Title :
The modified unlearning procedure for enhancing storage capacity in Hopfield network
Author :
Plakhov, A.Yu. ; Semenov, S.A.
Author_Institution :
Inst. of Phys. & Technol., Moscow, Russia
Abstract :
The authors propose a solvable iterative algorithm of unlearning type for self-correction of Hebbian connectivity. It is shown that, for almost all unlearning sequences, the resulting connection matrix asymptotically converges to the projector matrix. The corresponding convergence rate is calculated and confirmed with numerical simulations
Keywords :
Hebbian learning; Hopfield neural nets; convergence; iterative methods; matrix algebra; Hebbian connectivity; Hopfield network; connection matrix; convergence rate; neural nets; solvable iterative algorithm; storage capacity; unlearning procedure; Algorithm design and analysis; Associative memory; Convergence of numerical methods; Intelligent networks; Iterative algorithms; Neural networks; Neurons; Numerical simulation; Physics; Symmetric matrices;
Conference_Titel :
Neuroinformatics and Neurocomputers, 1992., RNNS/IEEE Symposium on
Conference_Location :
Rostov-on-Don
Print_ISBN :
0-7803-0809-3
DOI :
10.1109/RNNS.1992.268563
