Learning in multi-agent systems

The issues involved in applying machine learning algorithms to multi-agent systems were discussed. Issues about multi-agent learning, including the difference between single-agent learning and multi-agent learning, on-line and off-line learning methods, and mechanisms for social learning were presented. The different design options namely on-line versus off-line, reactive versus logic-based learning algorithms, and social learning algorithms inspired by animal learning were also presented. It was found that logic-based agents have the advantage of being able to naturally incorporate domain knowledge in the learning process, while artificial life approaches can be based on evidence from biology.