Profiles2020-2024
Nicolas Flammarion
EPFL
12Papers
11Last author
15Unique co-authors
0First author
0Solo
First and last author come from byline position, which some venues assign alphabetically.
Recent papers12
- Why Do We Need Weight Decay in Modern Deep Learning?Author 4 of 4NeurIPS 2024 · 101 citations
- First-order ANIL provably learns representations despite overparametrisationAuthor 3 of 3ICLR 2024 · 7 citations
- SGD vs GD: Rank Deficiency in Linear NetworksAuthor 3 of 3NeurIPS 2024 · 5 citations
- Saddle-to-Saddle Dynamics in Diagonal Linear NetworksAuthor 2 of 2NeurIPS 2023 · 68 citations
- Sharpness-Aware Minimization Leads to Low-Rank FeaturesAuthor 4 of 4NeurIPS 2023 · 48 citations
- On the spectral bias of two-layer linear networksAuthor 4 of 4NeurIPS 2023 · 27 citations
- Penalising the biases in norm regularisation enforces sparsityAuthor 2 of 2NeurIPS 2023 · 21 citations
- Gradient flow dynamics of shallow ReLU networks for square loss and orthogonal inputsAuthor 2 of 2NeurIPS 2022 · 92 citations
- Implicit Bias of SGD for Diagonal Linear Networks: a Provable Benefit of StochasticityAuthor 2 of 2NeurIPS 2021 · 135 citations
- Last iterate convergence of SGD for Least-Squares in the Interpolation regimeAuthor 3 of 3NeurIPS 2021 · 52 citations
- Sequential Algorithms for Testing Closeness of DistributionsAuthor 2 of 4NeurIPS 2021 · 4 citations
- Understanding and Improving Fast Adversarial TrainingAuthor 2 of 2NeurIPS 2020 · 366 citations
