Application of Some New Artificial Neural System Processing Principles to Pattern Classification

摘要

The overall goal of the Army Research Office (ARO) study is to provide a firm conceptual and theoretical foundation for future work in ANS by exploring the use of three information processing principles. This entails both experimental exploration and theoretical development of design laws that can quantitatively predict system behavior and performance. This interim report describes the pattern environment which will support the planned experimentation. The pattern environment is a simple version of spoken english (called Lenglish), where each letter is associated with an arbitrary sound spectrum and there is a continuous transition between the spectra of adjacent letters. The signal input to the ANS will be the Lenglish reading of selected source texts. The planned experimentation will use a model of unsupervised learning to discover feature detectors for spatiotemporal segments of the input language.