Bidirectional associative memory

Bidirectional associative memory, bidirectional associative memory english (short: BAM ) is a class of artificial neural networks and can be regarded as generalized Hopfield network. BAM is one of the group of back -coupled neural network.

Structure

A BAM network consists of an input layer and an output layer of n of m artificial neurons, both layers are connected to each other in both directions, wherein the weights are balanced. This leads to an mxn matrix of the weights, which are directed from to. The weights from to match the transposed matrix.

Training phase

In the training phase, the network learns an n-dimensional vector x to be associated with an m- dimensional vector y. These two vectors are applied to the input layer and output layer, and the weight matrix can be computed in a learning step. To this end, the following applies:

K = { 1, ..., l} for l vector pairs

Finally, all weight matrices are added to the resulting weight matrix.

Patterns restore

In a recall a noisy input vector is applied to and is allowed to expect the network simple, ie Neurons of the output layer calculate about her new condition and give it over again to continue. Then the process starts over, as long as until the steady decline in power of the network reaches a local minimum. Now the associated output vector can be removed.

And