A project carried by Yves Boubenec. 

Many neurodevelopmental disorders (NDDs) are associated with perturbations of functional interactions between cerebral areas. This makes the quantitative and non-invasive assessment of those interactions a potential efficient biomarker to ensure early detection and better prognosis of these pathologies. However, limitations in the currently available neuroimaging techniques in terms of spatiotemporal resolutions and clinical applicability has hindered the wide clinical deployement of such functional connectivity-based diagnosis.

Our project tackles this challenge by developing spatiotemporal functional connectivity (STFC), an innovative characterisation of multiscale (mesoscopic to large scale) brain activity expected to offer unprecedented sensitivity and specificity. It leverages functional UltraSound (fUS) neuroimaging, a cutting-edge brain imaging modality with high translational potential, and innovative post-processing techniques focusing on propagative brain activity. The ultimate goal is to identify potential fUS-based STFC biomarkers of NDDs suitable both for preclinical and clinical neonatal imaging.

First, we will develop a set of descriptor tools tailored to specifically capture propagative activity in neural (Optical and fUS) data and further derive a STFC processing technique, relying on these descriptors, to estimate repetitive propagative spatiotemporal patterns. Secondly, we will  benchmark the performance of these methods in scoring spontaneous cortical dynamics associated with distinct animal states (e.g., active, quiet, sleep phases, sedation), using calcium and intrinsic optical imaging data recorded synchronously in GCaMP6 mice. This will be a crucial step to retain the most pertinent (robustness, reproducibility) spatiotemporal descriptor of cortical  propagative activity extracted from both neuronal and hemodynamic-related signals. It will also enable to better characterize how hemodynamics, which can be non-invasively observed through functional ultrasound (fUS) are tied to underlying neuronal activity (calcium imaging). Furthermore the technique will also be used to score animal states in ferrets using fUS as ferrets have a mid-sized brain and folded cortex with closer homology with humans for frontal and sensory regions. Our final aim is to validate that fUS-based STFC can efficiently score pathological brain states. We will assess the potential of STFC applied on fUS data as an early biomarker for NDDs in a relevant cross-species (mice and ferrets) preclinical model of NDDs. We will implement the maternal immune activation (MIA) model in those two species and then track specific STFC signatures evidencing alterations in spontaneous cerebral dynamics in the  young adult (mice and ferrets) and at perinatal stages (ferret pups at P5-P12). Building from the biomarkers defined in ferrets, we will then search for comparable STFC signatures in fUS data acquired in human neonates (via a clinical trial external to this ANR proposal). This project will bridge the gap between animal models and human neonates to reveal early indicators of NDDs with a new type of analytical biomarkers, offering the potential for timely intervention and improved long-term outcomes for affected individuals.

 

A project carried by Hernan Anllo.

Integrating information coming from others into our own worldview is one of the most complex and efficient forms of social learning, as it largely mitigates the costs of exhaustive individual exploration through the capitalization of others’ experiences. However, while recent efforts have shed considerable light into how human learning is computationally modeled, very little is known about how language-based human-to-human transfer of knowledge can be computationally implemented in the context of learning new (or modifying old) goal-directed behaviours. Indeed, oftentimes we are taught how to proceed rather than having to discover optimal behaviors purely by ourselves. Yet, despite its ecological centrality, we have very limited mechanistic and computational understanding of these kinds of teaching processes. This project consists of an innovative set of work packages conceived to bridge this gap by combining cutting-edge methods in cognitive science, clinical and social psychology, with state-of-the-art techniques from Reinforcement Learning, Natural Language Processing and Large Language Models such as GPT-4. The main goal of this action is to develop an understanding of the computational makeup of different forms of verbal knowledge transfer (e.g., lessons, advice, suggestions), taking place between experienced learners (e.g., teachers, medical doctors) and novel naive learners (e.g, pupils, patients), in a precise sociocultural context (e.g., a lab, a classroom, a hospital), with concrete behavioral and experiential consequences (e.g, improving decision strategy, obtaining a clinical benefit).

A project carried by Srdjan Ostojic.

Mathematical models of neural networks are a key tool for understanding how millions of neurons in the brain work together to implement behaviorally relevant computations. Recent progress in artificial intelligence has provided new methods for generating network models that perform the same cognitive tasks as investigated with animals in systems neuroscience. Analyzing and interpreting such model networks has led to important insights into how computations emerge from collective dynamics of neural activity, yet the available models often lack key aspects of the biological structure of the brain.

A prominent level of structure in the mammalian cortex is the organization into  anatomically-defined regions, and a premise of much of cognitive neuroscience is that different regions play different computational roles. A mechanistic understanding of the functional significance of the multi-area structure however remains elusive due to the paucity of multi-area network models and multi-area electrophysiological recordings to constrain them.

In this project we will develop a new class of mathematically tractable multi-area recurrent neural networks that we will compare with recordings in behaving ferrets to identify the constraints that complex behavioral tasks impose on the structure of activity. The central goal of the project will be to test the hypothesis that a multi-area structure endows a network with compositionality, i.e. the capacity to robustly perform multiple tasks by combining simpler modularized computations. 

To this end, we will combine three complementary methods: (i) mathematical analyses of network dynamics;(ii) reverse-engineering of networks trained using back-propagation to perform multiple tasks; (iii) comparisons of predictions generated by model networks with cortical activity recorded in animals performing identical tasks.

A project carried by Nicolas Baumard.

Psychology is essential to understanding human history. When aggregated, changes in individual psychology - in the intensity of social trust, parental care or intellectual curiosity - can lead to significant changes in institutions, social norms and cultures. Yet the study of psychological changes over time remains difficult, since it is impossible to use standard methods (questionnaires, behavioral measures). Recent developments in psychology, however, suggest that cultural artifacts partly reflect the psychological traits of the individuals who produced or consumed them. It is therefore possible to carry out "historical psychology", i.e. to use the concepts and methods of psychology to trace and quantify the evolution of psychological traits over time.

A project carried by Tarryn Balsdon.

To interact with the dynamic world around us, we must process sensory information and plan behaviour efficiently. While behavioural efficiency has been studied as an exogenous speed-accuracy trade-off, it must be moderated by endogenous variables: confidence, urgency, effort, and value. This project seeks to understand the interplay of these variables in increasingly naturalistic behaviour. We will tease apart the role of these variables using experimental manipulations while developing a computational description of behaviour. This computational framework captures behavioural efficiency through the coordinated effort of perceptual, cognitive, motor, and metacognitive processes, to allow online updating in response to dynamic sensory environments. Electroencephalography will be used to develop this framework in line with biologically plausible neural mechanisms. First, we will examine how urgency could interact with confidence to drive faster actions, and the extent to which this operates at the level of motor processes. Second, we will examine how this confidence-urgency trade-off is moderated by the value or effort associated with behavioural outcomes. Finally, we will use a dynamic visual search task to examine how these processes are coordinated over several sub-decisions to achieve an overarching goal, as in naturalistic behaviour. This project will unravel how interconnected neural processes are coordinated to control how we moderate how quickly or deliberatively we plan behaviour.
 

A project carried by Sharon Peperkamp with Sho Tsuji and Anne Christophe.

Children around the globe learn language, a complex system with several interacting levels, with astonishing speed and efficiency. Despite enormous progress in the description of the time-course of language acquisition on the one hand, and a growing body of research showing the crucial role of social interactions in cognitive development on the other hand, we still lack an integrated understanding of how linguistic and social cues in the environment drive early language acquisition. Our project aims at investigating the mechanisms through which both linguistic and social aspects of children’s learning environments contribute to critical aspects of language acquisition across several levels of representation, from phonetics and phonology to the lexicon and morphosyntax.

First, we will consider the variability of cues in the language input, and investigate how cues at one linguistic level drive learning at another level, as it has become apparent that studying the development of linguistic levels independently poses limits on understanding the mechanisms of acquisition. We will thus gain a more precise and detailed understanding of the contribution of linguistic cues to language learning.

Second, we will expand our work to integrate a hitherto underexplored dimension, i.e. the role of social information. Children learn language from the people around them who enrich the linguistic input with social cues. In addition, they are not passive recipients of linguistic information, and their active role might play a key role in explaining their astonishing language development. We propose that social information acts as a gateway into language, facilitating the detection of subtle linguistic cues.

We will rely on a combination of well-established and innovative experimental methods, the latter including studying learning as it unfolds using interactive screens and toys to emulate social interaction, as well as infant online testing to assemble large and diverse datasets.
 

A project carried by Frédérique de Vignemont in collaboration with Alessandro Farnè, Jérémie Lafraire.

 

Space is rarely the focus of our perception but, as the pandemic made most of us aware of, we do monitor the space separating us from others, and this is so thanks to an evolutionary ancient mechanism that is constantly at play, or so it is argued. A specific network of sensorimotor neurons (in parietal, premotor and subcortical areas), first found in monkeys and later in humans, represents the space

closely surrounding the body, also known as peripersonal space (PPS). PPS can be seen as a buffer zone between the self and the world to be better prepared. Whether there is in PPS a snake to avoid, an apple to grasp, or an angry person to run away from, the brain processes them at both perceptual and sensorimotor levels in a manner different than if they were located farther away. However, we typically navigate in rich environments, constantly surrounded by multiple objects and congeners. Whether we are in a supermarket or at work in front of our desk, PPS is literally packed and there is a risk of computational overload. How, then, does the brain operate?

Despite more than 20 years of research on PPS in humans, we do not know. In this project we contrast two options:

(i) Proximity account: Proximity suffices to give a unique significance to stimuli and thus, to trigger PPS mechanisms because everything that is close-by matters. PPS processing is constantly activated. Factors other than proximity have only a modulatory effect.

(ii) Proximity+ account: Proximity does not suffice because the computations underlying PPS have neural, energetic, and cognitive costs. To manage costs, only some of our nearby stimuli gets high enough priority values to turn PPS processing ON.

In this multidisciplinary project, we shall theoretically and experimentally investigate the explanatory power of Proximity+ by assessing the role of three main dimensions in giving priority to stimuli surrounding us to switch PPS processing on: salience, valence, and social relevance.

A research project carried by Rocco Mennella.

Despite the severe clinical impact of socioemotional difficulties in patients with major depressive disorder (MDD), objective behavioral markers of social impairment are lacking. One of the two core symptoms of depression, anhedonia, the lack of motivation and/or pleasure in rewarding activities, has been suggested as a major determinant of altered social behavior, including decision-making in socioemotional contexts. On the other hand, there is suggestive evidence that anhedonia is also associated with measurable changes in postural and gait parameters in depression. The WANDERS project aims at examining the relationship between anhedonia, posture and gait, and socioemotional decision making. Based on our theoretical model, we hypothesize that anhedonia in the general population and in individuals with MDD causes changes in both posture/gait parameters and socio-emotional decision-making due to common neural determinants in fronto-striatal dopaminergic circuits.  In the first work package (WP1), we will induce reduced positive affect (anhedonia) in healthy participants via sleep deprivation to experimentally test our hypothesis. Participants will perform a social decision-making task, consisting in approaching or avoiding emotional individuals in virtual reality (VR) while three-dimensional (3D) motion parameters are recorded. WP1 will elucidate the relationship between anhedonia, postural and gait changes, and socioemotional decisions in the general population by combining VR, 3D motion analysis, and computational models. WP2 will translate these findings into the clinic to generalize the results of WP1 to a sample of MDD patients. The WANDERS project fits within the scope of the CES 28 by investigating a central dimension of human sociality in both healthy and clinical populations. This will help establishing a novel framework for understanding the affective determinants of social decisions and unravel objective behavioral markers of the quality of social life. The solid background of the scientific coordinator, together with the decisive help of a high-quality interdisciplinary network of collaborators, will allow realizing the project.