A study on the Convergence of different Learning Strategies in a Multi-teacher Environment

摘要

Learning strategy in a Multiteacher environment is not unique. Multiteacher environment provides an opportunity to learn different strategies. By processing the environmental response in different ways, different strategies can be learned. Three major learning strategies are, to find the action with the least weighted average penalty (LWAP), to find the action which has the maximum probability of collectively getting the majority approval (MPMA) and to find the action which maximum number of teachers independently agree is the best (MNTA). All of the learning algorithms expect MNTA are known to be absolutely expedient and ε-optimal in a stationary environment. In this paper, the rate of convergence and the accuracy of convergence of the three learning strategies are studied. For simulation a seven action set in a five teacher environment is considered.