Games in Machine Learning
MSc course · TU Wien · Summer semester 2025/26
How do we design and analyze learning algorithms when several agents interact, each with its own objective? This course develops the mathematical foundations of learning in games, connecting optimization, equilibrium problems, and machine learning.
Topics include:
- Foundations: convex optimization, game-theoretic solution concepts, and simultaneous and sequential games.
- Algorithms and analysis: variational inequalities, fixed-point methods, monotone operators, and operator splitting.
- Further perspectives: dynamical systems and acceleration, mirror methods, and applications to adversarial learning.
The course has concluded. Slides, annotated slides, exercises, and solutions are available below.
Lecturer: Tatjana Chavdarova
Course Materials
-
Lecture 1: Intro, course organization, game-theoretic solutions
• Slides - Part 1 • Slides - Part 2 • Exercises • Solutions
Covers:
• Motivation to Study Games in ML
• What Makes a Game? (an ML perspective)
• Games & Intelligence: Historical Remarks
• Game Theoretic Solution Concepts
• Optimization Intro
-
Lecture 2: Optimization Crash Course
• Slides-Part-1 • Annotated-Part-1 • Slides-Part-2 • Annotated-Part-2 • Exercises • Solutions
Covers:
• Optimization Intro
• Smooth Functions & Gradient Descent
• Convexity: Sets & Functions
• Gradient Descent Convergence on (Strongly) Convex Functions
• Stochastic Gradient Descent
-
Lecture 3: Two Player Simultaneous Games
• Slides • Annotated • Exercises • Solutions • Code-demo
Covers:
• Differentiable Two-player Games and Differentiable Nash Equilibria
• Minimax Theorem & Strong Linear Programming Duality Connection
• Methods: Gradient Descent, Extragradient, Proximal Point, Optimistic Gradient Descent, Lookahead
• PP perspective of ExtraGradient & Optimistic Gradient Descent
-
Lecture 4: Practical and Sequential Aspects of Two-Player Games
• Slides-Part-1 • Slides-Part-2 • Code-demo-GANs
Covers:
• Aspects when designing Two-player Algorithms; main case study: GANs
• Practical Aspects; w/ demo on Generative Adversarial Nets
• Solving Two-player Sequential (Stackelberg) Games
-
Lecture 5: Variational Inequalities
• Slides • Annotated slides • Exercises • Solutions
Covers:
• Variational inequalities as a unified framework for optimization and n-player games
• Complementarity problems and normal-cone formulations
• Operator classes: monotonicity, Lipschitz continuity, cocoercivity, and averaged operators
• Spectral viewpoints on variational inequalities and operator assumptions
• Monotone inclusions, fixed-point iteration, and convergence of averaged iterations
• Local convergence through the spectral radius
-
Lecture 6: Operator Theory Crash Course
• Slides • Annotated slides • Exercises • Solutions
Covers:
• Maximal monotone operators and Minty’s surjectivity theorem
• Firmly nonexpansive operators, resolvents, and reflected resolvents
• Proximal maps, projections, and resolvent identities
• Forward-backward and backward-forward splitting
• Peaceman–Rachford and Douglas–Rachford splitting and their convergence properties
• ADMM as Douglas–Rachford splitting in the dual space
-
Lecture 7: A dynamical systems view on accelerated optimization
Guest lecture by Michael Muehlebach
Covers:
• Gradient descent, momentum, and acceleration through continuous-time dynamics
• Geometric and spectral views of convergence and conditioning
• Edge of stability and implicit regularization in deep learning
• Sparse and low-rank training methods for transformer models
Lecture abstract
The lecture explores optimization algorithms through the lens of dynamical systems, highlighting how classical gradient methods, momentum schemes, and accelerated algorithms can be understood as discrete-time approximations of continuous dynamics. Starting from gradient descent and Nesterov acceleration, I will discuss how geometric and spectral viewpoints explain convergence rates, acceleration, and the role of conditioning in large-scale optimization. The lecture will then connect optimization theory to current questions in deep learning, including the edge-of-stability phenomenon and the implicit regularization effects of training dynamics. I will also discuss recent directions in structure-aware learning, with an emphasis on sparse and low-rank training methods for transformer models. Overall, the lecture shows how a dynamical systems perspective sheds light on classical optimization results while also informing the design of algorithms for modern deep learning problems.
About Michael Muehlebach
Michael Muehlebach studied mechanical engineering at ETH Zurich and specialized in robotics, systems, and control during his Master's degree. He received the B.Sc. and the M.Sc. in 2010 and 2013, respectively, before joining the Institute for Dynamic Systems and Control for his Ph.D. He graduated under the supervision of Prof. R. D'Andrea in 2018 and joined the group of Prof. Michael I. Jordan at the University of California, Berkeley as a postdoctoral researcher. In 2021 he started as an independent group leader at the Max Planck Institute for Intelligent Systems in Tuebingen, where he leads the group "learning and dynamical systems".
He is interested in a variety of subjects, including machine learning, dynamical systems, and optimization. During his Ph.D. he developed approximations to the constrained linear quadratic regulator problem, a central problem in control theory, and applied these to model predictive control. He also designed control and estimation algorithms for balancing robots and flying machines. His more recent work straddles the boundary between machine learning and optimization, and includes the analysis of momentum-based and constrained optimization algorithms from a dynamical systems point of view.
He received the Outstanding D-MAVT Bachelor Award for his Bachelor's degree and the Willi-Studer prize for the best Master's degree. His Ph.D. thesis was awarded with the ETH Medal and the HILTI prize for innovative research. He was also awarded a Branco Weiss Fellowship, Emmy Noether Fellowship, and an Amazon research grant, which fund his research group. He is also a member of the editorial board of Foundations and Trends in Machine Learning.
-
Lecture 8: Mirror Descent Methods
• Slides • Annotated slides • Exercises • Solutions
Covers:
• Choosing the geometry: limitations of Euclidean gradient descent and variable-metric methods
• Mirror maps, Bregman divergence, and dual norms
• Mirror descent: primal and dual views, and its relation to gradient descent
• Mirror-Prox for monotone games and variational inequalities
• Connection to extragradient and convergence rates for the averaged iterate
-
Lecture 9: Advanced Topics
Materials not posted.
References
Main course reference
Large-Scale Convex Optimization: Algorithms & Analyses via Monotone Operators
Ernest K. Ryu and Wotao Yin.
Cambridge University Press, 2022.
Further reading
Finite-Dimensional Variational Inequalities and Complementarity Problems
Francisco Facchinei and Jong-Shi Pang.
Springer, 2003.
Volume I ·
Volume II
Contact
For corrections, comments, or requests for the LaTeX source of the slides, please contact: tatjana.chavdarova[at]tuwien[dot]ac.at.
Thanks!