Jimin Heo

photo_me.jpeg

I am a Ph.D. student at UC Irvine, advised by Erik Sudderth. My research centers on generative models, diffusion, and variational inference, with a particular interest in adapting pre-trained diffusion models to inverse problems where observations are partial, noisy, or otherwise degraded. I’m also interested in how generative models interact with people particularly in accessibility and assistive contexts.

Before starting my Ph.D., I completed a Master of Data Science at UC Irvine. Earlier, I spent several years as a software / DevOps engineer in industry, most recently at Toss Payments, and previously at SK Inc. and Gurum Networks.

Publications

  1. CHI
    “It’s trained by non-disabled people”: Evaluating How Image Quality Affects Product Captioning with Vision-Language Models
    Kapil Garg, Xinru Tang, Jimin Heo, Dwayne R. Morgan, Darren Gergle, Erik B. Sudderth, and Anne Marie Piper
    In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI), 2026
    Honorable Mention for Best Paper (Top 5%) arXiv
  2. AISTATS
    VIPaint: Image Inpainting with Pre-Trained Diffusion Models via Variational Inference
    Sakshi Agarwal, Gabriel Hope, Jimin Heo, and Erik B. Sudderth
    In Proceedings of the 29th International Conference on Artificial Intelligence and Statistics (AISTATS), 2026
    arXiv Poster Website