Accession Number : ADA256227


Title :   Stability and Adaptation of Neural Networks.


Descriptive Note : Final rept. 1 Aug 88-31 Dec 91,


Corporate Author : UNIVERSITY OF SOUTHERN CALIFORNIA LOS ANGELES DEPT OF ELECTRICAL ENGINEERING


Personal Author(s) : Kosko, Bart


Report Date : 23 SEP 1992


Pagination or Media Count : 120


Abstract : This research studied the stability, adaptation, and robustness of neural networks and fuzzy systems. Key results include the stability of random adaptive bidirectional associative memories (RABAMs) and neural-fuzzy competitive and differential-Hebbian ABAMs, the introduction and analysis and testing of the differential competitive learning law, new theorems on the stochastic convergence of competitive learning for vector quantization, a universal approximation theorem for fuzzy systems, unsupervised schemes for Teaming fuzzy rules with neural networks with tests on truck-and-trailer control systems and coding and compression of still images and image sequences. Neural networks, unsupervised learning, robustness, stability, competitive learning, fuzzy systems, neural-fuzzy systems, phoneme recognition, image compression, truck and-trailer control systems.


Descriptors :   *NEURAL NETS, ADAPTATION, CODING, COMPRESSION, CONTROL, CONTROL SYSTEMS, CONVERGENCE, IMAGES, LEARNING, NETWORKS, PHONEMES, QUANTIZATION, RECOGNITION, SEQUENCES, STABILITY, TEST AND EVALUATION, THEOREMS, TRAILERS, TRUCKS, OPERATIONAL EFFECTIVENESS, STOCHASTIC PROCESSES.


Subject Categories : CYBERNETICS


Distribution Statement : APPROVED FOR PUBLIC RELEASE