INRIA Grenoble PhD Seminars

January 2026

Speakers :

• Alice Chevaux, PhD student at STATIFY team (Inria) : Toward Reliable Graph Analysis: Uncertainty Quantification for fMRI Connectivity
  Website Linkedin
Inferring brain graphs from fMRI relies on correlation matrices, yet standard estimators are unstable in high-dimensional, low-sample settings. We propose a general Bayesian framework that avoids structural assumptions and quantifies uncertainty through credible regions for correlation matrices. These regions enable application not feasible with point estimates: (i) diagnosing estimator instability, (ii) robust edge detection with posterior FWER control, and (iii) direct comparison of two fMRI scans via the posterior probability of matrix equality. This simple, assumption-light approach improves reliability and interpretability for downstream connectivity analyses.



• Manuel Campero Jurado, PhD student at DANCE team (Inria) : From Connectivity to Exposure: Flow-Aware Cycling Infrastructure Optimization
  Google Scholar
Cycling can play a major role in making cities healthier, cleaner, and more socially sustainable, yet everyday cycling remains limited in many places because many people perceive riding alongside motor traffic as unsafe or uncomfortable. Since the degree of separation between cyclists and motorized vehicles is a key determinant of perceived safety, this thesis formulates cycling-safety improvement as a budget-constrained network design problem. The cycling network is modeled as a weighted directed graph whose links are assigned a six-level Degree of Separation (DoS) derived from OpenStreetMap. We identify graph theory metrics that capture whether safe facilities form continuous corridors, analyze their monotonic behavior under DoS upgrades, and link them to perception-based survey data to support budgeted optimization. Because topology alone can misallocate investment when demand is low, we extend the framework to cyclist flows through an all-or-nothing assignment based on a generalized distance accounting for slope, angular changes, and traffic exposure. This yields a cyclist-weighted exposure metric and a bi-level feedback mechanism. Finally, we extend the framework to a multi-modal setting, where reducing cyclist exposure may increase car travel times through capacity changes.