Hyunmin Cho

Hi! I'm a integrated MS&Ph.D student at Korea University, Image Processing Algorithm Lab. advised by Prof. Kyong Hwan Jin.

My ultimate goal is to bridge the gap between human imagination and digital reality, creating a world that is seamlessly modifiable according to user intention. To that end, my research focuses on Generative Models and Controllable Generation, with an emphasis on understanding the internal dynamics of diffusion models and developing principled, training-free methods that steer sampling toward more faithful, diverse, and hallucination-resistant outputs.

Feel free to send me an e-mail if you want to have a chat!
Contact: hyun_cho@korea.ac.kr

	Balancing Fidelity and Diversity in Diffusion Models via Symmetric Attention Decomposition: Hopfield Perspective Hyunmin Cho, Woo Kyoung Han, and Kyong Hwan Jin^† 43rd International Conference on Machine Learning (ICML), 2026. Paper / Website / Code We characterize the pre-softmax attention matrix QK in transformers as an associative memory matrix encoding pairwise associations between input features.
	TAG: Tangential Amplifying Guidance for Hallucination-Resistant Sampling Hyunmin Cho^, Donghoon Ahn^, Susung Hong^, Jee Eun Kim, Seungryong Kim^†, and Kyong Hwan Jin^† (denotes equal contributions) 43rd International Conference on Machine Learning (ICML), 2026. Paper / Website / Code We introduce TAG, a theoretically grounded, training-free, computationally lightweight, and architecture-agnostic guidance method that operates solely on trajectory signals without modifying the underlying diffusion model.
	Reference-based Super-Resolution via Image-based Retrieval-Augmented Generation Diffusion Byeonghun Lee^, Hyunmin Cho^, Hong Gyu Choi, Soo Min Kang, Iljun Ahn, and Kyong Hwan Jin^† (denotes equal contributions) 20th International Conference on Computer Vision (ICCV), 2025. Paper / Website / Code We propose an image-based RAG framework (iRAG)* for realistic super-resolution, which employs a trainable hashing function to retrieve either real-world or generated references given an LR query.

	Integrated M.S & Ph.D in Electrical Engineering \| Korea University Mar 2024 - cont. Research: Generative Model & Hopfield Model Advisor: Prof. Kyong Hwan Jin
	B.S in Software \| Gachon University Mar 2020 - Feb 2024 Research: Image Super Resolution Advisor: Prof. Kiho Choi (currently in Kyung Hee University)

Gold Prize, 38th Workshop of Image Processing and Image Understanding (IPIU), 2026

Samsung Research, Next Generation Display Lab, (confidential contents), 2024.06-2025.05

Samsung Research, Next Generation Display Lab, (confidential contents), 2026.01-2026.12

Method and Apparatus for Lossless Implicit Neural Representation Based on Bipolar Vector Labeling and Recursive Single Weight Operation
Hyunmin Cho, Kyong Hwan Jin, Yongjun Lee, Jiwon Kim, Woo Kyoung Han
Korea Patent Application No. 10-2025-0150878

Website templates from Jon Barron.