Novelty detection based on relaxation time of a network of integrate-and-fire neurons

We propose a neural network model inspired from a simulated cortex model. Also, a new paradigm for pattern recognition by oscillatory neural networks is presented. The relaxation time of the oscillatory networks is used as a criterion for novelty detection. We compare the proposed neural network with Hopfield and backpropagation networks for a noisy digit recognition task. It is shown that the proposed network is more robust. This work could be a possible bridge between nonlinear dynamical systems and cognitive processes