On enhancing the deadlock-preventing object migration automaton using the pursuit paradigm

Probably, the most reputed solution for partitioning, which has applications in databases, attribute partitioning, processor-based assignment and many other similar scenarios, is the object migration automata (OMA). However, one of the known deficiencies of the OMA is that when the problem size is large, i.e., the number of objects and partitions are large, the probability of receiving a reward, which “strengthens” the current partitioning, from the Environment is not significant. This is because of an internal deadlock scenario which is discussed in this paper. As a result of this, it can take the OMA a considerable number of iterations to recover from an inferior configuration. This property, which characterizes learning automaton (LA) in general, is especially true for the OMA-based methods. In spite of the fact that various solutions have been proposed to remedy this issue for general families of LA, overcoming this hurdle is a completely unexplored area of research for conceptualizing how the OMA should interact with the Environment. Indeed, the best reported version of the OMA, the enhanced OMA (EOMA), has been proposed to mitigate the consequent deadlock scenario. In this paper, we demonstrate that the incorporation of the intrinsic properties of the Environment into the OMA’s design leads to a higher learning capacity and to a more consistent partitioning. To achieve this, we incorporate the state-of-the-art pursuit principle utilized in the field of LA by estimating the Environment’s reward/penalty probabilities and using them to further augment the EOMA. We also verify the performance of our proposed method, referred to as the pursuit EOMA (PEOMA), through simulation, and demonstrate a significant increase in the convergence rate, i.e., by a factor of about forty. It also yields a noticeable reduction in sensitivity to the noise in the Environment. The paper also includes some results obtained for a real-world application domain involving faulty sensors.