Adaptive mixtures of local experts are source coding solutions
Peter T. Szymanski, Michael D. Lemmon · 2002
Research in intelligent control uses radial basis function networks to learn and perform tasks. The method lacks rigorous quantitative analysis and a physical interpretation of the ideas present. Current research in intelligent control uses information theory to describe and quantify solutions to control problems. This approach lacks flexibility. The connection between the two approaches is identified using a comparison of the two approaches. The main result is that the computation of the stochastic mapping in the adaptive mixtures approach is the same as the stochastic mapping in the source coding approach. This implies that the adaptive mixtures approach produces optimal mappings based on the analysis and results from source encoding theory.>