Date |
Topics |
Lecture Notes |
Additional reading |
|
|
|
|
1/6 |
Introduction |
Lecture 1
|
1/9 |
Mistake bound model, Perceptron |
Lecture 2
|
1/13 |
Perceptron continued, Consistency model |
Lecture 3
|
(SSBD): Section 3.1
|
1/16 |
PAC model, Connections between learning models |
Lecture 4
|
|
1/20 |
Agnostic learning I: Uniform convergence for finite classes |
Lecture 5
|
Intro to statistical learning theory (Boucheron, Bousquet, Lugosi).
(SSBD): Chapter 4
(SSBD): Chapter 5 for No-Free-Lunch theorem
|
1/23 |
Agnostic learning II: Uniform convergence for infinite classes via the Growth Function |
Lecture 6
|
|
1/27 |
Agnostic learning III: Uniform convergence for infinite classes via the VC dimension |
Lectures 7 & 8
|
|
1/30 |
Agnostic learning IV: Examples of VC classes |
|
|
2/3 |
Compression schemes |
Lecture 9
|
(SSBD): Chapter 30
|
2/6 |
Rademacher complexity |
Lectures 10–13
|
|
2/10 |
Bounds based on Rademacher complexity
|
|
|
2/13 |
Properties of Rademacher complexity
|
|
|
2/24 |
Properties of Rademacher complexity continued
|
|
|
2/27 |
Support vector machines, margin bounds
|
Lecture 14
|
|
3/3 |
Prediction with expert advice, Exponentially weighted average forecaster
|
Lectures 15 & 16 (+ 17)
|
|
3/6 3/10 |
Mixability, Exp-concave losses
Decision-theoretic online learning; Hedge
|
|
|
3/13 |
Game theory (approximating Nash equilibria)
|
Lecture 18 (+19)
|
|
3/17 |
Online Convex Optimization, Online Gradient Descent
|
Lecture 19 (+20)
|
|
3/20 |
Online Gradient Descent continued
Online-to-batch conversion
|
Lecture 20
|
|
3/24 |
Non-stochastic multi-armed bandits, EXP3
|
Lectures 21 & 22
|
|
3/27 |
Analysis of EXP3
|
|
|
3/31 |
Project presentations |
|
4/3 |
Project presentations |
|