Alexandre Blain

Alexandre Blain

Research Scientist at SurgeCare · PhD in Machine Learning & Statistics

I am a Research Scientist at SurgeCare, a spinoff from Brice Gaudillière's lab at Stanford University. I'm working on building ImmuneMind, a foundation model trained on billions of single cells, and developing interpretability methods for biomarker discovery.

I previously studied at ENSAE Paris and completed the Master MVA (Mathematics, Vision and Learning) at École Normale Supérieure Paris-Saclay.

I hold a PhD in Machine Learning and Statistics from INRIA Parietal, advised by Bertrand Thirion and Pierre Neuvial. I am interested in interpretability and foundation models across all modalities.

Updates

  • Feb 2026 ImmuneMind awarded the AI Trailblazer Grant as part of the France 2030 plan!
  • May 2025 False Coverage Proportion control for Conformal Prediction accepted at ICML 2025.
  • Jan 2025 Joined SurgeCare as Research Scientist. I will be working on foundation models for single-cell cytometry data!
  • Dec 2024 Successfully defended my PhD thesis in Machine Learning and Statistics at INRIA.
  • May 2024 New preprint — When Knockoffs fail: diagnosing and fixing non-exchangeability of Knockoffs.
  • Dec 2023 False Discovery Proportion control for aggregated Knockoffs accepted at NeurIPS 2023.

Publications

False Coverage Proportion control for Conformal Prediction ICML 2025
A. Blain, B. Thirion, P. Neuvial
Distribution-free uncertainty quantification with improved statistical guarantees, applicable to any black-box model including neural networks.
Paper · Code
When Knockoffs fail: diagnosing and fixing non-exchangeability of Knockoffs Preprint
A. Blain, A. Reyero Lobo, J. Linhart, B. Thirion, P. Neuvial
A diagnostic tool based on Classifier Two-Sample Tests that detects violations of the knockoff exchangeability assumption — which cause massive false positive inflation — and an alternative construction that restores error control.
arXiv
False Discovery Proportion control for aggregated Knockoffs NeurIPS 2023
A. Blain, B. Thirion, O. Grisel, P. Neuvial
Controls the actual proportion of false discoveries (FDP) — beyond the usual FDR guarantee — for Knockoff-based variable selection in high dimension, with a new aggregation scheme that removes the randomness of classical Knockoff inference. Demonstrated on brain imaging and genomic data.
Paper · arXiv · Code
Notip: Non-parametric True Discovery Proportion control for brain imaging NeuroImage
A. Blain, B. Thirion, P. Neuvial
Distribution-free method for detecting statistically significant activations in high-dimensional data, with applications to fMRI. Oral at OHBM 2022 and MCP 2022.
Full text · Code

Software

Notip
Non-parametric True Discovery Proportion control for brain imaging.
KOPI
Knockoffs with FDP control and derandomization.
sanssouci.python
Python implementation of the SansSouci post-hoc inference framework.

Selected Talks