Lecture 1: Welcome to Machine Learning
Course Introduction & Learning ML Through a Cat’s Eyes
Overview
Welcome to Machine Learning! This course takes a unique “Learn by Building” approach where you’ll implement real ML systems from day one. In today’s hands-on session, we’ll explore the three fundamental paradigms of machine learning—supervised, unsupervised, and active learning—by building CatShop, an e-commerce system that thinks like a cat using Google’s Gemma-3 language model.
Learning Objectives
By the end of this lecture, you will:
- Experience all three ML paradigms through hands-on implementation
- Fine-tune Gemma-3 (270M parameters) using LoRA for efficient adaptation
- Observe catastrophic forgetting and model trade-offs
- Build a complete ML application from data to deployment
- Understand how active learning reduces labeling costs
Materials
Complete Pre-Lecture 1 Setup before the lecture!
Join the course Piazza forum (Required for participation)
Complete the Project Matchmaker Form by Mon 8/26, 12:00 PM ET. Required for Lecture 2: k-NN; counts toward participation.
Interactive Demo
Experience your trained model in action with the CatShop e-commerce demo, featuring real-time cat perspective classifications, confidence visualization, and interactive chat.
Built upon the WebShop environment with modifications for educational purposes.
Acknowledgments & Resources
Core Technologies
- Gemma-3 270M: Google’s efficient language model
- PEFT/LoRA: Parameter-efficient fine-tuning
- Transformers: Hugging Face’s model library
Dataset & Environment
- WebShop: E-commerce product data and environment
Yao et al., “WebShop: Towards Scalable Real-World Web Interaction with Grounded Language Agents”, NeurIPS 2022