Hierarchical predictive coding

202309201035

(reproduced from Carhart-Harris2019)

Part of the Free energy principle which describes how priors are encoded by the brain.

Sensory input arrives at sensory epithelia and is compared with descending predictions. The ensuing prediction error (blue arrows) — typically characterized as neuronal populations of superficial Pyramidal neurons) — is passed forward into hierarchies to update expectations at higher levels. The posterior expectations (teal circles, deep Pyramidal neurons) generate predictions of the representations found in lower levels via descending prediction errors (teal arrows). Recurrent neural message passing (dynamics) attempts to minimize amplitude of prediction errors at each level of the hierarchy, furnishing the best explanation for sensory input at multiple levels of hierarchical abstraction (why this redundancy?)

  • This process depends on the precision (importance/salience) afforded to ascending prediction errors (surprise) and precision (felt confidence) of posterior beliefs.

Utilized to account for:


Free energy principle
REBUS


Q&A