An extension of weighted strategy sharing in cooperative Q-learning for specialized agents
S.M. Eshgh, Majid Nili Ahmadabadi · 2004
Using other agents' experiences and knowledge, a learning agent may learn faster, make fewer mistakes, and create some rides for unseen situations. These benefits will be gained if the learning agents know the area of expertise and the expertness values of each other. In this paper, some Q-learning agents with different skills and expertness levels cooperate in learning. The agents use some criteria to judge others information and knowledge. Four expertness criterion, certainty and entropy measures are used to assign degrees of importance to others' Q-Tables. Effects of measuring these values based on their whole Q-Table, a portion of Q-Tables that reflects their proficiencies, and the states in Q-Tables on the learning quality are studied. Simple strategy sharing and two different weighted strategy-sharing methods are used to combine the acquired knowledge from different agents.