AN ONLINE UNSUPERVISED LEARNING-MACHINE FOR ADAPTIVE FEATURE-EXTRACTION

Adaptive feature extraction is useful in many information processing s ystems. In this paper we propose a learning machine implemented via a neural network to perform such a task using the tool principal compone nt analysis. This machine (1) is adaptive to nonstationary input, (2) is based on an un-supervised learning concept and requires no knowledg e of if, or when, the input changes statistically, and (3) performs on -line computation that requires little memory or data storage. Associa ted with this machine, we propose a learning algorithm (LEAP), whose c onvergence properties are theoretically analyzed and whose performance is evaluated via computer simulations. Two major contributions of thi s paper are: (1) Under appropriate conditions, we prove that the algor ithm will extract multiple principal components, when the learning rat e is constant; and (2) we identify a near optimal domain of attraction .