takefuji@uniks.ece.scarolina.EDU.UUCP (12/02/87)
A Conductance programmable "neural" chip based on a Hopfield model employs deterministically/stochastically controlled switched resistors Yutaka Akiyama*, Yoshiyasu Takefuji**, Yong B. Cho**, Yoshiaki Ajioka*, and Hideo Aiso* * Keio University Department of Electrical Engineering 3-14-1 Hiyoshi, Kouhoku-ku, Yokohama 223 JAPAN ** University of South Carolina Department of Electrical and Computer Engineering Columbia, SC 29208 (803)-777-5099 Abstract The artificial neural net models have been studied for many years. There has been a recent resurgence in the field of artificial neural nets caused by Hopfield. Hopfield models are suitable for VLSI implementations because of the simple architecture and components such as OP Amps and resistors. However VLSI techniques for implementing the neural models face difficulties dynamically changing the values of the conductances Gij to represent the problem constraints. In this paper, VLSI neural network architectures based on a Hopfield model with deterministically/stochastically controlled variable conductances are presented. The stochastic model subsumes both functions of the hopfield model and Boltzmann machine in terms of neural behaviors. We are under implementations of two CMOS VLSI neural chips based on the proposed methods. _______________________________________________________________________________ Multinomial Conjunctoid Statistical Learning Machines Yoshiyasu Takefuji, Robert Jannarone, Yong B. Cho, and Tatung Chen Unversity of South Carolina Department of ECE Columbia, SC 29208 (803)777-5099 ABSTRACT Multinomial Conjunctoids are supervised statistical modules that learn the relationships among binary events. The multinomial conjunctoid algorithm precluded the following problems that occur in existing feedforward multi-layerd neural networks:(a) existing networks often cannot detemine underlying neural architectures, for example how many hidden layers should be used, how many neurons in each hidden layer are required, and what interconnections between neurons should be made;(b) existing networks cannot avoid convergence to suboptimal solutions during the learning process; (c) existing networks require many iterations to converge, if at all, to stable states; and (d) existing networks may not be sufficiently general to reflect all learning situations. By contrast multinomial conjunctoids are based on a well-developed statistical decision theory framework, which guarantees that learning algorithms will converge to optimal learning states as the number of learning trials increases, and that convergence during each trial will be very fast. _________________________________________________________________________ Conjunctoids: Statistical Learning Modules for Binary Events Robert Jannarone, Kai Yu, and Y. Takefuji University of South Carolina Department of ECE Columbia, SC 29208 (803)777-7930 ABSTRACT A general family of fast and efficient PDP learning modules for binary events is introduced. The family (a) subsumes probabilistic as well as functional event associations; (b) subsumes all levels of input/output associations; (c) yields truly parallel learning processes; (d) provides for optimal parameter estimation; (e) points toward a workable description of optimal model performance; (f) provides for retaining and incorporating previously learned information; and (g) yields procedures that are simple and fast enough to be serious candidates for reflecting both neural functioning and real time machine learning. Examples as well as operationial details are provided. _________________________________________________________________________ If you need the full copies of those papers, please state which papers you are requesting through Email, phone, or USmail. For Multinomial and VLSI neural chips papers: Dr. Y. Takefuji University of South Carolina Deparment of Electrical and Computer Engineering Columbia, SC 29208 (803)777-5099 (803)777-4195 takefuji@uniks.ece.scarolina.edu For Conjuncoids papers: Dr. Robert Jannarone University of South Carolina Department of Electrical and Computer Engineering Columbia, SC 29208 (803) 777-7930 jann@uniks.ece.scarolina.edu Thank you...