论文信息 - Fault-tolerance in Iterative Learning Neural Networks

Fault-tolerance in Iterative Learning Neural Networks

The content-addressability of patterns stored in damaged Ising-spin neural network models is studied. After learning with the iterative Edinburgh-group algorithm, damage is introduced by independently cutting bonds with a probability p. Numerical results from simulations involving systems with up to 512 neurons show the retrieval properties of the resulting networks. Below saturation the networks are capable of re-learning, thereby adapting to the damage.

Norman Hendrich

[1] B. Forrest. Content-addressability and learning in neural networks , 1988 .

[2] Thomas B. Kepler,et al. Domains of attraction in neural networks , 1988 .

[3] Eva Koscielny-Bunde,et al. Effect of damage in neural networks , 1990 .

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

[5] E. Gardner,et al. Maximum Storage Capacity in Neural Networks , 1987 .

[6] J J Hopfield,et al. Neural networks and physical systems with emergent collective computational abilities. , 1982, Proceedings of the National Academy of Sciences of the United States of America.

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

[8] Sompolinsky,et al. Spin-glass models of neural networks. , 1985, Physical review. A, General physics.