Learning Automata for Pattern Classification

Abstract

In the previous two chapters we have seen some methods of employing collectives of learning automata to tackle different learning problems. In Chapter 2 we have presented the model of a set of automata engaged in a general game situation. The automata involved can be FALA or CALA. We have provided learning algorithms so that the automata can converge to (close approximation of) the optimal points of the game. We have also illustrated how such a model can be used for maximizing a function under noisy measurements (with no gradient information available), in applications such as system identification, and learning conjunctive concepts from noisy examples. In Chapter 3 we have seen how we can build much more powerful models by combining such teams of automata into networks. The structure of such networks can be very general, thus allowing a lot of flexibility in designing automata solutions to specific learning problems. As has been mentioned, we can also utilize other automata models such as PLA and GLA in such structures. We have also shown that with PLA we can have learning algorithms (for networks of automata) that converge to the global maximum of expectation of reinforcement.