Martin Müller is interested in developing efficient search methods for hard problems.

Search, plan, Go!

Martin Müller's main area of research is modern heuristic search, with its complex interactions between search, knowledge, simulations and machine learning. He and his research team work on understanding and improving Monte Carlo tree search, exploring and sampling in reinforcement learning, exploration in SAT, search and deep learning for Hex, and combinatorial game theory – especially developing efficient algorithms that combine search and subgame decomposition. He also works on domain-independent planning, random walk planning and motion planning and random sampling from time-changing discrete distributions.

Martin and his team have produced programs and algorithms for the games of Go, Amazons, Clobber and Hex. Many of his algorithms use all three of the modern heuristic search methods: search, knowledge and randomized simulations. He lead the development of the open source program Fuego, which in 2009 first won a 9x9 Go game on even terms against a top-ranked professional human player. Martin has worked on computer Go for 30 years and was the academic co-supervisor (along with Richard S. Sutton) of David Silver, who went on to become head of reinforcement learning at DeepMind and lead researcher on AlphaGo. With his students and colleagues, Müller has developed a series of successful game-playing programs, planning systems, and SAT solvers.