Dual-matching as a problem solved by neurons

This paper extends and generalizes a previously described information-theoretic formulation of single-neuron computation. This framework can be understood through a comparison of the stated goals of neural computation with the two fundamental problems of information theory; those of source and channel coding. Practically, this requires the evaluation of two Lagrangians that together allow a neuron to solve the dual-matching problem in information theory whereby system throughput capacity is maximized. The resulting model is suggestive of the brain's computational role as an intelligent controller that likewise matches the rate it acquires information to the rate it makes decisions and selects actions in service to its prevailing computational goals.