Seoul National University of Science and Technology Professor Eom Dae-ho's research team's paper accepted by world-class machine learning conference ICML 2026
Translated from Korean, summarized and contextualized by DistantNews.
TLDR
- A research team led by Professor Eom Dae-ho at the University of Seoul has had their paper accepted by the prestigious International Conference on Machine Learning (ICML 2026).
- The paper proposes a new defense technique called SS-TPT to overcome the vulnerability of large vision-language models to image manipulation attacks.
- ICML 2026 will be held in Seoul, marking the first time the conference is hosted offline in Korea.
The University of Seoul is proud to announce a significant achievement in artificial intelligence research, with Professor Eom Dae-ho's team securing a place at the highly esteemed International Conference on Machine Learning (ICML 2026). This recognition, alongside their previous success at CVPR 2026, underscores the innovative work being done in Korea's AI sector.
The research, a collaborative effort with Professor Kim Sun-oh of Dankook University, introduces 'SS-TPT: Stability and Suitability-Guided Test-Time Prompt Tuning for Adversarially Robust Vision-Language Models.' This novel approach addresses critical vulnerabilities in large vision-language models like CLIP, which are susceptible to adversarial attacks through image manipulation. By focusing on stability and suitability metrics, the team has developed a method to enhance the robustness of these models without altering their internal parameters.
This research presents a new direction for improving the robustness of vision-language models against adversarial attacks, following our work on CVPR 2026.
Hosting ICML 2026 in Seoul is a landmark event, highlighting Korea's growing prominence in the global AI landscape. This conference provides a vital platform for Korean researchers to showcase their advancements and engage with international peers. The success of Professor Eom's team at such a prestigious event is a testament to the quality and impact of Korean AI research, contributing to the nation's reputation as a leader in technological innovation. This achievement is particularly meaningful as it demonstrates Korea's capability to not only participate in but also to lead discussions at the forefront of AI development.
We plan to continue research on improving the reliability of vision-language models through ongoing collaboration with Professor Kim Sun-oh and contribute to enhancing AI reliability in real life.
DistantNews Editorial
Originally published by Hankyoreh in Korean. Translated, summarized, and contextualized by our editorial team with added local perspective. Read our editorial standards.
More Stories
From Our Blog
What Gets Lost (and Found) When News Is Translated
The Critical Role of Diaspora Media in Global News
What Travellers Should Know About Countries with Restricted Press
A Digital Nomad's Guide to Following Local News
How the Same Story Looks Different in Different Countries
