Seunghun Kang (강승훈)

Ph.D. Student, Department of Artificial Intelligence
Ajou University (아주대), Suwon, Korea
seunghun.kang at ajou.ac.kr
Google Scholar · GitHub · LinkedIn

About

I am Seunghun Kang, a Ph.D. student in the Department of Artificial Intelligence at Ajou University in Suwon, Korea, where I am advised by Prof. Jongbin Ryu in the Multi-Modal Artificial Intelligence Lab. I joined the lab in September 2021 as an undergraduate research intern and moved into the doctoral program in September 2023.

I am interested in building AI models that work efficiently and reliably in the real world, from clinical scans and mobile devices to embodied agents and industrial systems. My work spans four directions:

Education

Research Experience

News (A: award)

Publications (C: conference, J: journal, S: submission or preprint)

Test-time correction that verifies top-2 candidate labels by the divergence shift of a memory bank
[C.2] Seungmin Oh, Seunghun Kang, and Jongbin Ryu. Efficient Test-time Adaptation through Candidate Verification and Divergence Shifts.
Advances in Neural Information Processing Systems (NeurIPS), 2026

Keywords: “test-time adaptation, vision-language models, training-free adaptation, candidate verification, divergence shift, inference efficiency”

Transformation-aware prompts with an augmented contrastive space and adaptive re-calibration of the target
[C.1] Seungmin Oh, Seunghun Kang, and Jongbin Ryu. Re-calibrated Contrastive Loss for Transformation-Aware Prompt Conditioning in Vision-Language Models.
British Machine Vision Conference (BMVC), 2026

Keywords: “vision-language models, prompt learning, parameter-efficient transfer learning, contrastive loss, distribution shift, few-shot learning”

Robust rank pooling compresses sliding windows of a 3D brain MRI into 2D images and selects a robust set
[J.1] Seunghun Kang, Sungeun Kim, Hyun Woong Roh, Chang Hyung Hong, Sang Joon Son, and Jongbin Ryu. Efficient Alzheimer’s Disease Classification with Robust Rank Pooling.
Scientific Reports, 2026

Keywords: “Alzheimer’s disease classification, brain MRI, robust rank pooling, 3D-to-2D compression, dynamic images, efficient inference”

Naive multi-layer model compressed into a merged layer using layer imitation
[S.1] Zihao Ye, …, Seungmin Oh, Hankyul Kang, Seunghun Kang, Jongbin Ryu, et al. Evaluation of Winning Solutions of 2025 Low Power Computer Vision Challenge.
arXiv preprint, 2026

Keywords: “low-power computer vision, model compression, layer merging, layer imitation, mobile NPU”

Research Projects