Home
I am a researcher at onepoint, a member of the TALia laboratory, which is a joint research unit between Télécom Paris and onepoint.
My research work focuses on the areas of deep learning, NLP, and cognitive sciences.
Keywords: Deep Learning, NLP, Cognitive Sciences, Causality, Graphs
Research
Recent publications
- Plaud R., Perez-Lebel A., Saillenfest A., Bonald T., Le Morvan M., Varoquaux G. & Labeau M. (2026). Tailoring Strictly Proper Scoring Rules for Downstream Tasks: An Application to Causal Inference., ICML 2026.
Preprint: arxiv
- Saillenfest A. & Lemberger P. (2025). Nonlinear Concept Erasure: a Density Matching Approach, Proceedings of ECAI 2025.
Full version: arxiv
- Plaud R., Perez-Lebel A., Labeau M., Saillenfest A., Bonald T. (2025). To Each Metric Its Decoding: Post-Hoc Optimal Decision Rules of Probabilistic Hierarchical Classifiers, ICML 2025.
Preprint: arxiv
- Plaud R., Labeau M., Saillenfest A., Bonald T. (2024). Revisiting Hierarchical Text Classification: Inference and Metrics, CoNLL 2024.
- Lemberger, P., & Saillenfest, A. (2024). Explaining Text Classifiers with Counterfactual Representations, ECAI 2024.
See all my publications.
Current projects
I am interested in the causal evaluation of biases related to concepts such as gender or race in the field of NLP.
For research/education purpose, I built a dataset: Wikivitals-lvl5-04-2022.
This dataset was built from Wikivitals articles of level 5. It is a corpus of ~48.5k documents with an underlying graph structure (~2,3M edges) and a 3-level hierarchy of labels.
Supervision and teaching activities
PhD students
- (2024, September - …) Roman Plaud. Apprentissage et Manipulation de Représentations de Données
Textuelles Structurées (Learning and Manipulating Structured Text Data Representations). Co-supervision with Thomas Bonald and Matthieu Labeau.
Master’s students
Teaching
External links