Fast adaptive search on the line in dual environments

Abstract

A stochastic point location problem considers that a learning mechanism (agent, algorithm, etc.) searches the target point on a one-dimensional domain by operating a controlled random walk after receiving some direction information from a stochastic environment. A method named Adaptive Step Search has been the fastest algorithm so far for solving a stochastic point location problem, which can be applied in Particle Swarm Optimization (PSO), the establishment of epidemic models and many other scenarios. However, its application is theoretically restrained within the range of informative environment in which the probability of an environment providing a correct suggestion is strictly bigger than a half. Namely, it does not work in a deceptive environment where such a probability is less than a half. In this paper, we present a novel promotion to overcome the difficult issue facing Adaptive Step Search, by means of symmetrization and buffer techniques. The new algorithm is able to operate a controlled random walk in both informative and deceptive environments and to converge eventually without performance loss. Finally, experimental results demonstrate that the proposed scheme is efficient and feasible in dual environments.