Accession Number : ADA329960
Title : Biomorphic Networks for ATR and Higher-Level Processing
Descriptive Note : Quarterly rept. no. 11, 1 Jul-1 Oct 97
Corporate Author : MOORE SCHOOL OF ELECTRICAL ENGINEERING PHILADELPHIA PA
Personal Author(s) : Farhat, Nabil H.
PDF Url : ADA329960
Report Date : OCT 1997
Pagination or Media Count : 16
Abstract : There is considerable evidence that the basic functional unit for higher-level processing in the cortex is the netlet or neuronal assembly/(pool or group). This includes extensive analytical and modeling work of netlets carried out independently by several groups. Nearly all this work points to the possibility that netlet dynamics, namely its evolution in time, can be described by the discrete time evolution of the activity A(n), which is the percentage of neurons active at any instant of time. Plots of A(n+1) vs. A(n) obtained under a range of circumstances and assumptions are found to invariably resemble a distorted version of the quadratic or logistic map. The Logistic map is a nonlinear iterative map on the unit interval that exhibits complex orbits depending on the value of nonlinearity (control or bifurcation) parameter of the map. The similarity between the netlet's return map A(n+1) vs. A(n) and that of the logistic map has also been noted by Harth who also mentions that complex and unpredictable sequences A(n) were observed in some of their early simulations of netlets suggesting that certain regions of the netlet's parameter space may have led to observation of chaos in addition to the periodic and fixed point modalities they usually observed. In light of this evidence we have conjectured that cortical networks can be modeled and numerically studied in an efficient way by means of coupled populations of logistic processing elements. To test this conjecture we have studied the dynamics of such a network when it is subjected to external stimulus patterns that change in time.
Descriptors : *NEURAL NETS, COUPLING(INTERACTION), TIME DEPENDENCE, TARGET RECOGNITION, NONLINEAR SYSTEMS, PATTERN RECOGNITION.
Subject Categories : Bionics
Distribution Statement : APPROVED FOR PUBLIC RELEASE