Memory modeling, analysis, and design
Abstract
In this thesis, cognitive models of associative memory are developed. The cognitive view of memory involves information encoding, storage, and retrieval. Incoming information is first transformed into a form that can be processed by the memory system. For processed information to be remembered, storage must occur which entails transfer of the encoded information into memory. Finally, retrieval involves locating stored information when needed. In the first part of the thesis, a fully interconnected generalized Brain-State-in-a-Box neural network is used to model information storage and retrieval. In the second part, motivated by biological systems, sparsely interconnected neural associative memory is investigated. The design problems of fully and sparsely interconnected associative memories are formulated as constrained optimization programs, and “designer” networks for solving such programs in real time are proposed. Stability properties of the designer networks are analyzed using Barbalat's lemma. In the final part of the thesis, a neural network model is proposed to execute a complex memory search for information with sought attributes. The network is composed of chaotic neurons. The proposed model of a chaotic neuron is biologically inspired and accounts for the property of relative refractoriness, that is, gradual recovery of responsiveness of a biological neuron after a stimulus is applied to the neuron. Fuzzy logic is used to tune associative memory parameters for the purpose of directing network trajectory to patterns with sought features. Simulation results are presented to illustrate the results obtained.
Degree
Ph.D.
Advisors
Zak, Purdue University.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.