Predictive coding networks for temporal prediction
One of the key problems the brain faces is inferring the state of the world from a sequence of dynamically changing stimuli, and it is not yet clear how the sensory system achieves this task. A well-established computational framework for describing perceptual processes in the brain is provided by t...
Main Authors: | , , , , |
---|---|
Format: | Journal article |
Language: | English |
Published: |
Public Library of Science
2024
|
_version_ | 1811139257458229248 |
---|---|
author | Millidge, B Tang, M Osanlouy, M Harper, NS Bogacz, R |
author_facet | Millidge, B Tang, M Osanlouy, M Harper, NS Bogacz, R |
author_sort | Millidge, B |
collection | OXFORD |
description | One of the key problems the brain faces is inferring the state of the world from a sequence of dynamically changing stimuli, and it is not yet clear how the sensory system achieves this task. A well-established computational framework for describing perceptual processes in the brain is provided by the theory of predictive coding. Although the original proposals of predictive coding have discussed temporal prediction, later work developing this theory mostly focused on static stimuli, and key questions on neural implementation and computational properties of temporal predictive coding networks remain open. Here, we address these questions and present a formulation of the temporal predictive coding model that can be naturally implemented in recurrent networks, in which activity dynamics rely only on local inputs to the neurons, and learning only utilises local Hebbian plasticity. Additionally, we show that temporal predictive coding networks can approximate the performance of the Kalman filter in predicting behaviour of linear systems, and behave as a variant of a Kalman filter which does not track its own subjective posterior variance. Importantly, temporal predictive coding networks can achieve similar accuracy as the Kalman filter without performing complex mathematical operations, but just employing simple computations that can be implemented by biological networks. Moreover, when trained with natural dynamic inputs, we found that temporal predictive coding can produce Gabor-like, motion-sensitive receptive fields resembling those observed in real neurons in visual areas. In addition, we demonstrate how the model can be effectively generalized to nonlinear systems. Overall, models presented in this paper show how biologically plausible circuits can predict future stimuli and may guide research on understanding specific neural circuits in brain areas involved in temporal prediction. |
first_indexed | 2024-04-23T08:26:51Z |
format | Journal article |
id | oxford-uuid:3231034f-7c91-4192-b545-046658ee8bf0 |
institution | University of Oxford |
language | English |
last_indexed | 2024-09-25T04:03:13Z |
publishDate | 2024 |
publisher | Public Library of Science |
record_format | dspace |
spelling | oxford-uuid:3231034f-7c91-4192-b545-046658ee8bf02024-05-10T17:01:00ZPredictive coding networks for temporal predictionJournal articlehttp://purl.org/coar/resource_type/c_dcae04bcuuid:3231034f-7c91-4192-b545-046658ee8bf0EnglishSymplectic ElementsPublic Library of Science2024Millidge, BTang, MOsanlouy, MHarper, NSBogacz, ROne of the key problems the brain faces is inferring the state of the world from a sequence of dynamically changing stimuli, and it is not yet clear how the sensory system achieves this task. A well-established computational framework for describing perceptual processes in the brain is provided by the theory of predictive coding. Although the original proposals of predictive coding have discussed temporal prediction, later work developing this theory mostly focused on static stimuli, and key questions on neural implementation and computational properties of temporal predictive coding networks remain open. Here, we address these questions and present a formulation of the temporal predictive coding model that can be naturally implemented in recurrent networks, in which activity dynamics rely only on local inputs to the neurons, and learning only utilises local Hebbian plasticity. Additionally, we show that temporal predictive coding networks can approximate the performance of the Kalman filter in predicting behaviour of linear systems, and behave as a variant of a Kalman filter which does not track its own subjective posterior variance. Importantly, temporal predictive coding networks can achieve similar accuracy as the Kalman filter without performing complex mathematical operations, but just employing simple computations that can be implemented by biological networks. Moreover, when trained with natural dynamic inputs, we found that temporal predictive coding can produce Gabor-like, motion-sensitive receptive fields resembling those observed in real neurons in visual areas. In addition, we demonstrate how the model can be effectively generalized to nonlinear systems. Overall, models presented in this paper show how biologically plausible circuits can predict future stimuli and may guide research on understanding specific neural circuits in brain areas involved in temporal prediction. |
spellingShingle | Millidge, B Tang, M Osanlouy, M Harper, NS Bogacz, R Predictive coding networks for temporal prediction |
title | Predictive coding networks for temporal prediction |
title_full | Predictive coding networks for temporal prediction |
title_fullStr | Predictive coding networks for temporal prediction |
title_full_unstemmed | Predictive coding networks for temporal prediction |
title_short | Predictive coding networks for temporal prediction |
title_sort | predictive coding networks for temporal prediction |
work_keys_str_mv | AT millidgeb predictivecodingnetworksfortemporalprediction AT tangm predictivecodingnetworksfortemporalprediction AT osanlouym predictivecodingnetworksfortemporalprediction AT harperns predictivecodingnetworksfortemporalprediction AT bogaczr predictivecodingnetworksfortemporalprediction |