Research

PhD Defensemanuscript slides

Thinking Out of the (Black)-Box: Tools for machine learning audits in the presence of deceptive model providers

Robust ML Auditing using Prior KnowledgeICML25 arXiv poster code scholar

Jade Garcia Bourrée*, Augustin Godinot*, Martijn de Vos, Milos Vujasinovic, Sayan Biswas, Gilles Tredan, Erwan Le Merrer, Anne-Marie Kermarrec

How can auditors use prior knowledge to improve the robustness of their audit?
Queries, Representation & Detection: The Next 100 Model Fingerprinting SchemesAAAI25 arXiv poster code scholar

Augustin Godinot, Erwan Le Merrer, Camilla Penzo, François Taïani, Gilles Tredan

How do we evaluate fingerprinting methods and how can we do better?

Under manipulations, are there AI models harder to audit?SaTML24 arXiv slides poster scholar

Augustin Godinot, Erwan Le Merrer, Gilles Tredan, Camilla Penzo, François Taïani

Under platform manipulations, AI model cannot be audited in black box more robustly than by random sampling.

LDPC error correcting codesreport code

Generation and simulation of LDPC codes

Exposure diversity in music recommender systemsCNA21 journal slides code scholar

Study of exposure diversity through the lens of an Heterogeneous Networks diversity measure.

graphiareport code

Practical analysis of dynamical networks. Development of an analysis tool in Rust.
Real-time artificial horizon

Implementation of Hardware Abstraction Layer modules for the screen and I2C peripherals on the STM32F746 processor. Developed an artificial horizon as an example.