Some novel real/complex-valued neural network models

Traditional models of neurons are based on the assumption that a synapse is a lumped element represented by a scalar synaptic weight. But to faithfully model biological neurons, synapse is considered as a linear filter. Thus, a new model of continuous time neuron is discussed. It is described how such model leads to interesting neural networks. Also continuous time, complex-valued neuron is discussed. It is also described, how a synapse can be modeled as an FIR filter. Such a model of neuron leads to practically useful neural networks. A novel, continuous time associative memory is proposed. An approach to resolve the convergence of state of such an associative memory is discussed. Various interesting generalizations of neural networks are described.