IRUS Total

Revealing the (predictive) code of top-down signals in the brain

File Description SizeFormat 
Martin-C-2020-PhD-Thesis.pdfThesis97.45 MBAdobe PDFView/Open
Title: Revealing the (predictive) code of top-down signals in the brain
Authors: Martin, Christian-David
Item Type: Thesis or dissertation
Abstract: Reciprocal connections are common in the brain, yet little is known about their functional role. Top-down connections, in particular, remain functionally obscure in both neuroscience and in the nascent field of deep learning. On the theoretical side, predictive coding has been put forward as a framework that assigns specific roles to top-down and bottom-up connections in sensory information processing. It remains unclear, however, if and how the brain implements this predictive code. This work examined top-down signals in the auditory cortex and in the corticostriatal system in macaques in order to validate the claims put forward by predictive coding. This theory suggests there are imbalances in message passing up and down the cortical hierarchy; these imbalances imply cross-frequency couplings should predominate top-down. It is unknown whether these asymmetries are expressed in cross-frequency interactions in the brain. This work examined cross-frequency interactions across four sectors of the macaque auditory cortex. Predictive coding also applies in decision making, where it allows for action selection based on predicted reward (or value). This is more commonly known as reinforcement learning (RL) and is supported by the fronto-striatal systems in the brain. The computational mechanisms that drive learning in this system are unknown, however. This work drew on a recurrent neural network (RNN) model of the dlPFC-dSTR circuit in the brain together with recordings from macaques from the same regions to answer this question. Altogether, the findings are largely consistent with the predictive coding framework.
Content Version: Open Access
Issue Date: Nov-2019
Date Awarded: Mar-2020
URI: http://hdl.handle.net/10044/1/80155
DOI: https://doi.org/10.25560/80155
Copyright Statement: Creative Commons Attribution NonCommercial Licence
Supervisor: Schultz, Simon
Sponsor/Funder: Wellcome Trust (London, England)
National Institutes of Health
Funder's Grant Number: Wellcome Trust 4-Year Fellowship
Department: Bioengineering
Publisher: Imperial College London
Qualification Level: Doctoral
Qualification Name: Doctor of Philosophy (PhD)
Appears in Collections:Bioengineering PhD theses