Generalization learning in a perceptron with binary synapses (Q1040712)

There is a short discussion on the performance of the so called SBPI (stochastic belief propagation inspired)-algorithm applied to a perceptron with bipolar weights, inputs, outputs and zero threshold. A comparison to the CP (clipped perceptron algorithm) is done. One derives a set of partial differential equations which give an insight into the learning process under SBPI as well as CP learning rules.