M/F: Realtime Modification of Social Signals in Augmented Reality using Diffusion Models

Organisation/Company CNRS Department Laboratoire Interdisciplinaire des Sciences du Numérique Research Field Engineering Computer science Mathematics Researcher Profile First Stage Researcher (R1) Application Deadline 19 Sep 2026 - 00:00 (UTC) Country France Type of Contract Temporary Job Status Full-time Hours Per Week 35 Offer Starting Date 1 Oct 2026 Is the job funded through the EU Research Framework Programme? Not funded by a EU programme Is the Job related to staff position within a Research Infrastructure? No

Offer Description

Dr. Sandor, the principal advisor, leads the ARAI (Augmented Reality – Artificial Intelligence) team in the HCI department of the Laboratoire Interdisciplinaire des Sciences du Numérique (LISN, UMR 9015), Université Paris-Saclay/CNRS, which he co-founded with Dr. Nguyen in 2024. His recent research on video prediction and on real-time facial-expression modification with diffusion models is directly at the heart of this thesis.

Dr. Nguyen, the co-advisor, is the scientific coordinator of NeuroCoRe and co-leads this Ph.D. project with Dr. Sandor. Her expertise in collaborative XR—both remote and co-located—and in hybrid immersive platforms is central to the multi-user and user experience aspects of the thesis.

The thesis is embedded in the NeuroCoRe consortium (PEPR eNSEMBLE): collaborative XR and generative AI (LISN), neuroscience and decision science (CRNL), and computer vision, machine learning, and adolescent BPD psychiatry (Centre Borelli). The student benefits from this interdisciplinary environment, shared experimental infrastructure, and access to high-performance GPU computing at IDRIS (Jean-Zay). LISN is a key player in the French PEPR eNSEMBLE and CONTINUUM projects on the future of collaboration; this thesis provides strong synergy with them.

Consumer Augmented Reality (AR) smart glasses will increasingly let AI modify a user's view of the real world. This raises a largely unexplored question: can a user's perception of social signals—especially facial expressions—be deliberately modified in real time? Social interaction relies on such signals: people constantly read each other's emotional states to infer intentions, predict reactions, and guide their own behavior. This thesis studies how these signals, as perceived by an AR-glasses wearer, can be deliberately and convincingly modified in real time using diffusion models.