Epsilon capacity of neural networks
AIP Conference Proceedings 151 on Neural Networks for Computing
The capacity of the Hopfield associative memory
IEEE Transactions on Information Theory
The Science of Making ERORS: What Error Tolerance Implies for Capacity in Neural Networks
IEEE Transactions on Knowledge and Data Engineering
| Hi-index | 0.14 |
The authors investigate the computing capabilities of formal McCulloch-Pitts neurons when errors are permitted in decisions. They assume that m decisions are to be made on a randomly specified m set of points in n space and that an error tolerance of epsilon m decision errors is allowed, with 0or= epsilon 1/2. The authors are interested in how large an m can be selected such that the neuron makes reliable decisions within the prescribed error tolerance. Formal results for two protocols for error-tolerance-a random error protocol and an exhaustive error protocol-are obtained. The results demonstrate that a formal neuron has a computational capacity that is linear in n and that this rate of capacity growth persists even when errors are tolerated in the decisions.