Gabriel Loiseau

Hello! I am a third year PhD student in Natural Language Processing at Hornetsecurity and Inria Lille, advised by Marc Tommasi, Damien Sileo (Inria MAGNET team), Damien Riquet and Maxime Meyer (Hornetsecurity). My research focuses on the implementation of utility-preserving automatic text anonymization methods. The project is funded by a CIFRE aggreement, a French government-funded program promoting collaborative research between companies and research labs.

Prior to my PhD, I worked on designing a compact deep neural network with state-of-the-art performance for phishing detection at Vade (which got acquired by Hornetsecurity in 2024). During my Msc. I also worked on deep learning explainability for music generation, in the Algomus group at the CNRS.

You can find more information on my LinkedIn profile.

Works

My publications are available on the dedicated publications page and through my Google Scholar profile. A comprehensive list of my presentations can be found on the talks page, and you can explore the public software libraries I’ve developed and contributed to on my GitHub.

News

• September 2025. Our Tau-Eval paper has been accepted to EMNLP 2025.
• June 2025. I’ll be presenting our new evaluation framework for text anonymization at the 15th Atelier sur la Protection de la Vie Privée in Chasseneuil-du-Poitou (June 9-12).
• June 2025. New preprint available! We’ve developed a comprehensive evaluation framework and Python library for assessing text anonymization methods.
• April 2025. Presenting at NAACL (April 29 - May 4) on our latest PrivateNLP paper.
• March 2025. Paper accepted at PrivateNLP! Our work introduces a novel reinforcement learning approach for authorship obfuscation.

Contact Me

I’m always happy to discuss NLP research, potential collaborations, and new ideas! Feel free to reach out to me by e-mail (gabriel[dot]loiseau[at]inria[dot]fr).