This course provides an introduction to modern machine learning and data mining, with an emphasis on the probabilistic foundations and techniques underlying contemporary artificial intelligence. The course will study modern approaches to probabilistic modeling, uncertainty quantification, deep learning, generative modeling, and representation learning. Throughout the course, we will emphasize connections between mathematical and statistical principles and current developments in generative AI, multimodal learning, and foundation models. Another objective is to train students to use modern AI tools, especially large language models, effectively and critically in problem solving. The course therefore evaluates both traditional mathematical understanding and the ability to work productively with AI-assisted workflows.
Instructor: Prof. Yifan Hu.
Evaluation: 40% mid-term, 60% project, 20% bonus homework questions.
Mid-term exam:
2 hours closed-book written exam.
1 hour LLM-supported problem solving (to fix unsolved problems and solve new more challenging questions).
October 19.
LLM Policy: LLM is a must for the course. LLMs are prohibited during 2-hour closed-book exams but allowed during the last 1-hour exam. Whenever used, it should be mentioned clearly.
AI-Assisted Course Project:
The project is the primary assessment component of the course. Students (from a group of at most 4) will use modern AI tools as part of a substantial technical project, such as investigating an open question from the research literature or constructing an end-to-end data-to-decisions pipeline. The project will be evauated based on how students understands the problem, identifies difficulties, concrete milestone progress made. It will not be evaluated based on if the problem is solved or not. The final product of the project is a technical blog post.
Bonus homework questions:
Difficult problems will be given as bonus homework. The students are encouraged to try every means they could to address it.
Tentative Schedule
Tue, Sep. 8
Introduction to Modern Machine Learning and Probabilistic AI; Bayesian Linear Regression
Mon, Sep. 14
Gaussian Processes I
Mon, Sep. 21
Bayesian Optimization
Mon, Sep. 28
Variational Inference
Mon, Oct. 5
Bayesian Deep Learning
Mon, Oct. 12
Active Learning and Uncertainty Quantification
Mon, Oct. 19
Midterm Examination: 2-hour closed-book exam + 1-hour AI-assisted problem solving
Mon, Oct. 26
LLMs for Problem Solving: Hands-on Tutorial
Mon, Nov. 2
Generative Models: Latent-Variable Models, VAEs, and Diffusion Models
Mon, Nov. 9
Representation Learning I: Embeddings, Autoencoders, and Learning Useful Representations
Mon, Nov. 16
Representation Learning II: Self-Supervised Learning and Predictive Representations
Mon, Nov. 23
Contrastive Learning: Objectives, Negative Sampling, and Modern Methods
Mon, Nov. 30
Multimodal Representation Learning and Foundation Models
Mon, Dec. 7
Project Presentations + Poster sessions