Soobin Um

About Me

I am a Ph.D. candidate at the Kim Jaechul Graduate School of AI, KAIST, advised by Prof. Jong Chul Ye. My research focuses on generative modeling, with a special interest in robust and inclusive AI. I also explore the intersection of generative models and physics, particularly in photonic structures. I previously worked as a senior researcher at Agency for Defense Development (ADD) for 7.5 years, specializing in wireless communication systems and defense applications.

News

• 2025.07 – Gave an invited talk on MinorityPrompt and Boost-and-Skip at the Visual AI Lab, Princeton University.
• 2025.07 – Delivered a joint invited talk on diffusion-based inverse design at The McMahon Lab, Cornell University (with Dongjin Seo).
• 2025.05 – Our recent project, Boost-and-Skip, has been accepted to ICML 2025.
• 2025.04 – MinorityPrompt is selected as oral presentation (top 0.74%) at CVPR 2025.
• 2025.02 – Our paper, MinorityPrompt, is accepted to CVPR 2025.

Publications

Conference Papers

• Boost-and-Skip: A Simple Guidance-Free Diffusion for Minority Generation
  Soobin Um*, Beomsu Kim*, Jong Chul Ye
  ICML 2025
• Minority-Focused Text-to-Image Generation via Prompt Optimization
  Soobin Um, Jong Chul Ye
  CVPR 2025 (Oral presentation, top 0.74%)
• Physics-guided Optimization of Photonic Structures using Denoising Diffusion Probabilistic Models
  Dongjin Seo*, Soobin Um*, Sangbin Lee, Jong Chul Ye, Haejun Chung
  NeurIPS 2024 Workshop (ML4PS)
• Self-Guided Generation of Minority Samples Using Diffusion Models
  Soobin Um, Jong Chul Ye
  ECCV 2024
• Don't Play Favorites: Minority Guidance for Diffusion Models
  Soobin Um, Suhyeon Lee, Jong Chul Ye
  ICLR 2024
• A Fair Generative Model Using LeCam Divergence
  Soobin Um, Changho Suh
  AAAI 2023 (Oral presentation)

Preprints

• Physics-guided and fabrication-aware inverse design of photonic devices using diffusion models
  Dongjin Seo*, Soobin Um*, Sangbin Lee, Jong Chul Ye, Haejun Chung
  Submitted to ACS Photonics