A Study of Knowledge Acquisition Techniques -A Learning Method for Extracting Statistical Knowledge from Observation Data Based on Neural Networks

摘要

This paper presents a neural network method for learning the statistical knowledge (i.e., the a posteriori probability functions) from observation data sequences under the condition in which the a priori probability density functions about the observation sources are not provided by the observation environment. A suitable neural network architecture and a reinforcement learning algorithm are designed for learning the posterior probability functions from the observation data of two hypothesis sources. It is shown that the network architecture is an optimal model that can approach the target functions in arbitrarily small error. Simulation experiments conducted in the paper show that the neural-network based estimator can closely approximate the target functions, which confirms the validity of the proposed approach and the theoretical analysis of the paper.