Jongha Jon Ryu

Postdoctoral Associate at MIT

jon.jpg

Room 36-677

50 Vassar St

Cambridge, MA 02139

I am currently a postdoctoral associate at MIT, hosted by Gregory W. Wornell. Prior to joining MIT, I received my Ph.D. in Electrical Engineering from UC San Diego, where I was fortunate to be advised by Young-Han Kim and Sanjoy Dasgupta. My graduate study was generously supported by the Kwanjeong Educational Foundation. Before the graduate study, I received my B.S. in Electrical and Computer Engineering and B.S. in Mathematical Sciences (with minor in Physics) with the highest distinction from Seoul National University in 2015.

In general, I aim to develop principled and practical algorithms for machine learning and data science. My recent research topics include:

  • Designing algorithms for large-scale problems from first principles
  • Learning with efficient & reliable uncertainty quantification
  • Information-theoretic tools for machine learning and data science
  • Unifying principles in machine learning
    • unified view on density functional estimation with fixed-k-NNs [TIT2022]
    • unifying evidential deep learning methods for uncertainty quantification [NeurIPS2024]
    • unifying principles for fitting unnormalized distributions via noise-contrastive estimation [arXiv]

As an information theorist by training, I enjoy doing research by simplifying intricate ideas, unifying concepts, and generalizing them to address complex problems.

Check out my resume for more information.

news

Oct 23, 2024 In this fall, I have given talks on NeuralSVD at MERL, KAIST, KIAS, and Flatiron Institute.
Sep 25, 2024 Our paper on demystifying the sucess of evidential deep learning methods got accepted at NeurIPS 2024!
Sep 06, 2024 I have posted a substantially revised version of the arXiv preprint on minimax optimal learning with fixed-k-nearest neighbors, now including new results on density estimation.
Aug 21, 2024 One paper on gambling-based confidence sequences has been accepted at IEEE Transactions on Information Theory!
Jun 17, 2024 One paper on new techniques for better score estimation accepted at ICML 2024 Workshop on Structured Probabilistic Inference & Generative Modeling.

Selected publications

  1. TIT
    Nearest neighbor density functional estimation from inverse Laplace transform
    IEEE Trans. Inf. Theory, February 2022
  2. TIT
    On Confidence Sequences for Bounded Random Processes via Universal Gambling Strategies
    J. Jon Ryu, and Alankrita Bhatt
    IEEE Trans. Inf. Theory, February 2024
  3. AISTATS
    Parameter-Free Online Linear Optimization with Side Information via Universal Coin Betting
    J. Jon RyuAlankrita Bhatt, and Young-Han Kim
    In Proc. Int. Conf. Artif. Int. Statist. (AISTATS) , March 2022
  4. ICML
    Operator SVD with Neural Networks via Nested Low-Rank Approximation
    J. Jon RyuXiangxiang Xu, H. S. Melichan Erol, Yuheng BuLizhong Zheng, and Gregory W. Wornell
    In Proc. Int. Conf. Mach. Learn. (ICML) , July 2024
  5. ICML
    Gambling-Based Confidence Sequences for Bounded Random Vectors
    J. Jon Ryu, and Gregory W. Wornell
    In Proc. Int. Conf. Mach. Learn. (ICML) , July 2024
    Spotlight (top 3.5%)
  6. arXiv
    Learning with Succinct Common Representation with Wyner’s Common Information
    July 2022
    Submitted. A preliminary version of this manuscript was presented at the Bayesian Deep Learning Workshop at NeurIPS 2018, and an abridged version of the current manuscript was presented at the Bayesian Deep Learning workshop at NeurIPS 2021.