Performance enhancement of cooperative learning algorithms by improved decision making for context based application
Deepak Annasaheb Vidhate, Parag A. Kulkarni · 2016
Cooperation in learning (CL) can be understood in a multiagent system. In this the agents are capable of learning from both their own experiments and other agents' knowledge and expertise. Implementation of CL is a complicated task in the real world. In distributed systems several agents cooperate to achieve a common goal or accomplish a shared task. In particular, if there are different people or organizations with different goals and information, then a multiagent system (MAS) is needed to handle their interactions. In this paper, various issues related with cooperative machine learning are studied and implemented. A new set of improved cooperative learning algorithms is proposed in the paper. Expertness measuring criteria which were used in earlier work is further enhanced in proposed method. Six methods for measuring the agents' expertness are used i.e. Normal (Nrm), Absolute (Abs), Positive (P), Negative (N), Certainty (Cer) and Entropy (Ent). The novelty of this approach lies in the implementation of Weighted Strategy Sharing with expertness measuring criteria by means of Q-learning, Sarsa learning, Q(λ) and Sarsa(λ) learning algorithms. The paper shows implementation results and performance comparison of all these algorithms.