An analog neural network solution to the inverse problem of `early taction´

The authors examine an application of analog neural networks to low-level processing of tactile sensory data. In analogy to the term early vision, the authors call the first level of processing required in tactile sensing early taction. The problem of deblurring or deconvolution of data provided by an array of tactile sensors that is also assumed to be corrupted by noise is addressed. It is noted that this inverse problem is ill posed and that the technique of regularization may be used to obtain solutions. The theory of nonlinear electrical networks is utilized to describe energy functions for a class of nonlinear networks and to show that the equilibrium states of the proposed network correspond to regularized solutions of the deblurring problem. An entropy regularizer is incorporated into the energy function of the network for the recovery of normal stress distributions. An integrated circuit prototype of the proposed network is discussed