Hebbian Learning Rule

Hebbian learning adjusts the network's weights such that its output reflects its familiarity with an input. The more probable an input, the larger the output will become (on average). Unfortunately, plain Hebbian learning continually strengthens its weights without bound (unless the input data is properly normalized). There are only a few applications for plain Hebbian learning, however, almost every unsupervised and competitive learning procedures can be considered Hebbian in nature.

NeuroSolutions also supports the forced Hebbian learning rule. In forced Hebbian, the output of the component is substituted by a desired response for the purpose of weight update. This type of learning has been applied to heteroassociation.

NeuroSolutions Home Page