Will Hwang

Hi! I'm a Ph.D. candidate in Computer Science at Princeton University, where I am advised by Prof. Olga Russakovsky. My research interests include test-time training, efficient attention mechanisms such as linear attention, and long-context reasoning. I received my B.S. and M.S. from UChicago.

Email  /  CV  /  Scholar  /  X  /  LinkedIn

profile photo

Activities

  • 10-2026:  MERLIN (Multi-State Linear Attention with Generalized Second-Order Routing) preprint is posted on arXiv.
  • 09-2026:  Our work on FinnGen has been published in Nature.
  • 09-2026:  ReFINE was accepted as Spotlight at NeurIPS 2026!
  • 04-2026:  ReFINE received Best Paper Award at the ICLR 2026 TTU Workshop!
  • 02-2026:  ReFINE (Reinforced Fast Weights with Next-Sequence Prediction) preprint is posted on arXiv.
  • 01-2026:  COMPACT has been accepted to ICLR 2026.
  • 11-2025:  Our work on FinnGen is posted on medRXiv.
  • 10-2025:  Our work on NRIP1 R448G has been accepted to PNAS.
  • 08-2025:  I will be serving as an instructor for Princeton AI4ALL
  • 06-2025:  We are excited to present COMPACT at various workshops at CVPR 2025!
  • 04-2025:  COMPACT preprint is posted on arXiv.
  • 04-2025:  Our work on doubly exponential learning is accepted to CogSci 2025.
  • 08-2024:  Starting my Ph.D at Princeton.
  • 10-2023:  Our work on FinnGen single-cell multiome and immunophenotyping has been presented at ASHG 2023.
  • 07-2022:  Starting as a Computational Associate at the Broad Institute of MIT and Harvard.

Research

merlin MERLIN: Multi-State Linear Attention with Generalized Second-Order Routing
Hee Seung Hwang, Sanghyuk Chun, Xindi Wu, Olga Russakovsky, Zhiwei Deng
preprint

We route over multiple linear attention states using the mean and variance of each segment's key alignment with the query, improving long-context performance.

refine ReFINE: Reinforced Fast Weights with Next-Sequence Prediction
Hee Seung Hwang*, Xindi Wu*, Sanghyuk Chun, Olga Russakovsky, Zhiwei Deng
ICLR 2026 Workshop on Test-Time Updates, Best Paper Award
NeurIPS 2026 Spotlight (1.3%)

We present next-sequence prediction as a training objective for improving long-context modeling of fast weight architectures.

compact COMPACT: COMPositional Atomic-to-Complex Visual Capability Tuning
Xindi Wu*, Hee Seung Hwang*, Polina Kirichenko, Olga Russakovsky
ICLR, 2026

We present a compositional, complexity-aware VIT data recipe to improve visual reasoning in MLLMs.

doubly Learning a Doubly-Exponential Number of Concepts From Few Examples
Ilia Sucholutsky, Bonan Zhao, Hee Seung Hwang, Allison Chen, Olga Russakovsky, Tom Griffiths
CogSci, 2025

We present a novel minimal paradigm that explores how efficiently people can simultaneously learn visual and symbolic concepts.

finngen Population-scale immune multiome atlas reveals regulatory disease mechanisms
Masahiro Kanai, Toni M Delorey, Jarno Honkanen, Rodosthenis S Rodosthenous, Julianna Juvila, Shane Murphy, Isabella Teixeira-Soldano, Hee Seung Hwang, Juha Karjalainen, Jussi Halonen, Georgia Panagiotaropoulou, Yuanxiang Zhang, Cristin McCabe, Eric Chen, Kosaku Nanki, Toshimi Yoshida, Kai Liu, Marla Glean, Nitya Mehrotra, Emily P Finan, Daniel Chafamo, Yixiao Zhu, Mikko Arvas, Sanni Ruotsalainen, Zhili Zheng, Yoshihiko Tomofuji, Daiki Tokuyasu, Shinichi Namba, Kyuto Sonehara, BioBank Japan Project, FinnGen, Yukinori Okada, Mary P Reeve, Mitja Kurki, Caroline BM Porter, Orr Ashenberg, Wei Zhou, Kimmo Pitkänen, Jukka Partanen, Aarno Palotie, Daniel B Graham, Mark J Daly, Ramnik J. Xavier
Nature, 2026

We show that population-scale simultaneous profiling of chromatin accessibility and gene expression across immune cell types reveals multi-layered regulatory pathways connecting genetic variants to disease.

pnas Nuclear receptor coregulator NRIP1 R448G modulates T cell gut homing to control intestinal inflammation
Xiangjun Chen, Hee Seung Hwang, Bihua Li, Yanhua Zhao, Koushik Ghosh, Lei Deng, Elizabeth A. Creasey, Orr Ashenberg, Daniel B. Graham, Ramnik J. Xavier
PNAS, 122, 2025

We show that NRIP1 R448G leads to intestinal inflammation by activating T cell gut homing and inflammatory cytokine production.

Teaching

  • AI4ALL Instructor:  Princeton University, July 2025
  • Teaching Assistant:
    • Machine Learning and Pattern Recognition (ECE535), Fall 2026
    • Clinical AI (COS480), Spring 2026
    • Computer Vision (CS429), Fall 2025
  • Mentorship:  I am continuing to mentor students I've met along the way. Feel free to reach out!

Credit: Jon Baron.