Analysis of Reinforcement Learning Algorithms for Swarm Learning

Reinforcement learning has received much attention in the past decades. The three forms of reinforcement learning algorithms are Actor Critic learning, Q learning and Reinforcement Comparison. Q-learning is a form of model-free reinforcement learning with one drawback that is the overestimation (Rising Q) problem. To solve this problem Rough Sets approach is used. This has lead to the modification of the traditional Q learning algorithms to a new form of Q learning namely, Rough Q learning. Actor Critic Learning have a separate memory structure to explicitly represent the policy independent of the value function. Another form of reinforcement learning is Reinforcement Comparison. Using reinforcement comparison method (RC), a reference reward is equated with an average of previously received rewards. The Actor Critic and RC method is also made better by using the rough set approach. The results of the study are in form of various plots for all three forms of reinforcement learning, their variations and the effect of temperature on them.

Shamama Anwar is currently working as an Assistant Professor in the Department of Computer Science and Engineering at Birla Institute of Technology, Mesra, Ranchi, India. She has published research papers in esteemed journals like Taylor & Francis, Springer, etc.