Convergence analysis of a digital diffusion network

We propose a numerical procedure for approximating an analog diffusion network. The main idea, is to take advantage of the “separable” feature (of the noise) of the diffusion machine and use parallel processing method to develop recursive algorithms. In addition to the decreasing step size algorithm, constant step algorithms and procedures with periodic restarts are suggested. By means of weak convergence methods, the convergence of the algorithms is established. The algorithms may be useful for many large-scale optimization problems, including image segmentation