论文信息 - A dilution algorithm for neural networks

A dilution algorithm for neural networks

A dilution algorithm to enlarge the storage capacity per synapse alpha eff of neural networks is proposed. The algorithm is a hybrid method, where Hebb's rule is used to select a fraction of couplings to be removed. Afterwards the perceptron of optimal stability for the remaining couplings is learned. The authors present an analytical calculation and the results of the numerical simulations. In comparison with the fully connected or the randomly diluted perceptron, the effective storage capacity alpha eff is remarkably enlarged.

P. Kuhlmann | R. Garces | H. Eissfeller

[1] Thomas M. Cover,et al. Geometrical and Statistical Properties of Systems of Linear Inequalities with Applications in Pattern Recognition , 1965, IEEE Trans. Electron. Comput..

[2] R. Palmer,et al. The replica method and solvable spin glass model , 1979 .

[3] W. Krauth,et al. Learning algorithms with optimal stability in neural networks , 1987 .

[4] P. Leath,et al. The failure distribution in percolation models of breakdown , 1987 .

[5] E. Gardner,et al. Optimal storage properties of neural network models , 1988 .

[6] Opper. Learning times of neural networks: Exact solution for a PERCEPTRON algorithm. , 1988, Physical review. A, General physics.

[7] E. Gardner. The space of interactions in neural network models , 1988 .

[8] W. Kinzel,et al. Layered neural networks , 1989 .

[9] Michael Biehl,et al. The AdaTron: An Adaptive Perceptron Algorithm , 1989 .

[10] A. Komoda,et al. Quenched versus annealed dilution in neural networks , 1990 .

[11] Karl E. Kürten,et al. Quasi-optimized memorization and retrieval dynamics in sparsely connected neural network models , 1990 .

[12] K Y M Wong,et al. The Principle of Adaptation and Dilution Robustness in Neural Networks , 1991 .