Sensorimotor learning and information processing by Bayesian internal models

Fundamental to effective brain-machine interface and neuroprosthesis designs is an understanding of how sensory and motor information are encoded, integrated and adapted by the nervous system. Special session "Neural Information Processing by Bayesian and Internal Models" expounds two current theories of sensorimotor integration which posit that neural information may be encoded centrally as an "internal model" of the environment or as a stochastic state-space model that modulates the activity of spiking neurons. Underlying both theories is a possible role for Bayes' rule--as suggested by the recent findings that the brain may employ Bayesian internal models during certain types of sensorimotor learning in order to optimize task-specific performance and that the emergent activity of certain neural ensembles may be modeled as joint Bayesian point processes. These emerging concepts of neural signal processing have far-reaching implications in applications from rehabilitation engineering to artificial intelligence.