The development of models for neuronal systems have matured in recent years and they exhibit increasing complexity thanks to computer resources for simulation. In parallel, the increasing availability of data poses the challenge to quantitatively related those models to data, going beyond reproducing qualitative activity patterns and behavior. Model inference is thus becoming an indispensable tool for unraveling the mechanisms underlying brain dynamics, behavior, and (dys)function. A critical aspect of this endeavor is the ability to infer changes across multiple scales, from neurotransmitters and synaptic interactions to neural circuits and whole-brain networks. Recent approaches that have been adopted by the neuroscience community include methods for directed effective connectivity (e.g. dynamical causal modeling), simulation-based inference on whole-brain models, and active inference for understanding perception, action and behavior. They have significantly enhanced our ability to interpret data by modeling underlying mechanisms and neuronal processes. This workshop will bring together experts from diverse fields to explore the state-of-the-art methodologies, taking specific applications as examples to compare them and highlight remaining challenges.
Schedule:
* 9:00-9:20, Matthieu GILSON, INT, Marseille, France.
Introduction
* 9:20-9:45, Nina BALDY, INS, Marseille, France.
"Dynamic Causal Modeling in Probabilistic Programming Languages"
* 9:45-10:10, Cyprien DAUTREVAUX, INT, Marseille, France.
"Bayesian Inference of cortico-cortical effective connectivity in
networks of neural mass models with neuroanatomical prior"
* 10:10-10:35, Richard ROSCH, King's College London, UK.
"Integrating multimodal data through Bayesian Inference"
* 10:35-11:00, coffee break
* 11:00-11:25, Meysam HASHEMI, INS, Marseille, France.
"Simulation-based inference on virtual brain models of disorders"
* 11:25-11:50, Levin KUHLMANN, Monash University, Australia
"Bayesian vs Simulation Based Inference for neural population models"
* 11:50-12:15, Ivilin Peev STOIANOV, CNR, Padova, Italy
"Active inference: From Cortical Control to Neural Decoding"
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
# More details:
Matthieu Gilson, Institut de Neurosciences de la Timone, CNRS, Aix-Marseille University, France (email: matthieu.gilson@univ-amu.fr)
Title: TBA
Abstract:
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Nina Baldy, Institut de Neurosciences des Systèmes, INSERM, Aix-Marseille University, France (nina.baldy@univ-amu.fr)
Title: Dynamic Causal Modeling in Probabilistic Programming Languages
Abstract: Understanding brain dynamics requires models that capture both causality and nonlinearity. Dynamic Causal Modeling (DCM) provides a framework for inferring causal interactions among brain regions in response to experimental inputs. We apply Bayesian inference to a neurobiologically grounded generative model that simulates event-related potentials from MEG/EEG data, using nonlinear ordinary differential equations (ODEs) to describe the system’s latent and observed states. To address the multimodality caused by parameter degeneracy, we propose strategies such as hyperparameter tuning, informed initialization, and predictive-weighted model stacking. These enhance the reliability and accuracy of inference. We implement inference and model comparison in several probabilistic programming platforms, evaluating their computational efficiency and performance. Our findings demonstrate that gradient-based Markov Chain Monte Carlo methods, including a self-tuning Hamiltonian Monte Carlo and automatic Laplace approximation, effectively estimate posteriors and manage parameter degeneracy. This work extends DCM beyond standard variational methods, improving model accuracy and robustness. Ultimately, our approach advances the inversion of nonlinear state-space models and supports broader applications of DCM in neuroscience and neuroimaging.
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Cyprien Dautrevaux, Institut de Neurosciences de la Timone, CNRS, Aix-Marseille University, France (email: cyprien.dautrevaux@univ-amu.fr)
Title: Bayesian Inference of cortico-cortical effective connectivity in networks of neural mass models with neuroanatomical prior
Abstract: The characterization of how brain connectivity modulates its activity has become a cornerstone of neuroscience to study cognition and brain pathologies. Here we focus on models of neural masses that aim to fit neurophysiological signals like EEG and MEG, in particular by optimizing interactions between brain regions, following previous work on the dynamic causal model (DCM). Our aim is to improve current framework of effective connectivity estimation. To do such, we compare several modern Bayesian inference schemes for model inversion on data (Variational Inference, Markov Chain Monte-Carlo). We benchmark the identifiability of global parameters that relate to the connections between the cortical regions such as effective connectivity weights as well as some local parameters governing the interplay between several neuronal populations in each region. In this workshop I’ll present you the most recent results on the application of model inversion methods on MEG recordings in two tasks. Finally and importantly, I’ll present the role of anatomical connectivity as a prior in the Bayesian estimation with probabilistic tractography data. Our work provides a quantitative comparison between Bayesian methods in terms of estimation accuracy as well as required computational power.
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Richard ROSCH, Department for Basic and Clinical Neuroscience, Institute of Psychiatry, Psychology and Neuroscience, King's College London, London, UK (email: richard.rosch@kcl.ac.uk)
Title: Integrating multimodal data through Bayesian Inference
Abstract: With technological advances in neuroscience, there is an ever increasing array of quantitative information about cortical structure and function that is becoming available. However, integrating the different data types and perspectives remains challenging, yet there is a need for integrated multimodal models of human brain function that utilises the various data types. In this talk, I will introduce the dynamic causal modelling framework and how multimodal information can be integrated through the use of empirically informed priors in the generative model. I will illustrate this approach on worked examples, e.g. integrating intracranial EEG dynamics and receptor density distribution.
-------------------------------------------------------------
