Faites connaître cette offre !
Reference : UMR5824-TAIDAO-013
Workplace : ECULLY
Date of publication : Friday, October 04, 2019
Scientific Responsible name : Mehdi Khamasi (ISIR) and Mateus Joffily (GATE)
Type of Contract : PhD Student contract / Thesis offer
Contract Period : 36 months
Start date of the thesis : 1 December 2019
Proportion of work : Full time
Remuneration : 2 135,00 € gross monthly
Description of the thesis topic
The objective of our project is to investigate the neural and computational bases of causal learning. In particular, we will focus on causal learning in the context of goal-directed instrumental behaviors, which rely on learning rules determined by the contingency between actions and outcomes . In order to unravel the neural and computational bases of action-outcome causal learning, we need to lift two key barriers. The first barrier is the lack of neurocomputational models that formalise the above-mentioned theoretical framework to make predictions about the underlying neural computations. The second barrier is the lack of clear understanding of the brain network dynamics supporting action-outcome causal learning.
The selected candidate will contribute to lift the first barrier, by developing neurocomputational models that: i) formalise internal representations and computations predicted by causal learning theories (rational and Bayesian frameworks); ii) make predictions about the dynamics of neural activity (i.e., neurobiological plausibility) and fit single-participant behavioral patterns (i.e., computational flexibility). Among the possible learning models, two seem to provide the adequate theoretical and computational framework: Active Inference and Reinforcement learning. The former is a Bayesian approach postulating that human behaviour can be reduced to the minimization of variational free energy, which is an upper bound on Shannon surprise (Friston, 2010). The latter provides a complementary view and formalizes human behavior as a process that aims at maximizing cumulative reward (Sutton & Barto, 1998).
The selected candidate will help designing a new experimental protocol to test the specific predictions of these computational frameworks, and compare them with Bayesian model comparison methods. The experiment will involve human subjects and be realized with the facilities of the GATE (Experimental Economics) laboratory in Lyon, France. Then we will derive a computational model which best accounts for human behavior while they learn causalities, and make model-driven predictions for a new task involving brain imaging (fMRI, MEG, SEEG) performed in humans by partners of the project.
Applicants should be highly motivated and have a strong background in neurosciences, physics or related fields. Confirmed experience in computational modelling and programming skills are mandatory. Preference will be given to applicants with previous experience in causal learning models in neurosciences.
The thesis will take place mainly in the GATE-LSE in Lyon, with a secondary affiliation to ISIR in Paris. It will be co-directed by Mehdi Khamasi (ISIR) and Mateus Joffily (GATE).
We have obtained funding from the French Agence Nationale de la Recherche for 4 years project aiming at lifting the previously mentioned two barriers. The project involves both theoreticians (Mateus Joffily in GATE, Lyon, France; Mehdi Khamassi in ISIR, Paris, France; David Lagnado in UCL, London, UK) and experimentalists (Andrea Brovelli in INT, Marseille, France; Julien Bastin in GIN, Grenoble, France).
Applications should be submitted by this portal and the following documents should be sent by e-mail to Mehdi Khamassi (email@example.com) and Mateus Joffily (firstname.lastname@example.org): 1) a cover letter briefly describing experience, motivation and skills adapted for the position, as well as research interests; 2) complete CV and publication list; and 3) two letters of reference that should be sent directly to us by the evaluators.
We talk about it on Twitter!