Minhak Song

5 papers · 2023–2026 · 4 venues · across top CS/AI venues

Authors are matched by name, so researchers who share a name ("Yang Liu", "Wei Wang") appear as a single profile and their paper counts, venues and achievements are combined. How identity is handled.

Achievements

Jump to papers ↓

🌍 Conference Polyglot · Bronze 🌉 Interdisciplinary Bridge · Bronze 🗺️ Taxonomy Completionist (10) ❓ The Questioner

Venues

ICLR (2) ALT (1) ICML (1) NeurIPS (1)

Top co-authors

Chulhee Yun (4) Kwangjun Ahn (2) Boyao Liao (1) Tianyu Pang (1) Suvrit Sra (1) Geonhui Yoo (1) Ali Jadbabaie (1) Yaoqing Yang (1) Zhuoli Ouyang (1) Xiang Cheng (1)

Research topics

Machine Learning (4) Optimization & Theory · Machine Learning (4) Deep Learning (3) Architectures (2) Mathematics & Optimization (1) Mathematics (1) Optimization & Theory · Deep Learning (1) Techniques (1)

Keywords

training dynamics (2) progressive sharpening (2) edge of stability (2) stochastic gradient descent (2) ill-conditioned optimization (1) hessian eigenvalue (1) linear network (1) regression task (1) transformer optimization (1) linear attention (1) gradient alignment (1) network depth (1) step size (1) dataset difficulty (1) hessian spectrum (1) trajectory alignment (1) bifurcation theory (1) two-phase behavior (1) dominant subspace (1) hessian eigenspace (1)

Papers

Suspicious Alignment of SGD:A Fine-Grained Step Size Condition Analysis ALT 2026 Does SGD really happen in tiny subspaces? ICLR 2025 Understanding Sharpness Dynamics in NN Training with a Minimalist Example: The Effects of Dataset Difficulty, Depth, Stochasticity, and More ICML 2025 Linear attention is (maybe) all you need (to understand Transformer optimization) ICLR 2024 Trajectory Alignment: Understanding the Edge of Stability Phenomenon via Bifurcation Theory NeurIPS 2023