论文信息 - Statistical properties of the neuron active potential at stable states in Hopfield model

Statistical properties of the neuron active potential at stable states in Hopfield model

This article investigates the statistical properties of Hopfield model at its spurious stable states (SSS) and nonspurious stable states(N-SSS). It has been shown that, in Hopfield model, some stable states are learned patterns and the others are not. A stable state is spurious if it is not a learned pattern or it is nonspurious. In this article, by means of statistical neurodynamics, the probability distributions of the neuron active potential (NAP) at N-SSS and SSS are given, and the first moments and second moments of the absolute NAP are estimated. In Hopfield model, the NAP at SSS and N-SSS have almost the same distribution curves. The difference between the first moments at SSS and N-SSS is small. But the difference between the second moments is large. The statistical results of computer simulation on the NAP are consistent with the analysis results. The results can be used to improve the convergent property of Hopfield model and to design a parallel discrimination method to detect whether a recalling output is SSS or not. This method is real-time and need not refer to the learned patterns.

Yasumitsu Miyazaki | Xin-Min Huang

[1] D Psaltis,et al. Optical implementation of the Hopfield model. , 1985, Applied optics.

[2] E. Gardner. Structure of metastable states in the Hopfield model , 1986 .

[3] J J Hopfield,et al. Neurons with graded response have collective computational properties like those of two-state neurons. , 1984, Proceedings of the National Academy of Sciences of the United States of America.

[4] Yasumitsu Miyazaki,et al. Connected Associative Memory Neural Network with Dynamical Threshold Function , 1992 .

[5] John J. Hopfield,et al. Neural networks and physical systems with emergent collective computational abilities , 1999 .

[6] N. Farhat. Optoelectronic neural networks and learning machines , 1989, IEEE Circuits and Devices Magazine.

[7] Sompolinsky,et al. Storing infinite numbers of patterns in a spin-glass model of neural networks. , 1985, Physical review letters.

[8] Mahesan Niranjan,et al. A theoretical investigation into the performance of the Hopfield model , 1990, IEEE Trans. Neural Networks.

[9] Shun-ichi Amari,et al. Statistical neurodynamics of associative memory , 1988, Neural Networks.

[10] Shun-ichi Amari,et al. Mathematical foundations of neurocomputing , 1990, Proc. IEEE.

[11] M. Shiino,et al. Self-consistent signal-to-noise analysis and its application to analogue neural networks with asymmetric connections , 1992 .