Yeho Gwon

I am a M.S. student at POSTECH, South Korea, and I’m fortunate to be advised by Prof. Suha Kwak.

My ultimate goal is to enable machines to understand the world just like humans do. To achieve this, I have been focusing on dataset construction for model training and robustness for real-world applications.

Having served as both a first and second author on research projects, I am open to research collaboration opportunities. If you find my work interesting, please feel free to reach out to me!

Email: yeho.gwon@postech.ac.kr

News

Oct 1, 2025	🏆 I won the 2025 POSTECHIAN Fellowship award. Such a good start of my master’s degree!
Jul 22, 2025	🎉 Our paper on improving robustness of SAM (Segment Anything Model) has been accepted to NeurIPS 2025!
Jul 22, 2025	🎉 Our paper introducing GaRA-SAM has been accepted for an oral presentation at the ICCV 2025 workshop on Building Foundation Models You Can Trust!
Jul 22, 2025	🎉 Our paper on active learning has been accepted to TMLR.
Jun 5, 2025	❄️ Our paper on improving the robustness of SAM has been uploaded as an arxiv preprint.

Education

Sep, 2025 - Present	Pohang University of Science and Technology (POSTECH), Pohang, South Korea M.S. in Computer Science and Engineering
Feb, 2022 - Aug, 2025	Pohang University of Science and Technology (POSTECH), Pohang, South Korea B.S. in Computer Science and Engineering GPA: 4.0/4.3 (Summa Cum Laude)
Jun, 2024 - Aug, 2024	University of California, Berkeley, Berkeley, CA, USA Visiting Summer Sessions Student

Experience

Sep, 2025 - Present	Computer Vision Lab at POSTECH, Pohang, South Korea Research and Teaching Assistant Advisor: Prof. Suha Kwak
Feb, 2023 - Aug, 2025	Computer Vision Lab at POSTECH, Pohang, South Korea Research Intern Advisor: Prof. Suha Kwak Participate in research projects on active learning for visual recognition tasks.

Publications

* indicates equal contribution.

GaRA-SAM: Robustifying Segment Anything Model with Gated-Rank Adaptation

Sohyun Lee, Yeho Gwon, Lukas Hoyer, and Suha Kwak

Conference on Neural Information Processing Systems (NeurIPS), 2025
ICCV 2025 Workshop on Building Foundation Models You Can Trust (Oral)
Abstract Paper Code Project Page

Improving robustness of the Segment Anything Model (SAM) to input degradations is critical for its deployment in high-stakes applications such as autonomous driving and robotics. Our approach to this challenge prioritizes three key aspects: first, parameter efficiency to maintain the inherent generalization capability of SAM; second, fine-grained and input-aware robustification to precisely address the input corruption; and third, adherence to standard training protocols for ease of training. To this end, we propose gated-rank adaptation (GaRA). GaRA introduces lightweight adapters into intermediate layers of the frozen SAM, where each adapter dynamically adjusts the effective rank of its weight matrix based on the input by selectively activating (rank-1) components of the matrix using a learned gating module. This adjustment enables fine-grained and input-aware robustification without compromising the generalization capability of SAM. Our model, GaRA-SAM, significantly outperforms prior work on all robust segmentation benchmarks. In particular, it surpasses the previous best IoU score by up to 21.3%p on ACDC, a challenging real corrupted image dataset.
Enhancing Cost Efficiency in Active Learning with Candidate Set Query

Yeho Gwon*, Sehyun Hwang*, Hoyoung Kim, Jungseul Ok, and Suha Kwak

Transactions on Machine Learning Research (TMLR), 2025

Abstract Paper Code Project Page

This paper introduces a cost-efficient active learning (AL) framework for classification, featuring a novel query design called candidate set query. Unlike traditional AL queries requiring the oracle to examine all possible classes, our method narrows down the set of candidate classes likely to include the ground-truth class, significantly reducing the search space and labeling cost. Moreover, we leverage conformal prediction to dynamically generate small yet reliable candidate sets, adapting to model enhancement over successive AL rounds. To this end, we introduce an acquisition function designed to prioritize data points that offer high information gain at lower cost. Empirical evaluations on CIFAR-10, CIFAR-100, and ImageNet64x64 demonstrate the effectiveness and scalability of our framework. Notably, it reduces labeling cost by 48% on ImageNet64x64.