You are interested in applying for a position in the lab? Contact me: jean.daunizeau[at]gmail.com.

Below is a list (by no means, exclusive) of potential research projects:

Cognitive biases in learning and decision making

Why do we tend to find a hypothesis that we like more plausible than a hypothesis that we don't like? Why, instead of looking for contradiction, do we look for evidence that confirm our prior beliefs? Cognitive biases are mechanisms of the mind that induce systematic deviations from logical and rational thoughts. They have been demonstrated in a plethora of learning and decision contexts, but their expression seems to show high inter-individual variability. This lacks a coherent explanatory framework. This project's working hypothesis is that the expression of biases is determined by the allocation of mental resources to decisions, which is based upon a motivational cost-benefit trade-off. This would imply that inter-individual differences in parameters that control this type of trade-off (effort/confidence ratio, sensitivity to effort cost, etc...) are predictive of inter-individual differences in learning and/or decision making biases. The aim of this project is twofold. On the one hand, we aim at devoping a computational model that can relate the cognitive process of mental effort allocation to cognitive biases (in particular: overconfidence bias, optimism bias and confirmation bias). On the other hand, we aim at designing a set of learning and decision making experiments that can test the model's predictions (in particular: we aim to predict inter-individual differences in the expression of cognitive biases).

Exploration/exploitation arbitrages in instrumental learning tasks

Learning consists in the acquisition or modification of knowledge, behaviour or skill on the basis of novel information. Instrumental learning is a specific case of learning, in which reward and/or punishment feedback is provided, conditional upon the occurence of the response. When the action-reward contingency is unknown, agents must learn by trial and error. This induces a so-called exploration-exploitation dilemma (Berger-Tal et al., 2014). On the one hand, agents must explore to discover novel rewarding actions. On the other hand, they also must exploit their current knowledge in the aim of maximizing expected reward. So far, this dilemma was not given any firm computational basis. More precisely, exploratory behaviour is typically modelled as either stochastic and unspecific deviations from greedy strategies, or arising from, e.g., heuristic uncertainty bonus incentives. This project proposal primarily aims at modelling exploration-exploitation dilemmas from a semi-normative computational perspective. In brief, we will consider information-theoretic incentives that either favour (e.g., curiosity) or impair (e.g., perseveration) exploratory behaviour. The relative weight of action-reward contingencies, curiosity and perseveration onto decisions thus effectively balances exploitation against exploration. The primary aim of this project is twofold: (i) assessing the adaptive value of ensuing exploitation/exploration balances, and (ii) documenting the effect of neuropsychiatric conditions and subclinical states (e.g., mood, fatigue, etc...) onto the exploitation/exploration balance. The latter objective is made possible because we have acquired data in the context of instrumental learning tasks performed by different cohorts of neurological and psychiatric patients.

Neural encoding of value

Recall that "motivation" can be defined as the set of processes that generate goals and thus determine behaviour. Here, a goal is but a valuable state of the world. This is why the question of how the brain constructs, represents and processes value has reached such a central position in today's decision neuroscience. This project aims at (i) developping new models of value encoding in the brain, and (ii) evaluating them on both neuroimaging data acquired on healthy human subjects and invasive electrophysiological recordings acquired in monkeys. Although the neural code of value is virtually unknown, a good starting point here are the dual notions of receptive fields and population coding, borrowed from vision neuroscience. But alternative quantitative scenarios will be considered as well, which may be more specific to relevant cognitive contexts (e.g., learning and/or decision making). In addition, some of the macroscopic properties of the value code are already quite established (genericity, automaticity, auto-correlation, etc...). Ideally, plausible computational models would have to explain these properties from neurobiologically-realistic microscopic scenarios.

Neurocognitive models of mental resources' allocation

What makes you ponder a decision? Some decisions we make are implicit and automatic (e.g., where and how to move our limbs when we walk), and some require some form of explicit deliberation (e.g., choosing among different lunch meals). Such internal deliberation typically engages so-called “executive” resources, such as inhibitory control (in order to overcome distraction and/or automatic competing responses) and working memory (in order to manipulate behaviourally-relevant information). The allocation of such executive resources is referred to as “cognitive control”, which can be broken down into three core component functions (Shenhav et al., 2013): (i) regulation (i.e. governing or influencing lower-level information processing), (ii) specification (i.e. deciding where and how much resources have to be allocated) and (iii) monitoring (i.e. evaluating how well allocated resources meet task demands). This project proposal primarily aims at understanding the critical role of cognitive control in mediating the impact of decision difficulty and importance onto decision speed and accuracy, from both a computational and neurobiological perspective.

Mathematical modelling of the impact of decisions onto mental states (and back)

According to classical decision theory, we make decisions based upon beliefs and preferences: the rational choice is the one that maximizes expected value. In other words, preferences and beliefs cause behaviour. However, making a decision can be proven to change values (as if people thought: "since I chose it, I must have liked it"). This is referred to as "choice-induced preference change" (Izuma et al., 2013). Such experimental results lack a clear computational explanation: why should behaviour feedback onto preferences and beliefs? In addition, it raises a number of connected questions: (i) what is the underlying value updating mechanism?, (ii) does it work the same for beliefs and preferences?, and (iii) is it conditional upon attentional and/or mnesic factors? These are the questions we will address in this research project, using both experimental and modelling approaches.

Neural network models for integrating whole-brain neuroimaging and behavioural data

Neurobiological system modelling of neuroimaging data has become a standard tool for identifying the structure and plasticity of macroscale brain networks that respond to the experimental manipulation (e.g., sensory stimuli or task demands). These models, however, do not explain how distributed brain responses are causally involved in the production of behaviour (e.g. choices, reaction times). This project will be aimed at extending existing neural systems' models with behavioural outputs. Eventually, such models will serve to identifying a neural transfer function that would map experimental inputs to their behavioural response, through activity in the underlying large-scale brain dynamics. Informed with neuroimaging data, they will (i) provide a direct quantification of the behavioural relevance of brain connectivity, and (ii) predict behavioural deficits induced by specific anatomical lesions (and their recovery through plasticity).

AI modelling of big data from a "cognitive" smartphone app

In january 2015, our team has launched the BRAiN’US project, which aims at crowdscourcing neuroscientific knowledge (https://sites.google.com/site/brainusapp/). BRAiN'US gathers 8 psychological experiments into a smartphone app, designed to test cognitive processes engaged in social cognition, including executive functions, instrumental learning and theory of mind. In brief, the project mostly consists in analysing the (big) data that have already been acquired from these tests. To date, the sample size has reached about 32,000 participants, 1,800 of whom have declared a neurological and/or psychiatric pathology. The aim of the project is twofold:

(i) Developing AI models (in particular: cognitive computing models) that can be used to extract subject-specific parameters that quantify individual cognitive styles. This aspect of the project may or may not get inspiration from previous work on learning and mentalizing by the team, see: this theoretical (evolutionary) work, and its exerimental application on healthy adults, on seven primate species, and on people suffering from ASD.

(ii) Evaluating the diagnostic value of cognitive profiles derived from tests results. In other words: can we classify healthy individuals from neurological/psychiatric patients, and/or maybe re-segment heterogeneous pathological groups into distinct subclasses?

These results will also serve as feedback in the design of the next release of BRAiN’US (2.0). The added value of the project lies in the unique combination of big data and mathematical modelling, which has the potential to uncover subtle and/or nonlinear mechanisms.