Publications

You can also find my articles on my Google Scholar profile.

Conference Papers

Geometry-Correct Diffusion Posterior Sampling with Denoiser-Pullback Curvature Guidance and Manifold-Aligned Damping

Seunghyeok Shin *, Minwoo Kim *, Dabin Kim, Hongki Lim

Published in The Forty-Third International Conference on Machine Learning (ICML), 2026

CLAMP is a diffusion posterior sampler for inverse problems that replaces hand-tuned scalar likelihood guidance with geometry-aware, per-noise-level damped Gauss–Newton corrections in diffusion-state coordinates. It pulls likelihood sensitivity back through the denoiser, uses a one-sided curvature model with manifold-aligned rank-one damping, solves each correction with matrix-free GMRES, and advances sampling via a variance-preserving Langevin transition. OpenReview · PDF

Download Paper

FAST-DIPS: Adjoint-Free Analytic Steps and Hard-Constrained Likelihood Correction for Diffusion-Prior Inverse Problems

Minwoo Kim *, Seunghyeok Shin *, Hongki Lim

Published in The Fourteenth International Conference on Learning Representations (ICLR), 2026

FAST-DIPS is a training-free diffusion-prior inverse-problem solver that enforces a hard measurement-space feasibility constraint via an adjoint-free ADMM correction with an analytic (or forward-difference) step size, plus decoupled re-annealing. It supports linear/nonlinear forward operators without hand-coded adjoints and includes pixel, latent, and hybrid pixel→latent variants for faster, stable reconstructions. OpenReview · PDF · Code

Download Paper

Adaptive 3D Reconstruction via Diffusion Priors and Forward Curvature-Matching Likelihood Updates

Seunghyeok Shin, Dabin Kim, Hongki Lim

Published in The Thirty-Ninth Annual Conference on Neural Information Processing Systems(NeurIPS), Spotlight, 2025

FCM is a training-free, likelihood-guided diffusion update that auto-tunes step sizes via forward-mode autodiff and curvature probes, enabling fast, stable 3D point-cloud reconstruction from single/multi-view inputs. OpenReview · Project page · Code

Download Paper