Game Theory in AIC

Players



old guys

Viliam Lisý

Branislav Bošanský



new blood

Karel Horák

Karel Ha

Michal Šustr

Dominik Seitz

Petr Tomášek



training (almost) completed

Jiří Čermák (@Blindspot)

Karel Durkota



training in a galaxy far far away

Jakub Černý (@NTU)

Goals

... the goal is to win

  • Dynamic games with partial information (blind chess, poker, pursuit-evasion games, ...)
  • Security games (How to secure a computer network? Hackers are really smart ... )
  • Deception (How to deceive your opponent and not get fooled?)
  • Adversarial Machine Learning (Even a machine can learn ... be one step ahead!)

Methods

... this is what we do

  • equilibrium computation algorithms for dynamic games (extensive-form games, stochastic games)
  • counterfactual regret minimization and continual resolving (DeepStack)

Tools

... every Jedi needs his light-saber

  • we prove stuff (showing theoretical guarantees and/or limitations of our algorithms)
  • we implement and experimentally validate our algorithms (Java, C++, TensorFlow)
  • domain-independent implementation of all our algorithms (hg clone http://jones.felk.cvut.cz/repo/gtlibrary)

Play a game

... work on difficult and interesting challenges

  • Bachelors/Masters projects
  • PhD positions
  • part-time positions

... collaborate with other researchers/industry

  • projects and active collaboration with Carnegie Mellon University, UTEP, University of Alberta/DeepMind, IST Austria
  • projects with TrendMicro

Which game specifically can you play?

... only examples, you can propose your own

  • design an algorithm for the Scotland Yard board game (e.g., improve our domain-independent algorithm for this class of games)
  • design an algorithm for blind chess (Kriegspiel)
  • combine planners (e.g., A*) with game-theoretic algorithms (planning in an adversarial environment)
  • combine classifiers (e.g., simple image recognition, anomaly detection in computer network traffic) with game-theoretic algorithms (adversarial machine learning)
  • win over humans (build a mathematical model for human behavior for some game, design an algorithm that exploits human irrationality)
  • do the theory/math for combination of classifiers and game theory

Contact us: viliam.lisy@fel.cvut.cz bosansky@fel.cvut.cz