More than 150 years have passed since Helmholtz first described perception as a process of unconscious inference about the causes of sensations. His ideas have since inspired a wealth of literature investigating the mechanisms underlying these inferences. In recent years, much of this work has converged on the notion that the brain is a hierarchical generative model of its environment that predicts sensations and updates itself based on prediction errors. Here, we build a case for modeling tinnitus from this perspective, i.e. predictive coding. We emphasize two key claims: (1) acute tinnitus reflects an increase in sensory precision in related frequency channels and (2) chronic tinnitus reflects a change in the brain's default prediction. We further discuss specific neural biomarkers that would constitute evidence for or against these claims. Finally, we explore the implications of our model for clinical intervention strategies. We conclude that predictive coding offers the basis for a unifying theory of cognitive neuroscience, which we demonstrate with several examples linking tinnitus to other lines of brain research.
Keywords: Active inference; Bayes; Belief; Inference; Learning; Perception; Phantom perception; Prediction error; Predictive coding; Tinnitus; Unsupervised learning.
Copyright © 2019 Elsevier Ltd. All rights reserved.