|
|
| Academic Year |
2024/2025 |
| Name |
Dott. - MI (1380) Ingegneria dell'Informazione / Information Technology |
| Programme Year |
1 |
| ID Code |
063164 |
| Course Title |
GAMES IN MACHINE LEARNING (GML) |
| Course Type |
MONO-DISCIPLINARY COURSE |
| Credits (CFU / ECTS) |
5.0 |
| Course Description |
1. GML Motivation & Optimization Preliminaries (1st-day full lectures, 4.5h)
Motivation: (1h)
Game Theory Intro
What Makes a Game? (a Machine Learning perspective)
Examples: games in Machine Learning
Optimization Intro (1.5h)
Overview & Complexity analysis
Gradient Descent
Convexity: convex sets & (strongly) convex functions
Convex Optimization - Convergence (2h)
Gradient Descent Convergence on (strongly) convex functions
Stochastic Gradient Descent
2. Two-Player Games & Methods (2nd day: 4h lectures + 1h student presentation)
Two-Player Games
Two-Player Games:
normal form games, zero-sum and general-sum games
Minimax Theorem
Methods for Solving Two-Player Games:
Gradient Descent (GD)
Extragradient (EG)
Proximal Point (PP)
Optimistic Gradient Descent (OGD)
Lookahead (LA)
EG and OGD as Proximal Point Approximation
Framework Equilibrium Analysis: GANs case study
Examples of Two-Player Games: student presentations of primarily practical papers
3. Equilibrium-Finding Problems: Variational Inequality (3rd-day full lectures, 4.5h)
Variational Inequality Framework: Definition & Examples
Problem Classes & Operator Theory:
Lipschitz, Contractive, Averaged, Non-expansive, (strongly) Monotone
Spectral Viewpoint
Maximality, (Reflected) Resolvent
Monotone Inclusion
Fixed Point Iteration
Convergence of Averaged Operators
Local Convergence: Necessary and Sufficient Condition
4. Advanced VI Methods: (4th day: 4h lectures + 1h student presentations)
Mirror Descent & Mirror Prox (2h)
Motivation
Variable-Metric Methods
Mirror Descent: Mirror maps, Bregman divergence, MD algorithm
Mirror Prox
Operator Splitting: (2h)
forward↔backward, Douglas-Rachford Splitting
VI methods & research topics: student presentations of primarily theoretical papers
5. Guest Lecture and Paper presentations (5th day: student presentations + guest lecture)
Paper presentations by students
Guest Lecture TBD |
| Scientific-Disciplinary Sector (SSD)
|
|
SSD Code
|
SSD Description
|
CFU
|
|
ING-INF/05
|
INFORMATION PROCESSING SYSTEMS
|
5.0
|
|
|
Alphabetical group
|
Name
|
Teaching Assignment Details
|
|
From (included)
|
To (excluded)
|
|
A
|
ZZZZ
|
Chavdarova Tatjana
|
|
|
