Neuromorphic circuit implementation of isotropic sequence order learning

The isotropic sequence order (ISO) learning is an improved version of differential Hebbian learning algorithm. It uses a switch to turn on or off the learning at appropriate time instants to minimise the level of inherent instability possessed by the classical Hebbian learning. In this paper we present a novel analog very large scale integrated circuit (aVLSI) model to implement ISO learning. The circuit includes an integrate-and-fire (IF) neuron, two synapses and associated low-pass filters. By adjusting a set of input biases, the Cadence simulation results show that the predictive pathway of the circuit can effectively learn the inputs of the reflexive pathway in a fast and stable process.