Improved bounds for discretization of Langevin diffusions: Near-optimal rates without convexityBernoulli. 2022-08-01. Vol. 28, num. 3, p. 1577-1601. DOI : 10.3150/21-BEJ1343.
Accelerated SGD for Non-Strongly-Convex Least Squares
Trace norm regularization for multi-task learning with scarce data
Gradient flow dynamics of shallow ReLU networks for square loss and orthogonal inputs
Last iterate convergence of SGD for Least-Squares in the Interpolation regime
Is there an analog of Nesterov acceleration for gradient-based MCMC?Bernoulli. 2021-08-01. Vol. 27, num. 3, p. 1942-1992. DOI : 10.3150/20-BEJ1297.
Implicit Bias of SGD for Diagonal Linear Networks: a Provable Benefit of Stochasticity2021-06-16. 35th Conference on Neural Information Processing Systems (NeurIPS 2021), Virtual Conference, December 6-14, 2021.
On the effectiveness of adversarial training against common corruptions
A Continuized View on Nesterov Acceleration
On the Stability of Fine-tuning BERT: Misconceptions, Explanations, and Strong Baselines2021. 9th International Conference on Learning Representations, Virtual, May 4-8, 2021.
RobustBench: a standardized adversarial robustness benchmark
Square Attack: a query-efficient black-box adversarial attack via random search2020-08-28. European Conference on Computer Vision (ECCV 2020), [Online], August 23-28, 2020.
Understanding and Improving Fast Adversarial Training2020-07-06. Advances In Neural Information Processing Systems 33 (NeurIPS 2020), [Online], December 2020.
Sparse-RS: a versatile framework for query-efficient sparse black-box adversarial attacks