Predictive learning in rate-coded neuronal networks: a theoretical approach towards classical conditioning

A novel approach for learning of temporally extended, continuous signals is developed within the framework of rate-coded neurons. A new temporal Hebb-like learning rule is devised which utilises the predictive capabilities of bandpass-filtered signals by using the derivative of the output to modify the weights. The initial development of the weights is calculated analytically by applying signal theory and simulation results that are shown to demonstrate the performance of this approach. In addition, we show that only few units suffice to process multiple inputs with long temporal delays.