Mentor/s: Meysam HASHEMI <meysam.hashemi@gmail.com/ meysam.hashemi@univ-amu.fr>
Skill level: Intermediate to Advanced
Required skills: Simulations of differential equations, Python and Git; familiarity with neural mass models, and Bayesian inference in tools such as Numpyro/PyMC would be beneficial.
Time commitment: Full time (350 hours)
About: Bayesian inference on brain models translates into probabilistic estimation of latent and observed states within systems driven by network input and stimuli, modeled by high-dimensional nonlinear differential equations, with potentially correlated parameters. To address these challenges, advanced inference algorithms embedded in Probabilistic Programming Languages (PPLs) have shown remarkable results. In particular, No-U-Turn Sampler, and alternatives such as automatic variational and Laplace approximation in Numpyro and PyMC have demonstrated effectiveness in achieving reliable Bayesian inference even in the presence of multimodal parameter distributions.
Aims:
-
Implementation of dynamical brain models using JAX-based frameworks, such as NumPyro/PyMC, and in-silico validation.
-
Benchmark existing algorithms within to systematically identify their strengths and weaknesses in handling high-dimensional settings.
-
Monitor algorithm convergence for ensuring reliable inference in the presence of multimodal parameter distributions.
Final output will be a lightweight demonstrator with clear documentation, enabling users to quickly run a standardized example. Website: GitHub - ins-amu/DCM_PPLs: Implementing a DCM model of ERP in PPLs. · GitHub