Title :
Dynamic and static numerical modeling of actual gradient-type neural networks
Author :
Zurada, Jacek M. ; Kang, M.J.
Author_Institution :
Dept. of Electr. Eng., Louisville Univ., KY, USA
Abstract :
A discussion is presented of the performance modeling of gradient-type networks utilizing continuous activation functions, finite input resistance of neurons, and other parasitic components within the neural system. Both time-domain performance and static numerical modeling of the networks are characterized and compared. It is shown that the relaxation algorithm may not guarantee the numerical convergence while the vector field method does
Keywords :
convergence of numerical methods; modelling; neural nets; numerical methods; relaxation theory; time-domain analysis; continuous activation functions; finite input resistance; gradient-type; neural networks; neurons; numerical convergence; parasitic components; performance modeling; relaxation algorithm; static numerical modeling; time-domain performance; vector field method; Capacitance; Convergence; Electric resistance; Equations; Neural networks; Neurofeedback; Neurons; Numerical models; Time domain analysis; Voltage;
Conference_Titel :
Circuits and Systems, 1991., IEEE International Sympoisum on
Print_ISBN :
0-7803-0050-5
DOI :
10.1109/ISCAS.1991.176031
