Research
Large Language Models & NLP
We build efficient, trustworthy language models and agents — structure-aware document-editing agents, entity-level machine unlearning, annotation-free in-context translation, and dialogue systems — together with vision-language models for captioning and cultural understanding.
54 publications
863 citations
Selected Publications — 5 most representative of 54
Evaluating Visual and Cultural Interpretation: The K-Viscuit Benchmark with Human-VLM Collaboration
Builds cultural VLM benchmarks via human-VLM collaboration, producing K-Viscuit for Korean culture and revealing open-source models trail proprietary ones
Opt-Out: Investigating Entity-Level Unlearning for Large Language Models via Optimal Transport
Opt-Out uses optimal-transport Wasserstein distance to erase all knowledge of a target entity from an LLM while preserving its other capabilities
