Title :
A quantitative analysis of a variety of sparsely encoded memories
Author :
Huang, Yo-Ping ; Gustafson, Donald
Author_Institution :
Dept. of Electr. Eng., Texas Tech. Univ., Lubbock, TX, USA
Abstract :
Three different sparsely encoded associative memories (SEAMs) are quantitatively analyzed. The first two models are shown to have performance inferior to that of nonsparse systems due to stored-pattern mutual interference. Assuming that each neuron is working close to its threshold and using dynamic thresholding selection, the binary SEAM performance can be improved. The simulation results are consistent with the theoretical work. The sensitivity problem that arises in choosing the threshold range is also explored
Keywords :
content-addressable storage; neural nets; associative memories; dynamic thresholding; quantitative analysis; sensitivity problem; sparsely encoded memories; stored-pattern mutual interference; threshold range; Analytical models; Associative memory; Hamming distance; Hebbian theory; Interference; Neural networks; Neurons; Probes; Signal to noise ratio;
Conference_Titel :
Circuits and Systems, 1992. ISCAS '92. Proceedings., 1992 IEEE International Symposium on
Conference_Location :
San Diego, CA
Print_ISBN :
0-7803-0593-0
DOI :
10.1109/ISCAS.1992.229942
