Talk:Predictive coding

Untitled
Previous abstract: Predictive coding models suggest that the brain is constantly generating and updating hypotheses that predict sensory input at varying levels of abstraction. This framework is in contrast to the view that the brain integrates exteroceptive information through a predominantly feedforward process, with feedback connections playing a more minor role in cortical processing. Bodysurfinyon (talk) 02:42, 27 January 2019 (UTC)

Second bold item in lead
It is difficult to construe the second bold item in lead as a second definitional topic, as the phrase is simply too generic.

In each region, the model being propagated is compared to the sensory input and if they do not match, a Prediction Error is sent back up the network and the model is revised.

The emphasis is perhaps useful in conveying how central prediction error is within this theory, but the clash with Wikipedia convention seems too large to justify for a relatively small fish. &mdash; MaxEnt 16:43, 8 June 2019 (UTC)

Diagram
I made a diagram based on various diagrams. For more info, open the .svg as a test file. I'd love to hear suggestions for improvement. Bodysurfinyon (talk) 06:29, 14 June 2021 (UTC)


 * Thanks for doing this! I would suggest to simplify the diagram a bit, to make it easier to understand. I found a particularly simple diagram here . Maybe it could be changed into this direction. Apoptheosis (talk) 10:21, 29 October 2023 (UTC)

Suggested changes to improve the article
It seems that this article has a few pending issues that should be addressed. In general, some sections could be improved in their clarity, in the selection of work they discuss, and also by adding relevant connections that are missing. I felt that it would be a good idea to compile a list of concrete problems first. Maybe I will find time to implement them myself, but anyone else is also welcome to address them!

Apoptheosis (talk) 10:14, 29 October 2023 (UTC)
 * General framework
 * The language in the first paragraph is very technical. It might be a good idea to have an easily understandable high level description of the framework, followed by a more technical discussion that also encompasses the mathematical formalism.
 * "This is also the reason for what is nowadays called confirmation bias ..." This is stated as a fact, which is misleading, since it is a conjecture. I think this is also not relevant at this point but should rather be mentioned in a specialized section (e.g., in "Applications").
 * There are important connections missing, for example, sparse coding and related models should be mentioned, which are the non-hierarchical (no prediction) equivalent. For precision weighting, Kalman filter should be mentioned, etc.
 * Neural theory in predictive coding
 * It is to be contested that Predictive coding sensu Rao and Ballard 1999, despite its rich historical pedigree and the numerous testable preditions to be derived form it, is a *theory* in the accepted meaning of the word. — Preceding unsigned comment added by 141.83.70.133 (talk) 08:50, 21 November 2023 (UTC)
 * The paragraph about the testability of the theory is partly redundant with that in the "Challenges" section. Maybe these can be combined. I suggest separating "neural theory", "applications" and "empirical evidence" into individual sections.
 * "More generally, however, what seems to be required by the theory are (at least) two types of neurons" seems to be outdated. There are by now a few models that suggest other implementations of predictive coding, mostly based on dendritic computations (e.g.,  ).


 * Thanks for the improvements. The connections between predictive coding and cortical columns is important and I plan to restore it. 03:57, 9 July 2024 (UTC) Bodysurfinyon (talk) 03:57, 9 July 2024 (UTC)