About
I’m actively looking for PhD opportunities! I recently received my Master’s degree from the Electrical and Computer Engineering department at Carnegie Mellon University. Currently, I am a Research Associate at Carnegie Mellon University.
Research Interests
My research focuses on understanding how neural networks can learn useful representations from data more effectively. I am particularly interested in the basic learning rules behind neural networks, including both neural network architectures and optimization algorithms. My goal is to understand why common design choices work, when they become limiting, and how they can be improved to make learning more efficient and achieve better performance.
My projects study learning at different levels of a neural network, including how information is stored and updated, how it moves through a network, how model parameters are optimized, and how training data guides what a model learns. Rather than treating existing design choices as fixed conventions, I explore when they should be redesigned and when they should adapt to the data and task. In the long term, I am interested in neural networks that can increasingly shape their own learning process from experience, enabling more flexible, continual, and lifelong learning.
Contact
Email: ehong [at] andrew [dot] cmu [dot] edu
News
- 2026.09 — AK-Momentum accepted to NeurIPS 2026.
- 2026.08 — AK-Momentum (prev. DeltaMomentum) paper published on arXiv.
- 2026.06 — CD-RoPE paper accepted to Interspeech 2026.
- 2026.06 — AK-Momentum (prev. DeltaMomentum) paper accepted to ICML 2026 HiLD workshop.
- 2026.06 — Control-Theoretic Account of Muon’s Advantage paper accepted to ICML 2026 HiLD workshop.
- 2026.05 — Submitted two papers to NeurIPS 2026 Main Track.