Learning activation rules for associative networks

In neural networks involving associative recall, information is sometimes encoded using a local representation and predetermined, fired connection weights. Standard learning algorithms that alter weights are not useful in this situation as it is the activation rule that needs to be learned. To address this problem, we recently derived a supervised learning rule where a network is trained by changing the activation rule. We describe here the first successful use of this new learning rule on a real-world application modeling print-to-sound transformation