Featured Publications
arXiv preprint, 2026
Studies sequential reappearance in diffusion data-point unlearning and how deleted instances can remain forgotten across repeated unlearning requests.
The Fortieth Annual Conference on Neural Information Processing Systems, 2026
Identifies model-internal circuits associated with data utility and steers them to generate higher-quality, controllable training data for language models.
Current Research
Unlearning
반복적인 정보 삭제에도 모델의 성능을 유지할 수 있을까?
World Models
월드 모델은 실제 세계를 학습하는가, 데이터의 shortcut을 학습하는가?
Multi-Agent Systems
에이전트 간 hallucination은 어떻게 발생하고 전파되는가?
AI for Science
단백질 모델은 비생물학적 shortcut에 의존하는가?
Quantum Computing
측정과 피드백 비용을 줄여 양자 최적화를 가속할 수 있을까?
Latest News
Our paper, From Token Correction to Claim Admission: Training-Free Hallucination Mitigation for Large Vision-Language Models, is now available on arXiv.
Our paper, Rethinking Memorization Mitigation in Diffusion Models: Reinforcing Text Conditioning, is now available on arXiv.
TML joined Jeju National University for the AI Star Fellowship project workshop at Camphortree Hotel, Jeju, from October 1–3, 2026.
Our paper, Continual Data Unlearning in Diffusion Models via Transition-based Regularization, is now available on arXiv.
Our paper, Mechanistic Circuit Identification for Controllable Data Generation, has been accepted to NeurIPS 2026.
