Vivien Cabannes

My current research interests revolve around mechanistic interpretability and large-scale experimental design.
Other former interests include representation learning (notably with self-supervised learning), active learning, statistical learning theory, and weakly supervised learning.
Hobbies include art with AI and broader reflections on technology and society.

Publications

Deep Learning Understanding

Vivien Cabannes, Berfin Şimşek, and Alberto Bietti, Learning Associative Memories with Gradient Descent. Preprint, 2024.
Vivien Cabannes, Elvis Dohmatob, and Alberto Bietti, Scaling laws for associative memory. In ICLR (Spotlight), 2024.
Jenny Bao, Aaron Defazio, Alberto Bietti, and Vivien Cabannes, Hessian inertia. In ICML Learning Dynamics Workshop, 2023.
Alberto Bietti, Vivien Cabannes, Diane Bouchacourt, Hervé Jegou, and Léon Bottou, Birth of a transformer: a memory viewpoint. In NeurIPS (Spotlight), 2023.

Representation Learning & Self-Supervised Learning

Vivien Cabannes and Francis Bach, The Galerkin method beats Graph-Based Approaches for Spectral Algorithms. In AISTATS, 2024.
Vivien Cabannes, Bobak Kiani, Randall Balestriero, Yann LeCun, and Alberto Bietti, The SSL interplay: augmentations, inductive bias, and generalization. In ICML, 2023.
Vivien Cabannes, Alberto Bietti, and Randall Balestriero, On minimal variations for unsupervised representation learning. In ICASSP (Oral), 2023.

Active Learning

Charles Arnal*, Vivien Cabannes*, and Vianney Perchet, Mode estimation with partial feedback. Preprint, 2024.
Vivien Cabannes, Léon Bottou, Yann LeCun and Randall Balestriero, Active self-supervised Learning: a few low-cost relationships are all you need. In ICCV, 2023.
Vivien Cabannes, Vianney Perchet, Francis Bach, and Alessandro Rudi, Active labeling: streaming stochastic gradients. In NeurIPS, 2022.

Learning Theory

Vivien Cabannes, Open problem: learning with variational objectives on measures. In IEEE Big Data, 2023.
Vivien Cabannes and Stefano Vigogna, How many samples are needed to leverage smoothness? In NeurIPS, 2023.
Vivien Cabannes and Stefano Vigogna, A case of exponential convergence rates for SVM. In AISTATS, 2023.
Vivien Cabannes, Alessandro Rudi, and Francis Bach, Fast rates in structured prediction. In COLT, 2021.

Weakly Supervised Learning

Vivien Cabannes, From weakly supervised learning to active labeling. PhD thesis at Ecole Normale Supérieure, 2022.
Vivien Cabannes, Loucas Pillaud-Vivien, Francis Bach, and Alessandro Rudi, Overcoming the curse of dimensionnality with Laplacian regularization in semi-supervised learning. In NeurIPS, 2021.
Vivien Cabannes, Francis Bach, and Alessandro Rudi, Disambiguation of weak supervision with exponential convergence rates. In ICML, 2021.
Vivien Cabannes, Alessandro Rudi, and Francis Bach, Structured prediction with partial labelling through the infimum loss. In ICML, 2020.

Sampling

Vivien Cabannes, and Charles Arnal, Touring sampling with pushforward maps. In ICASSP (Best Paper Award for Industry), 2024.

Art with AI

Vivien Cabannes*, Thomas Kerdreux*, and Louis Thiry, Diptychs of human and machine perceptions. In NeurIPS Creativity Workshop, 2020.
Vivien Cabannes*, Thomas Kerdreux*, Louis Thiry*, and Tina & Charly*, Dialog on a canvas with a machine. In NeurIPS Creativity Workshop, 2019.

Technology and Society

Vivien Cabannes, Will the digital future be embodied? In Revue Esprit, 2022.

Curriculum Vitae

2022-2024: Postdoctoral researcher at Meta, supervised by Léon Bottou
2019-2022: PhD in Machine Learning, supervised by Francis Bach and Alessandro Rudi
2018: Year as a quant at Point72, supervised by Cyril Deremble
2017: Master in machine learning (MVA)
2014: Entrance at the ENS in math
2012: Baccalauréat
PDF version