On physical models of neural computation and their analog VLSI implementation

Examines computation in a framework where the problem is essentially that of extracting a signal from noise, filtering (selective amplification) or estimation. The discussion is relevant to computational tasks in sensory communication, such as vision, speech and natural language processing. We consider "real" systems, both natural (neural systems) and human engineered (silicon integrated circuits), where information processing takes the form of an irreversible physical process. We argue, and demonstrate experimentally, that it is possible to see the emergence of truly complex processing structures that are commensurate with the physical properties of the computational substrate and are therefore energetically efficient.<
>