Incremental Social Learning Applied to a Decentralized Decision-Making Mechanism: Collective Learning Made Faster

Abstract

Positive feedback and a consensus-building procedure are the key elements of a self-organized decision-making mechanism that allows a population of agents to collectively determine which of two actions is the fastest to execute. Such a mechanism can be seen as a collective learning algorithm because even though individual agents do not directly compare the available alternatives, the population is able to select the action that takes less time to perform, thus potentially improving the efficiency of the system. However, when a large population is involved, the time required to reach consensus on one of the available choices may render impractical such a decision-making mechanism. In this paper, we tackle this problem by applying the incremental social learning approach, which consists of a growing population size coupled with a social learning mechanism. The obtained experimental results show that by using the incremental social learning approach, the collective learning process can be accelerated substantially. The conditions under which this is true are described.