Course Summary
This graduate-level course combines lectures with paper readings. It covers several foundational topics in learning theory and explores how these ideas inspire and connect with modern research in machine learning and AI. Topics include typical basic learning theory topics such as PAC learning and VC dimensions and modern learning theory topics including diffusion models and reinforcement learning.
Administrative Information
Lectures: Tu, Thu, 9:30-10:45
Location: SN011
Instructor: Zhun Deng
Teaching Assistants: TBD Office hours: TBD Location: TBD
Requirements: This is a graduate-level course that requires some mathematical background.
Required background: probability, discrete math, calculus, analysis, linear algebra, algorithms and data structures
Grading and Collaboration
Grading: Attendancy (15%), paper presentation (35%), and a final course project (50%).
Collaboration: Collaboration on presentation and course projects are allowed.
Textbook and Readings
1. Understanding Machine Learning: From Theory to Algorithms, by Shai Shalev-Shwartz and Shai Ben-David, available here
2. (Draft) Reinforcement Learning: Theory and Algorithms, by Alekh Agarwal, Nan Jiang, Sham M. Kakade, and Wen Sun, available here
3. Tutorial on Diffusion Models for Imaging and Vision, available here
4. Algorithmic Foundations of Interactive Learning, available here
Additional papers for reading will be posted in the paper list as the course progresses.
Tentative Schedule
Lecture 1: Course introduction
Lecture 2: Introduction to PAC learning (related reading: Book 1, Chapter 1-3)
Lecture 3: Uniform convergence and VC dimension (related reading: Book 1, Chapter 4-6)
Lecture 4: Variational Auto-Encoder
Lecture 5: Denoising Diffusion Probabilistic Model (DDPM)
Lecture 6: Score-Matching Langevin Dynamics (SMLD）
Lecture 7: Reinforcement learning with Human Feedback I
Lecture 8: Reinforcement learning with Human Feedback II