Lune

Profiles2020-2024

Kwangjun Ahn

MIT EECS/LIDS

12Papers
0Last author
20Unique co-authors
9First author
0Solo

First and last author come from byline position, which some venues assign alphabetically.

Recent papers12

  1. Linear attention is (maybe) all you need (to understand Transformer optimization)Author 1 of 6ICLR 2024 · 92 citations
  2. Understanding Adam Optimizer via Online Learning of Updates: Adam is FTRL in DisguiseAuthor 1 of 4ICML 2024 · 25 citations
  3. How to Escape Sharp Minima with Random PerturbationsAuthor 1 of 3ICML 2024 · 17 citations
  4. Transformers learn to implement preconditioned gradient descent for in-context learningAuthor 1 of 4NeurIPS 2023 · 324 citations
  5. The Crucial Role of Normalization in Sharpness-Aware MinimizationAuthor 2 of 6NeurIPS 2023 · 36 citations
  6. Learning threshold neurons via edge of stabilityAuthor 1 of 6NeurIPS 2023 · 5 citations
  7. Understanding the unstable convergence of gradient descentAuthor 1 of 3ICML 2022 · 89 citations
  8. Mirror Descent Maximizes Generalized Margin and Can Be Implemented EfficientlyAuthor 2 of 4NeurIPS 2022 · 33 citations
  9. Reproducibility in Optimization: Theoretical Framework and LimitsAuthor 1 of 6NeurIPS 2022 · 32 citations
  10. Agnostic Learnability of Halfspaces via Logistic LossAuthor 2 of 5ICML 2022 · 3 citations
  11. Efficient constrained sampling via the mirror-Langevin algorithmAuthor 1 of 2NeurIPS 2021 · 77 citations
  12. SGD with shuffling: optimal rates without component convexity and large epoch requirementsAuthor 1 of 2NeurIPS 2020 · 83 citations
Dusk over the sea between two cliffs drawn in fine vertical lines