A global stochastic optimization particle filter algorithm

Summary We introduce a new online algorithm for expected loglikelihood maximization in situations where the objective function is multimodal or has saddle points. The key element underpinning probability distribution that concentrates on target parameter value as sample size increases and can be efficiently estimated by means of standard particle filter algorithm. This depends learning rate, such faster rate quicker desired search space, but less likely to escape from local optimum function. In order achieve fast convergence with slow our exploits acceleration property averaging, which well known stochastic gradient literature. Considering several challenging estimation problems, numerical experiments show high probability, successfully finds highest mode converges global maximizer at optimal rate. While focus this work maximization, proposed methodology its theory apply more generally optimization defined through an expectation.