Generating normative predictions with a variable-length rate code
- S. Thomas Christie, Cognitive Science, University of Minnesota, Minneapolis, Minnesota, United States
- Paul Schrater, University of Minnesota, Twin Cities, Minneapolis, Minnesota, United States
AbstractCognitive science is an archipelago of concepts and models, with cross-pollination between topics of interest often prohibited by incompatible approaches. Despite this, behavioral performance universally depends on information transmission between brain regions and is limited by physical and biological constraints. These constraints can be formalized as information theoretic constraints on transmission, which provide normative predictions across a surprising range of cognitive domains. To illustrate this, we describe a simple variable-length rate coding model built with Poisson processes, Bayesian inference, and an entropy-based decision threshold. This model replicates features of human task performance and provides a principled connection between a high-level normative framework and neural rate codes. We thereby integrate several disjoint ideas in cognitive science by translating plausible constraints into information theoretic terms. Such efforts to translate concepts, paradigms and models into common theoretical languages are essential for synthesizing our rich but fragmented understanding of cognitive systems.