Jongha Jon Ryu

I am a Postdoctoral Associate at MIT EECS/RLE, hosted by Gregory W. Wornell.
My research develops the mathematical and statistical foundations of scientific machine learning, with the aim of enabling scalable and reliable methods for scientific inference and modeling. I translate these foundations into tools for scientific discovery and large-scale engineering systems by designing algorithms for operator learning, generative modeling, and uncertainty quantification that scale to high-dimensional scientific problems.

research focus

• neural spectral methods for scalable operator learning
  • NestedLoRA for compact operators [ICML’24a]
  • NestedOMM for positive definite operators [NeurIPS’25b]
  • Koopman analysis via NestedLoRA for dynamical systems [NeurIPS’25a]
• principled methods for probabilistic and generative modeling
  • Score-of-Mixture generative modeling framework [ICML’25a, Spotlight]
  • unified NCE-based lens for energy-based models [ICML’25b]
  • corrected InfoNCE for density-ratio estimation [arXiv’25]
• reliable techniques for uncertainty quantification
  • time-uniform confidence sets for:
    • bounded random variables [TransIT’24]
    • bounded random vectors [ICML’24b, Spotlight]
    • nonnegative random variables [COLT’25] (applied to off-policy contextual bandits)
  • diagnosis of evidential deep learning [NeurIPS’24]

background

Before MIT, I received my Ph.D. in Electrical Engineering from UC San Diego, advised by Young-Han Kim and Sanjoy Dasgupta, supported by the Kwanjeong Educational Foundation. I hold dual B.S. degrees in Electrical and Computer Engineering and Mathematical Sciences, with a minor in Physics, from Seoul National University, graduating with the highest distinction.

[CV] (last updated: 11/16/2025)