Neural networks mediating linearizable dynamic redundant sensori-motor reflexes characterized by minimum of Hermitian norm

In general both sensory and motor systems of natural reflexes are redundant. The hypothesis is put forward that the neural network mediating a given linearizable dynamic redundant reflex is characterized by minimizing the Hermitian norm of the neural transfer function matrix for all frequencies of the reflex operating range. Such a neural network resolves the problem of redundancies, minimizes effects of noise added to sensory signals, and minimizes motor effort. Furthermore, it can be found or approximated by a suitable combination of learning and forgetting rules in adaptive neural nets.<>