From Isolation to Cooperation: An Alternative View of a System of Experts

摘要: We introduce a constructive, incremental learning system for regression problems that models data by means of locally linear experts. In contrast to other approaches, the experts are trained independently and do not compete during learning. Only when prediction query is required cooperate blending their individual predictions. Each expert minimizing penalized local cross validation error using second order methods. this way, an able find distance metric adjusting size shape receptive field in which its predictions valid, also detect relevant input features bias on importance dimensions. derive asymptotic results our method. variety simulations properties algorithm demonstrated with respect interference, speed, accuracy, feature detection, task oriented