Critical storage capacity of the J = ± 1 neural network
作者： W Krauth , M Opper
DOI: 10.1088/0305-4470/22/11/012
关键词: Artificial neural network 、 Mathematics 、 Gray code 、 Algorithm 、 Applied mathematics
摘要: For neural networks in which the coupling Jij are allowed to take on values = 1 or -1, authors determine numerically critical storage capacity for random unbiased patterns as a function of stability. They use an exact enumeration scheme based Gray code and continuous distribution control finite-size effects. Results presented N ≤ 25; they indicate optimal αc ≈ 0.82 (N → ∞).
harvard.edu
sci-hub.se 参考文章(5)
David Sherrington, Scott Kirkpatrick, Solvable Model of a Spin-Glass Physical Review Letters. ,vol. 35, pp. 1792- 1796 ,(1975) , 10.1103/PHYSREVLETT.35.1792
E Gardner, The space of interactions in neural network models Journal of Physics A. ,vol. 21, pp. 257- 270 ,(1988) , 10.1088/0305-4470/21/1/030
W Krauth, M Mezard, Learning algorithms with optimal stability in neural networks Journal of Physics A. ,vol. 20, ,(1987) , 10.1088/0305-4470/20/11/013
E Gardner, Maximum Storage Capacity in Neural Networks EPL. ,vol. 4, pp. 481- 485 ,(1987) , 10.1209/0295-5075/4/4/016
Santosh S. Venkatesh, Epsilon capacity of neural networks Neural Networks for Computing. ,vol. 151, pp. 440- 445 ,(2008) , 10.1063/1.36217