was found at home, restaurants and workplaces, whereas low acceptability was found for outdoor and transport environments. The participants, from Singapore's modern tropical environment spent an average of 96% of their time indoors. Weak associations were reported between acceptabilities and measured physical parameters taken independently. Clustering data by location, subject's sleeping ventilation habit, air-conditioning operation status and the changes in physical parameters over a designated time period enhanced the understanding of the acceptability results. In general, acceptability was lower for those who slept in air-conditioned environments than for those who slept without air-conditioning. The carbon dioxide mixing ratio was critical for PAQA predictions but not for TA. The Gaussian process (GP) had a better predictive power than a multiple linear regression approach. Using GP, we found that a general predictive model had comparable simulation performance as for individual predictive models. The longitudinal experiment has demonstrated effectiveness for TA and PAQA analysis, which could be beneficial to future studies in personal comfort prediction.

@inproceedings{2016_2C_metroeye,
  title={Metroeye: smart tracking your metro trips underground},
  author={Gu, Weixi and Jin, Ming and Zhou, Zimu and Spanos, Costas J and Zhang, Lin},
  booktitle={International Conference on Mobile and Ubiquitous Systems: Computing, Networking and Services (MobiQuitous)},
  url_pdf={metroeye_paper.pdf},
  url_link={https://dl.acm.org/doi/10.1145/2968219.2971437},
  url_poster={metroeye.pdf},
  keywords={Smart city, Data mining},
  abstract={Subway has become the first choice of traveling for people in metropolis due to its efficiency and convenience. Yet passengers have to rely on subway broadcasts to know their locations because popular localization services (e.g. GPS and wireless localization technologies) are often unavailable underground. To this end, we propose MetroEye, a fine-grained passenger tracking service underground. MetroEye leverages smartphone sensors to record ambient contextual features, and infers the state of passengers (including stop, running, and interchange) during a metro trip using a Conditional Random Field (CRF) model. MetroEye further provides arrival alarm services based on individual passenger state, and aggregates crowdsourced interchange durations to guide passengers for intelligent metro trip planning. Experimental results within 6 months across over 14 subway trains in 3 major cities demonstrate that MetroEye outperforms the state-of-the-art.},
  pages={84--93},
  year={2016},
  note={<font style="color:#FF0000">(Best Paper Runner-up)</font>}
}

Subway has become the first choice of traveling for people in metropolis due to its efficiency and convenience. Yet passengers have to rely on subway broadcasts to know their locations because popular localization services (e.g. GPS and wireless localization technologies) are often unavailable underground. To this end, we propose MetroEye, a fine-grained passenger tracking service underground. MetroEye leverages smartphone sensors to record ambient contextual features, and infers the state of passengers (including stop, running, and interchange) during a metro trip using a Conditional Random Field (CRF) model. MetroEye further provides arrival alarm services based on individual passenger state, and aggregates crowdsourced interchange durations to guide passengers for intelligent metro trip planning. Experimental results within 6 months across over 14 subway trains in 3 major cities demonstrate that MetroEye outperforms the state-of-the-art.

@article{2016_1J_mapsentinel,
  title={Mapsentinel: Can the knowledge of space use improve indoor tracking further?},
  author={Jia, Ruoxi and Jin, Ming and Zou, Han and Yesilata, Yigitcan and Xie, Lihua and Spanos, Costas},
  journal={Sensors},
  volume={16},
  number={4},
  pages={472},
  year={2016},
  publisher={Multidisciplinary Digital Publishing Institute},
  abstract={Estimating an occupant’s location is arguably the most fundamental sensing task in smart buildings. The applications for fine-grained, responsive building operations require the location sensing systems to provide location estimates in real time, also known as indoor tracking. Existing indoor tracking systems require occupants to carry specialized devices or install programs on their smartphone to collect inertial sensing data. In this paper, we propose MapSentinel, which performs non-intrusive location sensing based on WiFi access points and ultrasonic sensors. MapSentinel combines the noisy sensor readings with the floormap information to estimate locations. One key observation supporting our work is that occupants exhibit distinctive motion characteristics at different locations on the floormap, e.g., constrained motion along the corridor or in the cubicle zones, and free movement in the open space. While extensive research has been performed on using a floormap as a tool to obtain correct walking trajectories without wall-crossings, there have been few attempts to incorporate the knowledge of space use available from the floormap into the location estimation. This paper argues that the knowledge of space use as an additional information source presents new opportunities for indoor tracking. The fusion of heterogeneous information is theoretically formulated within the Factor Graph framework, and the Context-Augmented Particle Filtering algorithm is developed to efficiently solve real-time walking trajectories. Our evaluation in a large office space shows that the MapSentinel can achieve accuracy improvement of 31.3% compared with the purely WiFi-based tracking system. },
  url_link = {https://www.mdpi.com/1424-8220/16/4/472},
  url_pdf = {mapsentinel.pdf},
  keywords = {Smart city, Data mining}
}

Estimating an occupant’s location is arguably the most fundamental sensing task in smart buildings. The applications for fine-grained, responsive building operations require the location sensing systems to provide location estimates in real time, also known as indoor tracking. Existing indoor tracking systems require occupants to carry specialized devices or install programs on their smartphone to collect inertial sensing data. In this paper, we propose MapSentinel, which performs non-intrusive location sensing based on WiFi access points and ultrasonic sensors. MapSentinel combines the noisy sensor readings with the floormap information to estimate locations. One key observation supporting our work is that occupants exhibit distinctive motion characteristics at different locations on the floormap, e.g., constrained motion along the corridor or in the cubicle zones, and free movement in the open space. While extensive research has been performed on using a floormap as a tool to obtain correct walking trajectories without wall-crossings, there have been few attempts to incorporate the knowledge of space use available from the floormap into the location estimation. This paper argues that the knowledge of space use as an additional information source presents new opportunities for indoor tracking. The fusion of heterogeneous information is theoretically formulated within the Factor Graph framework, and the Context-Augmented Particle Filtering algorithm is developed to efficiently solve real-time walking trajectories. Our evaluation in a large office space shows that the MapSentinel can achieve accuracy improvement of 31.3% compared with the purely WiFi-based tracking system.

@INPROCEEDINGS{2016_1C_winips,
  author={H. {Zou} and  {Ming Jin} and H. {Jiang} and L. {Xie} and C. {Spanos}},
  booktitle={IEEE Conference on Computer Communications (INFOCOM) Workshops }, 
  title={WinIPS: WiFi-based non-intrusive IPS for online radio map construction}, 
  year={2016},
  volume={},
  number={},
  pages={1081-1082},
  doi={10.1109/INFCOMW.2016.7562263},
  url_link={https://ieeexplore.ieee.org/abstract/document/7562263},
  url_pdf={infocom_winips.pdf},
  abstract={Existing WiFi fingerprinting-based Indoor Positioning System (IPS) suffers from two major bottlenecks. One is that the offline site survey process is extremely time-consuming and labor-intensive. The other is that the offline calibrated received signal strength (RSS) fingerprint database is vulnerable to environmental dynamics. To address these issues comprehensively, in this paper, we propose WinIPS, a WiFi-based non-intrusive IPS that enables automatic online radio map construction and adaptation for calibration-free indoor localization. WinIPS is able to capture data packets transmitted in the existing WiFi traffic and extract the RSS values and MAC addresses of both access points (AP) and mobile devices (MD) in a non-intrusive manner. By leveraging APs as online reference points for radio map construction, we can completely remove the needs of laborious offline site survey process. The constructed radio map is more robust to environmental dynamics since it is updated automatically in real-time. Extensive experimental results verify the superiority of WinIPS in terms of RSS estimation accuracy and localization accuracy, and these merits make it more suitable for practical large-scale implementation.},
  keywords={Data mining, Smart city}}
  

Existing WiFi fingerprinting-based Indoor Positioning System (IPS) suffers from two major bottlenecks. One is that the offline site survey process is extremely time-consuming and labor-intensive. The other is that the offline calibrated received signal strength (RSS) fingerprint database is vulnerable to environmental dynamics. To address these issues comprehensively, in this paper, we propose WinIPS, a WiFi-based non-intrusive IPS that enables automatic online radio map construction and adaptation for calibration-free indoor localization. WinIPS is able to capture data packets transmitted in the existing WiFi traffic and extract the RSS values and MAC addresses of both access points (AP) and mobile devices (MD) in a non-intrusive manner. By leveraging APs as online reference points for radio map construction, we can completely remove the needs of laborious offline site survey process. The constructed radio map is more robust to environmental dynamics since it is updated automatically in real-time. Extensive experimental results verify the superiority of WinIPS in terms of RSS estimation accuracy and localization accuracy, and these merits make it more suitable for practical large-scale implementation.

@INPROCEEDINGS{2016_1C_smart,
  author={I. C. {Konstantakopoulos} and L. J. {Ratliff} and M. {Jin} and C. {Spanos} and S. S. {Sastry}},
  booktitle={International Workshop on Science of Smart City Operations and Platforms Engineering (SCOPE)}, 
  title={Smart building energy efficiency via social game: a robust utility learning framework for closing–the–loop}, 
  year={2016},
  volume={},
  number={},
  pages={1-6},
  doi={10.1109/SCOPE.2016.7515054},
  url_link={https://ieeexplore.ieee.org/abstract/document/7515054},
  url_pdf={Scope2016CameraReady.pdf},
  abstract={Given a non-cooperative, continuous game, we describe a framework for parametric utility learning. Using heteroskedasticity inference, we adapt a Constrained Feasible Generalized Least Squares (cFGLS) utility learning method in which estimator variance is reduced, unbiased, and consistent. We extend our utility learning method using bootstrapping and bagging. We show the performance of the proposed method using data from a social game experiment designed to encourage energy efficient behavior amongst building occupants. Using occupant voting data we simulate the game defined by the estimated utility functions and show that the performance of our robust utility learning method and quantify its improvement over classical methods such as Ordinary Least Squares (OLS).},
  keywords={Data mining, Optimization, Game theory, Smart city, Energy system}}

Given a non-cooperative, continuous game, we describe a framework for parametric utility learning. Using heteroskedasticity inference, we adapt a Constrained Feasible Generalized Least Squares (cFGLS) utility learning method in which estimator variance is reduced, unbiased, and consistent. We extend our utility learning method using bootstrapping and bagging. We show the performance of the proposed method using data from a social game experiment designed to encourage energy efficient behavior amongst building occupants. Using occupant voting data we simulate the game defined by the estimated utility functions and show that the performance of our robust utility learning method and quantify its improvement over classical methods such as Ordinary Least Squares (OLS).

@INPROCEEDINGS{2016_1C_inverse,
  author={I. C. {Konstantakopoulos} and L. J. {Ratliff} and M. {Jin} and C. J. {Spanos} and S. S. {Sastry}},
  booktitle={IEEE Conference on Decision and Control}, 
  title={Inverse modeling of non-cooperative agents via mixture of utilities}, 
  year={2016},
  volume={},
  number={},
  pages={6327-6334},
  doi={10.1109/CDC.2016.7799243},
  url_link={https://ieeexplore.ieee.org/document/7799243},
  url_pdf={InverseModel_2016.pdf},
  abstract={We describe a new method of parametric utility learning for non-cooperative, continuous games using a probabilistic interpretation for combining multiple utility functions - thereby creating a mixture of utilities - under non-spherical noise terms. We present an adaptation of mixture of regression models that takes in to account heteroskedasticity. We show the performance of the proposed method by estimating the utility functions of players using data from a social game experiment designed to encourage energy efficient behavior amongst building occupants. Using occupant voting data we simulate the new game defined by the estimated mixture of utilities and show that the resulting forecast is more accurate than robust utility learning methods such as constrained Feasible Generalized Least Squares (cFGLS), ensemble methods such as bagging, and classical methods such as Ordinary Least Squares (OLS).},
  keywords={Smart city, Optimization, Data mining}}
  

We describe a new method of parametric utility learning for non-cooperative, continuous games using a probabilistic interpretation for combining multiple utility functions - thereby creating a mixture of utilities - under non-spherical noise terms. We present an adaptation of mixture of regression models that takes in to account heteroskedasticity. We show the performance of the proposed method by estimating the utility functions of players using data from a social game experiment designed to encourage energy efficient behavior amongst building occupants. Using occupant voting data we simulate the new game defined by the estimated mixture of utilities and show that the resulting forecast is more accurate than robust utility learning methods such as constrained Feasible Generalized Least Squares (cFGLS), ensemble methods such as bagging, and classical methods such as Ordinary Least Squares (OLS).

@INPROCEEDINGS{2015_3C_sbp,
  title={Sensing by proxy: Occupancy detection based on indoor CO2 concentration},
  author={Jin, Ming and Bekiaris-Liberis, Nikolaos and Weekly, Kevin and Spanos, Costas and Bayen, Alexandre},
  booktitle={UBICOMM 2015},
  url_pdf={sbp.pdf},
  url_link={https://www.iaria.org/conferences2015/UBICOMM15.html},
  abstract={Sensing by proxy, as described in this study, is a sensing paradigm which infers latent factors by “proxy” measurements based on constitutive models that exploit the spatial and physical features in the system. In this study, we demonstrate the efficiency of sensing by proxy for occupancy detection based on indoor CO2 concentration. We propose a link model that relates the proxy measurements with unknown human emission rates based on a data-driven model which consists of a coupled Partial Differential Equation (PDE) – Ordinary Differential Equation (ODE) system. We report on several experimental results using both a CO2 pump that emulates human breathing, as well as measurements of actual occupancy by performing controlled field experiments, in order to validate our model. Parameters of the model are datadriven, which exhibit long-term stability and robustness across all the occupants experiments. The inference of the number of occupants in the room based on CO2 measurements at the air return and air supply vents by sensing by proxy outperforms a range of machine learning algorithms, and achieves an overall mean squared error of 0.6569 (fractional person), while the best alternative by Bayes net is 1.2061 (fractional person). Building indoor occupancy is essential to facilitate heating, ventilation, and air conditioning (HVAC) control, lighting adjustment, and occupancy-aware services to achieve occupancy comfort and energy efficiency. The significance of this study is the proposal of a paradigm of sensing that results in a parsimonious and accurate occupancy inference model, which holds considerable potential for energy saving and improvement of HVAC operations. The proposed framework can be also applied to other tasks, such as indoor pollutants source identification, while requiring minimal infrastructure expenses.},
  keywords={Smart city, Energy system, Data mining},
  volume={14},
  year={2015},
  note={<font style="color:#FF0000">(Best Paper Award)</font>}
}

Sensing by proxy, as described in this study, is a sensing paradigm which infers latent factors by “proxy” measurements based on constitutive models that exploit the spatial and physical features in the system. In this study, we demonstrate the efficiency of sensing by proxy for occupancy detection based on indoor CO2 concentration. We propose a link model that relates the proxy measurements with unknown human emission rates based on a data-driven model which consists of a coupled Partial Differential Equation (PDE) – Ordinary Differential Equation (ODE) system. We report on several experimental results using both a CO2 pump that emulates human breathing, as well as measurements of actual occupancy by performing controlled field experiments, in order to validate our model. Parameters of the model are datadriven, which exhibit long-term stability and robustness across all the occupants experiments. The inference of the number of occupants in the room based on CO2 measurements at the air return and air supply vents by sensing by proxy outperforms a range of machine learning algorithms, and achieves an overall mean squared error of 0.6569 (fractional person), while the best alternative by Bayes net is 1.2061 (fractional person). Building indoor occupancy is essential to facilitate heating, ventilation, and air conditioning (HVAC) control, lighting adjustment, and occupancy-aware services to achieve occupancy comfort and energy efficiency. The significance of this study is the proposal of a paradigm of sensing that results in a parsimonious and accurate occupancy inference model, which holds considerable potential for energy saving and improvement of HVAC operations. The proposed framework can be also applied to other tasks, such as indoor pollutants source identification, while requiring minimal infrastructure expenses.

@INPROCEEDINGS{2015_3C_brief,
  author={M. {Jin} and C. J. {Spanos}},
  booktitle={IEEE Conference on Decision and Control (CDC)}, 
  title={BRIEF: Bayesian Regression of Infinite Expert Forecasters for single and multiple time series prediction}, 
  year={2015},
  volume={},
  number={},
  pages={78-83},
  doi={10.1109/CDC.2015.7402089},
  url_link={https://ieeexplore.ieee.org/document/7402089},
  url_pdf={brief_cdc.pdf},
  abstract={Bayesian Regression of Infinite Expert Forecasters (BRIEF) as proposed in the study is a prediction algorithm for time-varying systems. The method is based on regret minimization by tracking the performance of an inifinite pool of experts for single and multiple time series. The inverse correlation weighted error (ICWE) employed in BRIEF takes into account the dependency structure among multiple time series, which can also be adapted to multi-step ahead predictions. Theoretical bounds show that the cumulative regret grows at rate O(log T) with respect to the oracle that can select the best strategy in retrospect. As the per round regret vanishes, BRIEF is indistinguishable to the oracle when the horizon increases. Also since the bound applies to any choice of input subject to the euclidean norm constraint, the method can be applied to adversarial settings. Experimental results verify that BRIEF excels in single and multiple steps ahead prediction of ARMAX simulated data and building energy consumptions.},
  keywords={Machine learning, Optimization}}
  

Bayesian Regression of Infinite Expert Forecasters (BRIEF) as proposed in the study is a prediction algorithm for time-varying systems. The method is based on regret minimization by tracking the performance of an inifinite pool of experts for single and multiple time series. The inverse correlation weighted error (ICWE) employed in BRIEF takes into account the dependency structure among multiple time series, which can also be adapted to multi-step ahead predictions. Theoretical bounds show that the cumulative regret grows at rate O(log T) with respect to the oracle that can select the best strategy in retrospect. As the per round regret vanishes, BRIEF is indistinguishable to the oracle when the horizon increases. Also since the bound applies to any choice of input subject to the euclidean norm constraint, the method can be applied to adversarial settings. Experimental results verify that BRIEF excels in single and multiple steps ahead prediction of ARMAX simulated data and building energy consumptions.

@INPROCEEDINGS{2015_2C_soundloc,
  author={R. {Jia} and M. {Jin} and Z. {Chen} and C. J. {Spanos}},
  booktitle={IEEE International Conference on Automation Science and Engineering (CASE)}, 
  title={SoundLoc: Accurate room-level indoor localization using acoustic signatures}, 
  year={2015},
  volume={},
  number={},
  pages={186-193},
  doi={10.1109/CoASE.2015.7294060},
  url_link={https://ieeexplore.ieee.org/document/7294060},
  url_pdf={soundloc.pdf},
  url_media={https://www.technologyreview.com/view/529176/an-indoor-positioning-system-based-on-echolocation/},
  abstract={Room-level indoor localization is of particular interest in the energy-efficient smart building, as services, such as lighting and ventilation, can be targeted towards individual rooms based on occupancy instead of an entire floor. Hence, this paper focuses on identifying the room where a person or a mobile device is physically present. Existing room-level localization methods, however, require special infrastructure to annotate rooms with special signatures. SoundLoc is a room-level localization scheme that exploits the intrinsic acoustic properties of individual rooms and obviates the needs for infrastructures. As we will show in the study, rooms' acoustic properties can be characterized by Room Impulse Response (RIR). Nevertheless, obtaining precise RIRs is a time-consuming and expensive process. The main contributions of our work are the following: First, a cost-effective RIR measurement system is designed and the Noise Adaptive Extraction of Reverberation (NAER) algorithm is developed to estimate room acoustic parameters in noisy conditions. Second, a comprehensive physical and statistical analysis of features extracted from RIRs is performed. Also, SoundLoc is evaluated using the dataset consisting of ten (10) different rooms and the overall accuracy of 97.8% has been achieved.},
  keywords={Data mining, Smart city, Energy system},
  note={<a style="color:#FF0000" href="http://www.technologyreview.com/view/529176/an-indoor-positioning-system-based-on-echolocation/">(Featured in 'MIT Technology Review')</a>}}
  

Room-level indoor localization is of particular interest in the energy-efficient smart building, as services, such as lighting and ventilation, can be targeted towards individual rooms based on occupancy instead of an entire floor. Hence, this paper focuses on identifying the room where a person or a mobile device is physically present. Existing room-level localization methods, however, require special infrastructure to annotate rooms with special signatures. SoundLoc is a room-level localization scheme that exploits the intrinsic acoustic properties of individual rooms and obviates the needs for infrastructures. As we will show in the study, rooms' acoustic properties can be characterized by Room Impulse Response (RIR). Nevertheless, obtaining precise RIRs is a time-consuming and expensive process. The main contributions of our work are the following: First, a cost-effective RIR measurement system is designed and the Noise Adaptive Extraction of Reverberation (NAER) algorithm is developed to estimate room acoustic parameters in noisy conditions. Second, a comprehensive physical and statistical analysis of features extracted from RIRs is performed. Also, SoundLoc is evaluated using the dataset consisting of ten (10) different rooms and the overall accuracy of 97.8% has been achieved.

@inproceedings{2015_2C_rest,
  title={REST: a reliable estimation of stopping time algorithm for social game experiments},
  author={Jin, Ming and Ratliff, Lillian J and Konstantakopoulos, Ioannis and Spanos, Costas and Sastry, Shankar},
  booktitle={ACM/IEEE International Conference on Cyber-Physical Systems (ICCPS)},
  pages={90--99},
  year={2015},
  url_pdf={rest.pdf},
  url_link={https://dl.acm.org/doi/abs/10.1145/2735960.2735974},
  url_code={https://github.com/jinming99/REST},
  abstract={Through a social game, we integrate building occupants into the control and management of an office building that is instrumented with networked embedded systems for sensing and actuation. The goal of the social game is to both incentivize building occupants to be more energy efficient and learn behavioral models for occupants so that the building can be made sustainable through automation. Given a generative model for the occupants behavior in the competitive environment created by the social game, we develop a method for learning the parameters of the behavioral model as we conduct the experiment by adopting a learning to learn framework. Using tools from statistical learning, we provide bounds on the parameter inference error. In addition, we provide an algorithm for computing the stopping time required for a specified level of confidence in estimation. We show the performance of our algorithm in several examples.},
  keywords={Smart city, Optimization, Energy system}
}

Through a social game, we integrate building occupants into the control and management of an office building that is instrumented with networked embedded systems for sensing and actuation. The goal of the social game is to both incentivize building occupants to be more energy efficient and learn behavioral models for occupants so that the building can be made sustainable through automation. Given a generative model for the occupants behavior in the competitive environment created by the social game, we develop a method for learning the parameters of the behavioral model as we conduct the experiment by adopting a learning to learn framework. Using tools from statistical learning, we provide bounds on the parameter inference error. In addition, we provide an algorithm for computing the stopping time required for a specified level of confidence in estimation. We show the performance of our algorithm in several examples.

@INPROCEEDINGS{2015_1C_powerpred,
  author={M. {Jin} and L. {Zhang} and C. J. {Spanos}},
  booktitle={IEEE International Conference on Automation Science and Engineering (CASE)}, 
  title={Power prediction through energy consumption pattern recognition for smart buildings}, 
  year={2015},
  volume={},
  number={},
  pages={419-424},
  doi={10.1109/CoASE.2015.7294115},
  url_link={https://ieeexplore.ieee.org/document/7294115},
  url_pdf={power_pred.pdf},
  abstract={In this paper, we propose a Non-negative Mixture of Experts (NME) model for smart buildings that is capable of making accurate power forecasting by recognizing characteristic consumption patterns. The model uses prediction error as a metric to guide the feature learning process subject to non-negativity constraints. The objective is to understand and model energy consumption behaviors in commercial buildings at the appliance level so as to facilitate dynamic pricing and demand response. Application of the NME model to a large dataset of device power measurements results in the discovery of meaningful energy usage patterns that are characteristic of the working and idle states of the building space, with the additional advantage that the learned features also optimize the energy prediction model. The model can be learned by stochastic gradient descent, which is suitable for large-scale problems, and an online version is also suggested.},
  keywords={Machine learning, Smart city, Energy system}}

In this paper, we propose a Non-negative Mixture of Experts (NME) model for smart buildings that is capable of making accurate power forecasting by recognizing characteristic consumption patterns. The model uses prediction error as a metric to guide the feature learning process subject to non-negativity constraints. The objective is to understand and model energy consumption behaviors in commercial buildings at the appliance level so as to facilitate dynamic pricing and demand response. Application of the NME model to a large dataset of device power measurements results in the discovery of meaningful energy usage patterns that are characteristic of the working and idle states of the building space, with the additional advantage that the learned features also optimize the energy prediction model. The model can be learned by stochastic gradient descent, which is suitable for large-scale problems, and an online version is also suggested.

@inproceedings{2015_1C_apec,
  title={APEC: Auto planner for efficient configuration of indoor positioning system},
  author={Jin, Ming and Jia, Ruoxi and Spanos, Costas},
  booktitle={Int. Conf. Mobile Ubiquitous Comput. Syst. Services Technol.(UBICOMM)},
  pages={100--107},
  year={2015},
  url_pdf={apec.pdf},
  url_link={https://www.iaria.org/conferences2015/AwardsUBICOMM15.html},
  note={<font style="color:#FF0000">(Invitation for IARIA Journals)</font>},
  abstract={Fingerprints-based methods have been prevailing in indoor positioning systems, whereas they have certain drawbacks that fingerprints collection in the offline phase requires considerable manpower and time. Auto Planner for Efficient Configuration (APEC) systematically exploits router setups and fingerprints allocations over space by taking into account user preferences and budget constraints. The task of configuration is formulated as an optimization problem, whose objective is the expected loss based on the Hierarchical Bayesian Signal Model (HBSM) and theoretical results on the misclassification rates. To reduce the computational complexity of large-scale problems, two heuristics are employed, i.e., the coordinate descent and the router-fingerprints decoupling, which are validated by simulation analysis. Experiments with three mobile devices (Android, iPad, iPhone) in two setups (7 or 9 access points) verify that the expected loss is a reliable predictor of the actual loss of the system (objective consistency), and that APEC outperforms the random and uniform approaches (solution superiority). Since APEC focuses on the system configuration in the planning stage, it can be combined with other fingerprinting processes in the online phase to improve the utility of the system},
  keywords={Optimization, Data mining, Smart city, Energy system}
}

Fingerprints-based methods have been prevailing in indoor positioning systems, whereas they have certain drawbacks that fingerprints collection in the offline phase requires considerable manpower and time. Auto Planner for Efficient Configuration (APEC) systematically exploits router setups and fingerprints allocations over space by taking into account user preferences and budget constraints. The task of configuration is formulated as an optimization problem, whose objective is the expected loss based on the Hierarchical Bayesian Signal Model (HBSM) and theoretical results on the misclassification rates. To reduce the computational complexity of large-scale problems, two heuristics are employed, i.e., the coordinate descent and the router-fingerprints decoupling, which are validated by simulation analysis. Experiments with three mobile devices (Android, iPad, iPhone) in two setups (7 or 9 access points) verify that the expected loss is a reliable predictor of the actual loss of the system (objective consistency), and that APEC outperforms the random and uniform approaches (solution superiority). Since APEC focuses on the system configuration in the planning stage, it can be combined with other fingerprinting processes in the online phase to improve the utility of the system

@ARTICLE{2015_0J_modeling,
  author={K. {Weekly} and N. {Bekiaris-Liberis} and M. {Jin} and A. M. {Bayen}},
  journal={IEEE Transactions on Control Systems Technology}, 
  title={Modeling and Estimation of the Humans' Effect on the CO2 Dynamics Inside a Conference Room}, 
  year={2015},
  volume={23},
  number={5},
  pages={1770-1781},
  doi={10.1109/TCST.2014.2384002},
  abstract={We develop a data driven, partial differential equation-ordinary differential equation model that describes the response of the carbon dioxide (CO 2 ) dynamics inside a conference room, due to the presence of humans, or of a user-controlled exogenous source of CO 2 . We conduct three controlled experiments to develop and tune a model whose output matches the measured output concentration of CO 2 inside the room, when known inputs are applied to the model. In the first experiment, a controlled amount of CO 2 gas is released inside the room from a regulated supply, and in the second and third experiments, a known number of humans produce a certain amount of CO 2 inside the room. For the estimation of the exogenous inputs, we design an observer, based on our model, using measurements of CO 2 concentrations at two locations inside the room. We perform several simulation studies for the illustration of our results.},
  url_link={https://ieeexplore.ieee.org/abstract/document/7004837},
  url_pdf={modelco2.pdf},
  keywords = {Control theory, Data mining, Smart city}}
  

We develop a data driven, partial differential equation-ordinary differential equation model that describes the response of the carbon dioxide (CO 2 ) dynamics inside a conference room, due to the presence of humans, or of a user-controlled exogenous source of CO 2 . We conduct three controlled experiments to develop and tune a model whose output matches the measured output concentration of CO 2 inside the room, when known inputs are applied to the model. In the first experiment, a controlled amount of CO 2 gas is released inside the room from a regulated supply, and in the second and third experiments, a known number of humans produce a certain amount of CO 2 inside the room. For the estimation of the exogenous inputs, we design an observer, based on our model, using measurements of CO 2 concentrations at two locations inside the room. We perform several simulation studies for the illustration of our results.

@inproceedings{2014_3C_presence,
  title={Presencesense: Zero-training algorithm for individual presence detection based on power monitoring},
  author={Jin, Ming and Jia, Ruoxi and Kang, Zhaoyi and Konstantakopoulos, Ioannis C and Spanos, Costas J},
  booktitle={ACM Conference on Embedded Systems for Energy-Efficient Buildings},
  pages={1--10},
  year={2014},
  url_link={https://dl.acm.org/doi/abs/10.1145/2674061.2674073},
  url_pdf={/presencesense.pdf},
  url_arXiv={https://arxiv.org/abs/1407.4395},
  abstract={Non-intrusive presence detection of individuals in commercial buildings is much easier to implement than intrusive methods such as passive infrared, acoustic sensors, and camera. Individual power consumption, while providing useful feedback and motivation for energy saving, can be used as a valuable source for presence detection. We conduct pilot experiments in an office setting to collect individual presence data by ultrasonic sensors, acceleration sensors, and WiFi access points, in addition to the individual power monitoring data. PresenceSense (PS), a semi-supervised learning algorithm based on power measurement that trains itself with only unlabeled data, is proposed, analyzed and evaluated in the study. Without any labeling efforts, which are usually tedious and time consuming, PresenceSense outperforms popular models whose parameters are optimized over a large training set. The results are interpreted and potential applications of PresenceSense on other data sources are discussed. The significance of this study attaches to space security, occupancy behavior modeling, and energy saving of plug loads.},
  keywords={Machine learning, Data mining, Energy system, Smart city}
}

Non-intrusive presence detection of individuals in commercial buildings is much easier to implement than intrusive methods such as passive infrared, acoustic sensors, and camera. Individual power consumption, while providing useful feedback and motivation for energy saving, can be used as a valuable source for presence detection. We conduct pilot experiments in an office setting to collect individual presence data by ultrasonic sensors, acceleration sensors, and WiFi access points, in addition to the individual power monitoring data. PresenceSense (PS), a semi-supervised learning algorithm based on power measurement that trains itself with only unlabeled data, is proposed, analyzed and evaluated in the study. Without any labeling efforts, which are usually tedious and time consuming, PresenceSense outperforms popular models whose parameters are optimized over a large training set. The results are interpreted and potential applications of PresenceSense on other data sources are discussed. The significance of this study attaches to space security, occupancy behavior modeling, and energy saving of plug loads.

@INPROCEEDINGS{2014_2C_social,
  author={L. J. {Ratliff} and M. {Jin} and I. C. {Konstantakopoulos} and C. {Spanos} and S. S. {Sastry}},
  booktitle={Annual Allerton Conference on Communication, Control, and Computing (Allerton)}, 
  title={Social game for building energy efficiency: Incentive design}, 
  year={2014},
  volume={},
  number={},
  pages={1011-1018},
  doi={10.1109/ALLERTON.2014.7028565},
  url_link={https://ieeexplore.ieee.org/document/7028565},
  url_pdf={social_game.pdf},
  abstract={We present analysis and results of a social game encouraging energy efficient behavior in occupants by distributing points which determine the likelihood of winning in a lottery. We estimate occupants utilities and formulate the interaction between the building manager and the occupants as a reversed Stackelberg game in which there are multiple followers that play in a non-cooperative game. The estimated utilities are used for determining the occupant behavior in the non-cooperative game. Due to nonconvexities and complexity of the problem, in particular the size of the joint distribution across the states of the occupants, we solve the resulting the bilevel optimization problem using a particle swarm optimization method. Drawing from the distribution across player states, we compute the Nash equilibrium of the game using the resulting leader choice. We show that the behavior of the agents under the leader choice results in greater utility for the leader.},
  keywords={Game theory, Optimization, Smart city, Energy system}}

We present analysis and results of a social game encouraging energy efficient behavior in occupants by distributing points which determine the likelihood of winning in a lottery. We estimate occupants utilities and formulate the interaction between the building manager and the occupants as a reversed Stackelberg game in which there are multiple followers that play in a non-cooperative game. The estimated utilities are used for determining the occupant behavior in the non-cooperative game. Due to nonconvexities and complexity of the problem, in particular the size of the joint distribution across the states of the occupants, we solve the resulting the bilevel optimization problem using a particle swarm optimization method. Drawing from the distribution across player states, we compute the Nash equilibrium of the game using the resulting leader choice. We show that the behavior of the agents under the leader choice results in greater utility for the leader.

@INPROCEEDINGS{2014_2C_environment,
  author={M. {Jin} and H. {Zou} and K. {Weekly} and R. {Jia} and A. M. {Bayen} and C. J. {Spanos}},
  booktitle={Annual Conference of the IEEE Industrial Electronics Society (IECON)}, 
  title={Environmental sensing by wearable device for indoor activity and location estimation}, 
  year={2014},
  volume={},
  number={},
  pages={5369-5375},
  doi={10.1109/IECON.2014.7049320},
  url_pdf={environmental_sensing.pdf},
  url_poster={environmental_sensing_poster.pdf},
  url_link={https://ieeexplore.ieee.org/document/7049320},
  abstract={We present results from a set of experiments in this pilot study to investigate the causal influence of user activity on various environmental parameters monitored by occupant-carried multi-purpose sensors. Hypotheses with respect to each type of measurements are verified, including temperature, humidity, and light level collected during eight typical activities: sitting in lab / cubicle, indoor walking / running, resting after physical activity, climbing stairs, taking elevators, and outdoor walking. Our main contribution is the development of features for activity and location recognition based on environmental measurements, which exploit location- and activity-specific characteristics and capture the trends resulted from the underlying physiological process. The features are statistically shown to have good separability and are also information-rich. Fusing environmental sensing together with acceleration is shown to achieve classification accuracy as high as 99.13%. For building applications, this study motivates a sensor fusion paradigm for learning individualized activity, location, and environmental preferences for energy management and user comfort.},
  keywords={Data mining, Smart city, Energy system}}
  

We present results from a set of experiments in this pilot study to investigate the causal influence of user activity on various environmental parameters monitored by occupant-carried multi-purpose sensors. Hypotheses with respect to each type of measurements are verified, including temperature, humidity, and light level collected during eight typical activities: sitting in lab / cubicle, indoor walking / running, resting after physical activity, climbing stairs, taking elevators, and outdoor walking. Our main contribution is the development of features for activity and location recognition based on environmental measurements, which exploit location- and activity-specific characteristics and capture the trends resulted from the underlying physiological process. The features are statistically shown to have good separability and are also information-rich. Fusing environmental sensing together with acceleration is shown to achieve classification accuracy as high as 99.13%. For building applications, this study motivates a sensor fusion paradigm for learning individualized activity, location, and environmental preferences for energy management and user comfort.

@INPROCEEDINGS{2014_1C_model,
  author={Z. {Kang} and M. {Jin} and C. J. {Spanos}},
  booktitle={Annual Conference of the IEEE Industrial Electronics Society (IECON)}, 
  title={Modeling of end-use energy profile: An appliance-data-driven stochastic approach}, 
  year={2014},
  volume={},
  number={},
  pages={5382-5388},
  doi={10.1109/IECON.2014.7049322},
  url_link={https://ieeexplore.ieee.org/abstract/document/7049322},
  url_pdf={model_use.pdf},
  abstract={In this paper, the modeling of building end-use energy profile is comprehensively investigated. Top-down and Bottom-up approaches are discussed with a focus on the latter for better integration with occupant information. Compared to the Time-Of-Use (TOU) data used in previous Bottom-up models, this work utilizes high frequency sampled appliance power consumption data from wireless sensor network, and hence builds an appliance-data-driven probability based end-use energy profile model. ON/OFF probabilities of appliances are used in this model, to build a non-homogeneous Markov Chain, compared to the duration statistics based model that is widely used in other works. The simulation results show the capability of the model to capture the diversity and variability of different categories of end-use appliance energy profile, which can further help on the design of a modern robust building power system.},
  keywords={Energy system, Data mining}}
  

In this paper, the modeling of building end-use energy profile is comprehensively investigated. Top-down and Bottom-up approaches are discussed with a focus on the latter for better integration with occupant information. Compared to the Time-Of-Use (TOU) data used in previous Bottom-up models, this work utilizes high frequency sampled appliance power consumption data from wireless sensor network, and hence builds an appliance-data-driven probability based end-use energy profile model. ON/OFF probabilities of appliances are used in this model, to build a non-homogeneous Markov Chain, compared to the duration statistics based model that is widely used in other works. The simulation results show the capability of the model to capture the diversity and variability of different categories of end-use appliance energy profile, which can further help on the design of a modern robust building power system.