About
I am a researcher at the Korea AI Safety Institute (AISI) within ETRI, working on explainable and robust AI-generated content and deepfake detection.
If you’re interested in collaborating, feel free to reach out! 💌
Research Interests
- AI Safety
- AI-Generated Contents Detection & Attribution
- Mechanistic Interpretability
- Trustworthy AI
- Robustness and Generalization
Publications
- SAGE: Semantic-Agnostic Image Embedding for Generalized AI-Generated Image Detection
Seongho Kim, Jaehyun Choi, Dahye Kim, Sungwon Yi, Jang-Ho Choi (NeurIPS), 2026 - Stability-Weighted Direction Regularization Disentangles Generator Shortcuts from Detection Signal
Jang-Ho Choi, Seongho Kim, Jaehyun Choi, Dahye Kim, Sungwon Yi, Eunho Yang (NeurIPS), 2026 - Dissect and Prune: Enhancing Robustness in AI-Generated Image Detection
Dahye Kim, Jaehyun Choi, Hyun Seok Seong, Seongho Kim, Donghun Lee, Sungwon Yi, Jang-Ho Choi (ICML), 2026
PosterICML 2026 arXiv2606.10309 CodeGitHub - Unsupervised Segmentation incorporating Shape Prior via Generative Adversarial Networks
Dahye Kim, Byung-Woo Hong (ICCV), 2021
PaperICCV 2021 Project Pageshape-gan-seg CodeGitHub
Academic Service
- NeurIPS — ⛰️ Top Reviewer (2026)