Taesung Kwon

Welcome!

I am a Postdoctoral Researcher at KAIST AI. Currently, I am leading research on efficient generative modeling supported by the Sejong Science Fellowship (KRW 500M / USD 350K). I received my Ph.D., M.S., and B.S. in the Dept. of Bio and Brain Engineering from KAIST. During my Ph.D., I was co-advised by Prof. Jong Chul Ye and Prof. Mooseok Jang, and I completed my M.S. advised by Prof. Jong Chul Ye.

Experience

• KAIST AI, InnoCORE Postdoctoral Researcher
  Advisor: Prof. Jong Chul Ye
  Mar 2026 - Present, Seoul, Korea
• Disney Research, ML Group, Research Intern
  Mentor: Vinicius Azevedo
  May 2025 - Aug 2025, Zürich, Switzerland
• ETH Zürich, Dept. of Computer Science, Visiting Researcher
  Joint research program with Disney Research
  May 2025 - Aug 2025, Zürich, Switzerland

Research Interests

My research aims to advance efficient generative modeling for sustainable spatial intelligence. I pursue this through three core directions:

• Efficient Generative Modeling: Designing efficient and scalable frameworks for generative models.
• Generative Priors as Spatial Intelligence: Leveraging generative models to solve visual and scientific problems (e.g., imaging through scattering media).
• Generative Policies for Physical AI: Making vision-language-action and world-action models fast and reliable enough for real-time robot manipulation.

Grants

• Sejong Science Fellowship | 2026 – 2031
  Principal Investigator (PI)
  KRW 500,000,000 (~ USD 350,000)

News

• [2026.09] Our paper (AVIS) has been accepted to NeurIPS 2026.
• [2026.05] Our paper (AGSM) has been accepted to ICML 2026 as a Spotlight paper.
• [2026.03] Our two papers (FCDM - main, Zero4D - findings) have been accepted to CVPR 2026.
• [2026.03] I joined KAIST AI as a Postdoctoral Researcher.
• [2026.02] I secured a 5-year research grant as a PI (Sejong Science Fellowship).
• [2025.12] Our paper (VDPS) has been published in IEEE TPAMI (IF: 20.4).
• [2025.06] Our paper (VISION-XL) has been accepted to ICCV 2025.
• [2025.05] I joined Disney Research as a Research Intern.
• [2025.01] Our two papers (SVI, ViBiDSampler) have been accepted to ICLR 2025.

Publications

Patents

• Image Generation Using Convolution Diffusion Models
  U.S. Patent App. No. 19/449,217, 2026

• Method and Apparatus for Video Restoration Beyond Dynamic Scattering Media Using Video Diffusion Posterior Sampling
  U.S. Patent App. No. 19/815,604, 2026
  KR Patent App. No. 10-2026-0031241, 2026

• Method and Apparatus for High-Resolution Video Restoration Using Latent Image Diffusion Models
  PCT App. No. PCT/KR2026/007929, 2026
  KR Patent App. No. 10-2026-0049062, 2026

• Method and Apparatus for Accelerating Video Restoration Using Autoregressive Diffusion Models
  KR Patent App. No. 10-2026-0161283, 2026

• Method and Apparatus for Video Generation Based on On-Manifold Steering
  KR Patent App. No. 10-2026-0159951, 2026

• Method and Apparatus for Generating Intermediate Video Frames via Bidirectional Sampling
  KR Patent App. No. 10-2025-0114505, 2025

• Method and Apparatus for Solving Video Inverse Problem using Image Diffusion Models
  KR Patent App. No. 10-2025-0171861, 2025

• Method and Apparatus for Low-Dose X-Ray Computed Tomography Image Processing Based on Efficient Unsupervised Learning Using Invertible Neural Network
  U.S. Patent No. 12,346,997, 2025
  KR Patent No. 10-2643601, 2024

Services

• Conference Reviewers: CVPR, ICCV, ECCV, NeurIPS, ICLR, ICML, ACM MM
• Journal Reviewers: IEEE TPAMI, IEEE TCI

Media

• Featured on YTN News: TPAMI research on zero-shot video restoration, 2025
  YouTube

Powered by Jekyll and Minimal Light theme.