External fields in attractor neural networks with different learning rules

The effects of external fields on the retrieval properties of highly dilute attractor neural networks with general classes of learning rules are examined. It can be shown that external fields increase basins of attraction making even perfect retrieval possible for relatively high loading. The application of different classes of noise distributions on the stimulus field indicates that certain first-order transitions occurring are peculiarities of the type of noise. Optimally adapted networks in the presence of external fields extend the critical loading above which perfect retrieval is impossible. In the presence of external fields in the high-temperature regime, Hebb networks also retrieve better than rules with more optimal performances at low temperatures.