Predictive coding and the Bayesian brain: Intractability hurdles that are yet to be overcome

Introduction An influential conjecture in neuroscience states that a brain organized according to principles of hierarchical predictive coding (HPC) can effectively and efficiently perform genuine Bayesian inferences, providing a neurally plausible account of how Bayesian models of cogntion may be physically implemented in the brain. However, each key subcomputation invoked in HPC potentially hides a computationally intractable problem. We identify ways in which computational modelers may or may not overcome these ‘intractability hurdles.’ Figure 1. A schematic depiction of the way in which the top-down chain (left) and the bottom-up chain (right) in the HPC framework interact. The red arrows are labeled by the key computational transformations PREDICTION, ERROR COMPUTATION and HYPOTHESIS UPDATING.