phenomenon · first seen 5 Oct 2026
edge of stability
Also: EoS, edge of stability phenomenon
Training regime where sharpness hovers near the threshold 2/learning rate under gradient descent with large step sizes.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Mind the Drift: Diagonal Linear Networks Under Large Learning RatesPapers · 5 Oct 2026 · Aniket Sanyal, Tom Jacobs, Rebekka Burkholz
- A Solvable Model of Adaptive Learning Rate Rescaling: Acceleration, Stability & ScalingPapers · 5 Oct 2026 · Itay Lavie, Clarissa Lauditi, Cengiz Pehlevan
- Non-Euclidean Gradient Descent Has Many Edges of StabilityPapers · 10 Oct 2026 · Peter Zaika, Zakhar Shumaylov
- Measuring the local shape of learning-rate transfer curvesPapers · 15 Oct 2026 · Satoki Ishikawa, Rio Yokota, Ryo Karakida
- A Minimal ReLU Classifier on Non-Linearly Separable Data: Convergence and Implicit Bias at and beyond the Edge of StabilityPapers · 16 Oct 2026 · Quan Nguyen, Robin Yadav, Param Mody, Mark Schmidt
- Muon Operates Beyond the Edge of StabilityPapers · 16 Oct 2026 · Niko Covic, Baraq Lipshitz, Alexander Padula, Stefan Visnjic
Latest sources
- High-dimensional limit theorems for adaptive variants of SGD I: Ballistic limitsPapers · 14 Nov 2026 · Aukosh Jagannath, Varnan Sarangian
- Classification with Large-Step Gradient Descent: Unifying Self-Stabilization and Convergence RatesPapers · 12 Nov 2026 · Salma Tarmoun, Lachlan Ewen MacDonald, Konstantinos Emmanouilidis, Ziqing Xu, Re…
- Characterizing the Edge of Stability in Variational Training Without PriorsPapers · 11 Nov 2026 · Juraj Marusic, Jonathan Wenger, Beau Coker, John P. Cunningham
- Adversarial Training with Large Step Sizes: Implicit Bias, Optimization, and SharpnessPapers · 5 Nov 2026 · Yi Feng, Andrea Paudice
- Unifying Central Flow, Rod Flow, and the Edge CouplingPapers · 5 Nov 2026 · Eric Regis, Sinho Chewi
- Symmetry Breaking at the Edge of Stability via Central Flow AnalysisPapers · 4 Nov 2026 · Naima Elosegui Borras, Georgios Arvanitidis
- A Self-Generated Floor on Adam's $\varepsilon$: Evidence from Power-Law Matrix SensingPapers · 1 Nov 2026 · Lngel Kyo
- Stochastic learning flow: moment equations for stochastic gradient descent at finite step sizePapers · 30 Oct 2026 · Christian Schmid, James M Murray
- Does LLM Pre-Training Typically Occur at the Edge of Stability?Papers · 27 Oct 2026 · Yuhang Cai, Haodong Wen, Haofeng Huang, Deyi Liu, Yiyuan Ma, Kaifeng Lyu
- Physics-Inspired Optimizers beyond the Edge of Stability: Period-Two Orbits and Delayed InstabilityPapers · 19 Oct 2026 · Alexander Itin
- Muon Operates Beyond the Edge of StabilityPapers · 16 Oct 2026 · Niko Covic, Baraq Lipshitz, Alexander Padula, Stefan Visnjic
- A Minimal ReLU Classifier on Non-Linearly Separable Data: Convergence and Implicit Bias at and beyond the Edge of StabilityPapers · 16 Oct 2026 · Quan Nguyen, Robin Yadav, Param Mody, Mark Schmidt