Jung Yeon Park

Jung Yeon Park

Ph.D. Student

Northeastern University

Biography

Research intern @ The AI Institute, PhD candidate @ Northeastern University.

I am a Ph.D. candidate at the GRAIL Lab and Geometric Learning Lab at Northeastern University, advised by Lawson Wong and Robin Walters. My research interests are in imitation learning and symmetry. I aim to apply strong inductive biases like symmetry or expert demonstrations to learning algorithms so that they can be more efficient and generalize. I’ve also worked on making imitation learning more robust with learned dynamics models, predicting complex dynamics over meshes, and on multiresolution tensor models for spatial analysis.

Previously, I received my B.S. and M.S. at KAIST and researched prediction models for complex semiconductor tools. Before coming to Northeastern, I worked as a software engineer and systems administrator at Samsung Electronics.

Interests
  • Symmetry and Equivariance
  • Imitation Learning
  • Representation Learning
Education
  • Ph.D. in Computer Science, 2019 ~ present

    Northeastern University

  • M.S. in Computer Science, 2022

    Northeastern University

  • M.S. in Industrial Systems Engineering, 2016

    KAIST

  • B.S. in Industrial Systems Engineering, 2014

    KAIST

Recent Publications

Symmetric Models for Radar Response Modeling. In NeurReps, 2023.
A General Theory of Correct, Incorrect, and Extrinsic Equivariance. In NeurIPS, 2023.
The Surprising Effectiveness of Equivariant Models in Domains with Latent Symmetry. In ICLR, (notable-top-25%), 2023.
Learning Symmetric Embedding Networks for Equivariant World Models. In ICML, 2022.
Generator Surgery for Compressed Sensing. In NeurIPS Deep Inverse Workshop, 2020.
Multiresolution Tensor Learning for Efficient and Interpretable Spatial Analysis. In ICML, 2020.
On the Consequences of Un-Modeled Dynamics to the Optimality of Schedules in Clustered Photolithography Tools. In WSC, 2019.
Exit Recursion Models of Clustered Photolithography Tools for Fab Level Simulation. South Korea Patent Office, 1018856190000, 2018.

Contact