Error Computation in the Diseased Brain using Machine Learning

Error Computation in the Diseased Brain using Machine Learning

By In PhD proposals 2019 On April 1, 2019

Project: Error Computation in the Diseased Brain using Machine Learning

Laboratory: Adaptive Decision-making Lab

Affiliation: Institute for Psychiatry and Neuroscience INSERM U1266
Address: Université Paris Descartes

LAB Director
Name: Dr. Abhishek Banerjee
Phone number: +41754173319

Name: Dr. Abhishek Banerjee (IPNP) and Dr. Stefano Palminteri (ENS)
Phone number: +41754173319

Subject Keywords: Reinforcement Learning, Neural Circuit, Synaptic Plasticity, Neurodevelopmental Disorders, In vivo imaging
Summary of lab’s interests: Our research in the ‘Adaptive Decision-making lab’ focuses on the study of neural circuits behind goal-directed flexible behaviour. We employ a range of molecular, genetic, opto-physiology tools combined with computational and mathematical framework to study how activity in specific neuronal cell-types, synapses and mesoscale circuits contributes to goal-directed learning in awake behaving animals and how dysfunctions of such mechanisms lead to neurological disorders.
Project summary: Mammalian brain during development and maturation learn from experience by flexibly updating behavioural strategies to optimise a desired outcome by earning reward or avoiding punishment. A key learning mechanism based on predictive coding hypothesis is to evaluate the mismatch between an internal model of the world and the actual behavioural interactions, and to update the internal model accordingly. Despite the fact that failure to flexibly update internal models are central to many brain disorders (including autism spectrum disorders; ASD), the underlying neural circuits are poorly understood at the synaptic and computational levels.

Inspired by reinforcement learning concepts, we propose a new theory-driven experimental approach to identify dysfunctions in predictive coding principles in the animal models ASD. Our central hypothesis is that in ASD, interactions and interplay of feedforward (FF) ‘learning-‘ and feedback (FB) ‘teaching signals’ between distinct higher and lower-order brain areas are compromised. We will test this hypothesis by measuring and modelling neural activity in distinct mouse cortical areas during goal-directed flexible behaviours, and by specifically manipulating FF and FB inputs using optogenetic tools. Our interdisciplinary collaboration with complementary expertise will lay the groundwork for developing computational hypotheses to study dysfunctions in mechanisms underlying flexible-learning in ASD.

Interdisciplinary aspect of the project: In this proposal, we combine three key disciplines: biology, mathematics, and engineering, that all provide complementary perspectives on neural circuit function. Using a highly interdisciplinary approach, we aim to identify computational principles underlying adaptive decision-making and their dysfunctions in neurological disorders. Neurophysiological experiments are necessary to obtain experimental data on neuronal activity under conditions of error processing, and to dissect the underlying cellular and circuit mechanisms using causal intervention. Such data will then be compared to test predictions on the dynamics that are inferred from theoretical concepts of reinforcement learning and predictive coding using computational models.
Funding: NA