Accession Number : ADA296789

Title :   Connectionist Models for Intelligent Computation.

Descriptive Note : Final rept. 1 May 91-30 Apr 94,

Corporate Author : MARYLAND UNIV COLLEGE PARK LAB FOR PLASMA RESEARCH

Personal Author(s) : Chen, H. H. ; Lee, Y. C.

PDF Url : ADA296789

Report Date : 28 JUL 1994

Pagination or Media Count : 9

Abstract : This final report covers the work done by our group of neural network computing at the University of Maryland for the past three years. We studied the neural network's capability of processing temporal or sequential data. Recurrent neural networks were used to perform inference cn grammers. An external memory stack was constructed to work with the neural network to perform inferences on context free languages. And finally, a spatially homogeneous locally connected recurrent neural network that could simulate any given turing machine, including the universal Turing machine was devised. It is capable of performing universal computations and demonstrated the universal power of recurrent neural network architectures. To train these sequential neural net machine, we have investigated the forward propagating learning algorithms. (KAR) P. 1

Descriptors :   *ALGORITHMS, *NEURAL NETS, *COMPUTATIONS, *LEARNING, DATA PROCESSING, MARYLAND, PROGRAMMING LANGUAGES, COMPUTER ARCHITECTURE, SEQUENCES, MEMORY DEVICES, ARTIFICIAL INTELLIGENCE, STACKING.

Subject Categories : Cybernetics

Distribution Statement : APPROVED FOR PUBLIC RELEASE