April 2022: I am currently not taking on new graduate or undergraduate students
If you want to come to Alberta, please read about Computing Science Graduate Studies and click the "How to Apply" button.
The recent and current research in our group covers a number of themes in search and learning. You can also look at some of our Publications and Talks.
These are just some ideas, and new ones pop up constantly. They should give you an indication of which possible research directions we are looking at right now.
While most Go positions are far too complicated to allow a complete analysis, full-scale late stage Go endgame puzzles can be solved exactly with the search method of decomposition search. This method implements a divide and conquer approach based on concepts from combinatorial game theory.
I this project we want to study how modern Go programs based on the Alpha Zero algorithms learn to play such difficult puzzles, and whether there are limits to how well they can learn them.
This project is part of the theme Evaluation of deep RL learning against exact solutions in two player games
Reconnaissance Blind Chess (RBC) is a variant of chess where the opponent's moves are hidden from the player. Before each move, a player can view a 3x3 region of the board, in order to learn more about the opponent's actions.
This can be a joint project with J. Fürnkranz and his group at JKU Linz.
Some References:
See the CGT project ideas page.
MSc student Zeyi Wang is currently working on an open source reimplementation of DeepMind's MuZero architecture. Once this MuZero re-implementation is available, we should study in more detail how the algorithm works, how it can be applied to other problems, and how it can be improved. For example,
Some References:
Created: Aug 25, 2012 Last modified: Apr 10, 2022
Martin Müller