Towards a cooperative machine learning simulator
Jason M. Black · 2006
Cooperative multi-agent domains are of current interest due to their relevance to goals in both robotics and networking. Such problems are non-trivial and can be exponentially more difficult to solve as the problem space increases in size. A common approach to solving these problems is through the application of machine learning. Techniques such as reinforcement learning, genetic algorithms, and genetic programming, have been studied and applied in this domain with varying degrees of success. However, researchers usually create new implementations of a problem domain instead of using previous implementations created by other researchers. While a growing understanding of the relationships between problem domains and machine learning techniques has emerged over time, there have been few attempts to measure the performance of machine learning techniques at the level of the entire research community. The goal of this research is the design of a non-trivial simulation environment geared towards the efficient addition and testing of machine learning algorithms given a wide variety of conditions, with a concentration on cooperative tasks.