On some new associative properties of neural networks with an asymmetric Hebbian rule

Recurrent associative neural networks with a new model of connection weights, which we call an asymmetric Hebbian rule, are discussed. By using this model, we extend associative functions of neural networks. Networks with the model has associative functions such as conditional association, robustness to unmemorized patterns and cooperative association are discussed. We also propose a model of multiple association modes realized by using the properties of the same model. Multiple levels of hidden memories can be embedded and they are activated or deactivated by controlling threshold values.