Speaker: Cameron Browne, (Queensland University of Technology (QUT) in Brisbane, Australia)
A New Approach to General Game Playing
General Game Playing (GGP) is a field of AI research that involves developing software for playing a range of games well, rather than playing a single game expertly. This talk outlines a new approach to GGP, based on a ‘ludemic’ model that decomposes games into fundamental conceptual units, to address several limiting factors of current GGP approaches. The resulting system, called LUDII, will use a Monte Carlo Tree Search (MCTS) approach enhanced by artificial neural networks (ANNs) for move planning, based on the state-of-the-art Go programme AlphaGo but extended to general games. A drawback of machine learning methods such as MCTS and ANNs is that they are ‘black box’ approaches that can provide good solutions to problems but without any indication of how those solutions were derived. An advantage of the ludemic model is that it explicitly labels the underlying concepts involved, which may be exploited to explain strategies automatically learnt by the move planner in human-comprehensible terms. This talk will outline the ideas behind LUDII, its operation, its intended purpose as a platform for AI research into automatic strategy learning, transfer and explanation, and why this is important.
|Date||May 26, 2017 (Fri) 16:00 - 17:00|