Dayal Singh Kalra

4 papers · 2023–2025 · 3 conferences · across top CS/AI conferences

Achievements

🌍 Conference Polyglot (3) 🌉 Interdisciplinary Bridge 🐣 Hot Topic Early Bird 🐝 Cross-Pollinator (15) ❓ The Questioner (2)

Conferences

NIPS (2) ICLR (1) ICML (1)

Top co-authors

Maissam Barkeshli (4) Tianyu He (2) Darshil Doshi (1) Tao Tao (1)

Keywords

stochastic gradient descent (2) neural network optimization (1) loss landscape (1) hyperparameter tuning (1) learning rate (1) adam optimizer (1) hessian eigenvalue (1) training dynamics (1) learning rate warmup (1) edge of stability (1) neural network (1) phase diagram (1)

Papers

Universal Sharpness Dynamics in Neural Network Training: Fixed Point Analysis, Edge of Stability, and Route to Chaos ICLR 2025

(How) Can Transformers Predict Pseudo-Random Numbers? ICML 2025

Why Warmup the Learning Rate? Underlying Mechanisms and Improvements NIPS 2024

Phase diagram of early training dynamics in deep neural networks: effect of the learning rate, depth, and width NIPS 2023