Super-Resolution and Signal Recovery Using Models of Neural Networks

Content addressable memory (CAM) based on models of neural networks [1], [2], offer capabilities that are useful in information processing, signal recovery, and pattern recognition. These include speed (stemming from their inherent parallelism and massive interconnectivity), robustness (stemming from their fault tolerant and soft-fail nature) and most significantly, relative to the subject matter of this meeting, their ability to recognize a partial input i.e., when the initializing input is an incomplete version of one of the stored entities. The latter two features are in fact synonymous with the realization of super-resolution where a function is recovered from a noisy or imperfect part. These attractive features are traceable to the highly nonlinear and iterative nature of feedback employed in such CAMs.