Neural computing for data recovery

This paper presents binary threshold networks to recover regularized LS estimates from degraded images. The binary networks consist of nonlinear processing elements configured to optimize the objective function. The optimization takes place at the bit-level on partitions of these networks. Update procedures and algorithms are outlined. In addition, alternate objective criteria are expressed in partitions to recover the LS estimate. Regularization is introduced to control the rate of convergence of the LS estimate.