Random Cloud Least Mean-Squares Adaptive Filters

This work proposes a simple class of random search algorithms based on a cloud of random agents that recursively seek the optimal mean-square solution. In the first kind, a sequence of random clouds directly seek the optimal solution; in the second, the cloud orbits an adaptive filter estimate in order to explore the neighborhood during adaptation. Both kinds have the same two parameters, cloud population and spread, and are easy to tune at a linear complexity in the population size for the cloud generation. They outperform standalone adaptive filters as well as the Particle Swarm Optimization algorithm in the system identification setup.