BraiNets

Brain Networks - From neural computations to cognition

BraiNets is a research team of the Institut de Neurosciences de la Timone in computational neuroscience working at the interface between system and cognitive neuroscience, network neuroscience, machine learning and artificial intelligence. The BraiNets team exploits most advanced theoretical approaches and computational tools from artificial intelligence and statistics to study how brain interactions support neural computations underlying cognitive functions.

ResearchMembersPublications

The main research axes of BraiNets are:

Andrea Brovelli. Brain interactions and human learning.

The research projects of Andrea Brovelli (DR, CNRS) investigate the role of brain interactions in goal-directed learning and adaptive behaviors. Methodologically, they combine functional connectivity and information theoretical analysis of brain interactions, multimodal brain imaging and neurophysiology (MEG, intracranial EEG, LFP and fMRI), and computational models of human learning. Publications and citation record are on Google Scholar.

*Neural interactions in prefrontal cortex encoding reward and punishement prediction errors during goal-directed learning (Combrisson et al., eLife 2024)*

Emmanuel Daucé. Learning, motor control and Information seeking in the brain

We focus on developing models of information-seeking action selection, integrating insights from cognitive science and neural systems. The work explores (i) how visual processing exploits the log-polar structure of retinal inputs to target relevant visual information through saccades (ii) how information seeking interacts with learning and decision-making during goal-directed action selection and/or target-oriented visuo-motor adaptation. Additionally (iii) we investigate population coding and brain decoding in cortico-basal dynamics, providing insights into neural circuits involved in motor control and decision-making in normal an pathological condition (Parkinson’s disease).

Matthieu Gilson. Anatomo-functional modelling of healthy and pathological brain neuronal networks

Model-based analysis of fMRI data by anatomo-functional model constrained by anatomical MRI data. The model effective connectivity is optimized to reproduce functional MRI (BOLD) statistics and interpreted as directional interactions betwen cortical areas (Gilson et al., Netw Neurosci 2020)

To uncover ho neuronal circuit process information, we develop models of neuronal networks to link their dynamics with their functions based on distributed representations of stimuli and internal processes (e.g. motivation). This model-based approach is useful to interpret neuroimaging and electrophysiological data (spike trains, fMRI, EEG, MEG) in terms of brain communication at the macroscopic level and in terms of neuronal computations at the microscopic level.

Bruno L. Giordano studies how the cerebral auditory network transforms input acoustic patterns into a semantic knowledge of the auditory environment. This goal is achieved by assessing the extent to which computational models (acoustic representations, and sound and language artificial neural networks) predict electrophysiology (MEG), neuroimaging (fMRI), and behavioral responses to natural sounds.

JD Lemaréchal develops Bayesian Inference in the context of modelling of electrophysiological data. In particular, he focuses on a biophysical model of high-gamma activity and proposes techniques to infer connectivity parameters from data simulated in different conditions, and from empiral dataas well. One of the state-of-the-art bayesian techniques used to estimate parameters from data is called “Simulation based inference”.

The BraiNets team develops toolbox for the analysis and modelling of brain dynamics.

Frites is a Python toolbox for networks-level analysis of brain data based on functional connectivity methods and statistics.

HOI (Higher Order Interactions) is a Python package for network-level analysis of brain data beyond pairwise interactions using information-theoretical metrics.

pyMOU is a Python package for simulation and connectivity estimation using the multivariate Ornstein-Uhlenbeck process (MOU): https://github.com/MatthieuGilson/pyMOU