Project

Back to overview

Predictive coding, predictive routing and cortical oscillations in speech processing

English title Predictive coding, predictive routing and cortical oscillations in speech processing
Applicant Giraud Mamessier Anne-Lise
Number 182855
Funding scheme Project funding
Research institution Département des neurosciences fondamentales Faculté de Médecine Université de Genève
Institution of higher education University of Geneva - GE
Main discipline Neurophysiology and Brain Research
Start/End 01.01.2019 - 30.06.2022
Approved amount 857'766.00
Show all

Keywords (3)

Predictive coding; Neural oscillations; Speech

Lay Summary (French)

Lead
Codage prédictif, "routage" prédictif et oscillation neurales dans le traitement de la parole
Lay summary
Dans la perception de la parole comme dans de nombreux autres domaines de la cognition, les oscillations neurales sont supposées assumer des fonctions différentes qu'il demeure difficile de relier de façon causale à des opérations computationnelles et des algorithmes de traitement de l'information précis. Récemment, une propriété importante a été révélée chez l'homme et le singe: le fait que les informations ascendantes et descendantes dans le cerveau sont associées à des familles de fréquence oscillatoires différentes, qui s'expriment de façon alternées, avec des fréquences plus élevées pour l'information ascendante que pour l'information descendante. Ces propriétés offrent une opportunité particulièrement intéressante pour relier la notion d'oscillations neurales avec la théorie du codage prédictif, dans un nouveau cadre théorique que nous appelons "Oscillation-Based Predictive Routing" ou OPR. Le but de ce projet est de conceptualiser la perception de la parole comme un processus d'analyse-par-synthèse discontinu, dans lequel les notion d'oscillation descendante et ascendante sont associées aux notions de prediction et d'erreur de prédiction. Pour cela nous combinons des expériences chez l'humain réalisées au moyen de l'électrophysiologie intracortical et de la magnétoencephalographie (MEG), avec la modélisation neuro-computationnelle.
Direct link to Lay Summary Last update: 03.12.2018

Responsible applicant and co-applicants

Employees

Project partner

Associated projects

Number Title Start Funding scheme
163040 Predictive coding and cortical oscillations in speech processing 01.01.2016 Project funding
163040 Predictive coding and cortical oscillations in speech processing 01.01.2016 Project funding

Abstract

Background : In the domain of speech perception as well as in many other domains of cognition, neural oscillations have been assigned various roles, which all remain difficult to causally relate to specific computational operations 1-3. Recently, however, an important feature was experimentally identified in humans and monkeys: the notion that feed-forward and feedback information operate in both different frequency ranges 4-7 and alternating temporal windows 8, with higher frequencies for feedforward than feedback information. This feature offers a particularly interesting opportunity to connect the notions of neural oscillations and oscillation coupling, with the theory of predicting coding 9,10, under a novel framework that we refer to as Oscillation-based Predictive Routing (OPR).Aims/hypotheses: The goal of our follow-up project is to conceptualize speech perception as a discontinuous analysis-by-synthesis 11-13 process, in which notions of up- and down-going neural oscillations are connected to those of predictions and prediction errors. In particular, we will seek to determine whether there could be a computational advantage for i) predictions and prediction errors to be conveyed through distinct frequency channels, with lower frequencies for the former than the latter, and ii) alternating temporal windows for feedforward prediction errors and feedback predictions. Under the OPR hypothesis, a slower oscillation for feedback than feedforward information transfer could reflect the integration of more information in both space and time when updating predictions than when encoding sensory stimuli,10 while alternating windows of information transfer could permit to un-mix vertical (hierarchical message passing) and horizontal (local prediction updating). This two-fold assumption will be central to our proposed work for the next three years.Methods: We will carry out experiments in humans using intracortical macro- and micro-electrode recordings (Utah array) and MEG, to characterize the involvement of distinct oscillation frequency ranges in parsing, encoding and predicting speech. In parallel, we will continue the modelling engaged during the first three years period, which aims at 1- understanding the mechanistic and computational function of oscillations in cortical speech processing, and 2- exploring their potential causal role in conveying and coding predictions and prediction errors. We will pursue biophysical modelling and expand to three hierarchical levels our previously developed oscillatory network model (Baroni et al., in prep.), to address the computational relevance of oscillations in speech segmentation, encoding and decoding. In parallel, we will continue exploring whether oscillations can be useful in a predictive coding model of speech processing that emphasizes function over biological plausibility (Hovsepyan et al., in prep.), by involving beta in addition to theta and gamma oscillations.Expected results: We expect the combination of these experimental and modelling data to challenge and possibly refine the OPR hypothesis. The expected results reach beyond the domain of speech processing; by assessing the usefulness of cortical oscillations in a generic organization mode of information processing, as predictive coding/routing, we hope to uncover generic principles relevant for all domains of cognitive neurosciences.
-