method · first seen 8 Oct 2026
gradient flow
The continuous-time limit of gradient descent used to analyze how neural network training dynamics evolve.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Rare Gate Disagreements Can Limit Plasticity: When Gradient Flow Mispredicts Finite-Batch SGDPapers · 8 Oct 2026 · Ruoyu Zhao, Mingxuan Zhang, Jianbo Dai, Jiaqi Wu, Chenyu Zhu, Tong Che
- A "feature ODE" describing the learning behavior of shallow MLPs on simple functionsPapers · 19 Oct 2026 · Joseph Turnbull, Berkan Ottlik, Jamie Simon
- When Parameter Gradient Flows Close in Representation SpacePapers · 20 Oct 2026 · Shengye Tao
- Gradient Descent on Two ReLU Neurons: Global Landscape and Bifurcation DynamicsPapers · 30 Oct 2026 · Binghua Li, Mengzhe Li, Denny Wu, Tianhao Wang
- Gradient Flows over Convex Bodies for Learning Optimal RegularizersPapers · 12 Nov 2026 · Nilava Metya, Oscar Leong, Yong Sheng Soh
Latest sources
- Gradient Flows over Convex Bodies for Learning Optimal RegularizersPapers · 12 Nov 2026 · Nilava Metya, Oscar Leong, Yong Sheng Soh
- Gradient Descent on Two ReLU Neurons: Global Landscape and Bifurcation DynamicsPapers · 30 Oct 2026 · Binghua Li, Mengzhe Li, Denny Wu, Tianhao Wang
- When Parameter Gradient Flows Close in Representation SpacePapers · 20 Oct 2026 · Shengye Tao
- A "feature ODE" describing the learning behavior of shallow MLPs on simple functionsPapers · 19 Oct 2026 · Joseph Turnbull, Berkan Ottlik, Jamie Simon
- Rare Gate Disagreements Can Limit Plasticity: When Gradient Flow Mispredicts Finite-Batch SGDPapers · 8 Oct 2026 · Ruoyu Zhao, Mingxuan Zhang, Jianbo Dai, Jiaqi Wu, Chenyu Zhu, Tong Che