Collaborative Agent-Based Learning with Limited Data Exchange (Extended Abstract)
Xavier Rafael, Palou Michael Rovatsos · 2009
We describe a collaborative agent-based learning model suit- able for environments with limited data exchange and pro- vide an overview of its empirical evaluation. techniques since MAS can contribute a number of crucial capabilities that may be useful for solving DDM problems. While different approaches exist in the literature, a recent Multiagent Learning Framework (MALEF) (4) attempts to ensure adherence to agency aspects like collaboration be- tween heterogeneous classifiers, decentralised learning con- trol, and the autonomy of self-directed learning processes. This abstract framework uses communication and collabora- tion among the different local classification learning agents. The learning agents perform a series of consecutive learning steps using two functions: classifier training and measure- ment of the resulting classifier quality. They may addition- ally perform integration operations using different parts of the knowledge received from other learning agents. Also, a number of very generic categories of integration operations are proposed in the framework.