INRIA Grenoble PhD Seminars

June 2026

Speakers :

• Ivan Baheux-Blin, PhD student at PRIVATICS team (Inria) : Studying Android in the Age of Restricted Transparency
  Linkedin Slides
   With billions of users and a vast diversity of devices, Android represents one of the largest and most influential digital ecosystems in the world. Its scale makes it a crucial target for researchers, auditors, and regulators seeking to understand how mobile platforms operate and whether they comply with privacy and consumer protection requirements.
   Yet, despite Android’s open-source foundations, analyzing the platform has become increasingly challenging. Over the years, new security mechanisms and platform restrictions have reduced visibility into system behavior, often limiting the ability of independent actors to observe, measure, and verify how data is collected and used. In many respects, the mobile ecosystem is evolving toward a model that offers users, and those studying the platform, less control and transparency than traditional web environments.
   This presentation reflects on a PhD journey that began with studying Android itself and gradually shifted toward a broader question: how can researchers and regulators investigate digital platforms when the platforms actively constrain observation? Through experiences navigating technical barriers, changing platform requirements, and researcher-pentester collaborations, I will discuss the growing tension between security, transparency, and accountability, and explore the adversarial landscape faced by those working to verify whether privacy promises and legal obligations are truly being respected.


• Imen Bajar, PhD student at PRIVATICS team (Inria) : Beyond the Mask : The illusion of privacy in brain MRI
  Linkedin Slides

   Brain magnetic resonance imaging data contain highly sensitive personal information, including facial features and the unique anatomical structure of each brain (“brain fingerprint”). While defacing methods aim to remove facial information, the extent to which they can protect against re-identification based on brain features remains unclear. In this study, we assess four state-of-the-art defacing methods fsl_deface, PyDeface, PyFaceWipe, and mri_reface on T1-weighted scans from the OASIS 2 dataset, extracting 15 neuroanatomical features with FreeSurfer. Using a similarity-based re-identification attack with probabilistic linear discriminant analysis and Canberra distance, individuals remain identifiable with success rates of 66–73% and low equal error rates. Our results show that current defacing approaches only partially protect privacy: the brain itself carries sufficient identity information for re-identification. These findings highlight the need for anonymization strategies that obscure brain fingerprints while preserving the utility of MRI data.
   Index Terms— Privacy, MRI, Defacing, Brain, Re-identification.