Topics of Interest
- Monte Carlo Tree Search
- Proof Number Search
- Heuristic Search
- Combinatorial Game Theory
- Imperfect Information Games
- Computational Social Choice
My PhD: Solving Games and All That
Strategy games constitute a good testbed for AI techniques. The task of building an intelligent player should be a lot easier than building an intelligent general agent. The world of a game is indeed much simpler than the physical world: for instance the goal and the dynamics and the possible interactions of a game are well defined and are known to the players which is not always the case in the real world.
Still, interesting games are usually complex. An intelligent player is expected to take a decision without an exhaustive search of the possible outcomes, not to repeat the same mistakes again and again, to be able to play well different games. Moreover, some games might involve chance (backgammon), hidden information (phantom go) or both (most card games). Thus, building an intelligent player is not trivial and has motivated decades of active research over the globe.
I implemented the GaDeLaC forward chaining based compiler for the Game Description Language. It is described in this paper and can be downloaded from its home page. GaDeLaC is written in OCaml and generates OCaml code to be further compiled to machine code. Writing a C backend is naturally an interesting direction that I am working on with Hilmar Finnsson.
During my stay at the TUM, I took the Model Checking class by Dr. Stefan Schwoon. This class was very interesting and I decided to take advantage of my Master's internship to investigate this field further under Stefan's supervision. My report is available here.
MCTS for Hex
When I needed to choose a Bachelor thesis topic, I knew I would like to dive into Artificial Intelligence. I was delighted when Pr. Tristan Cazenave offered me to work on applying UCT to the game of Hex. First of all, I knew Hex and enjoyed playing it, mainly on Little Golem. Then my program would compete at the Computer Olympiads, which was surely a motivation in itself.
My thesis went pretty well and Tristan is now my PhD supervisor. My report is available here, but it is written in French.