[Not loaded yet]
There might be a bit of misconception here. What the paper very convincingly shows is that visual cortex does not compute global oddball prediction errors and does not receive any top-down predictions that could be used to compute such prediction errors.
Jul 14, 2025 15:44The construction of the interpretation, however, appears to be built on the assumption that if something is predictable in principle it will be predicted by the brain. I would look to behavioral relevance to make guesses about what the brain predicts.
Stimulus surround (i.e. statistics of natural stimuli) and visuomotor coupling are much (by orders of magnitude) stronger predictors of visual input (that require no additional pretraining…). No animal prior to the advent of modern neuroscience has ever seen repeating sequences of images.
To claim that sensory responses of visual cortical neurons are not prediction errors (based on the data the authors have) is unwarranted. There is ample evidence for stimulus surround and visuomotor prediction errors.
The title of this paper should be «sensory responses in visual cortex cannot be explained as stimulus history prediction errors» – I am confident the authors will rectify this in the next version of the manuscript.
While I am quite convinced that predictive processing, even its more reasonable non-hierarchical form, is wrong, there is very good evidence of prediction error responses.
p.s. @tyrellturing.bsky.social what would be the alternative to PE neurons for “comparing predictions to sensory input” – prediction error dendrites, or ensembles?
