General Framework for Parameter Learning and Optimization in Stochastic Environments

The existing strategies on Stochastic Point Location (SPL) are adopted to search a point or parameter on a real line under stochastic environment, which have been demonstrated to be effective. However, one problem which still has not been addressed yet is how to learn a point in multidimensional space under stochastic environment. This problem will become more difficult when the environment itself is a deceptive one in which the probability (p) of the correct response emitted from the environment is p < 0.5. In this paper, a general framework is proposed to deal with all the above-mentioned problems. A key aspect of our method worth mentioning is that it can transform learning a point in space into one of searching d optimal points on d different super lines, where d is the dimension size. Finally, the excellent performance of our proposed algorithm has been proved by our rigorous mathematical proof and validated by a great number of experiments.