Chanwoong Yoon

I am a Master's student at Korea University, advised by Professor Jaewoo Kang. My research focuses on making language models more interpretable, safer, and reliable.

🚀 Starting September 2025, I will be joining Georgia Tech as a visiting researcher, working with Professor Alan Ritter. I'll primarily focus on LLM moderation.

Research Areas: Natural Language Processing, AI Safety, Model Interpretability

Email / CV / Google Scholar / LinkedIn / X / Github

Research Interests

AI Safety: My recent work focuses on advanced LLM moderation. I'm developing methods that move beyond simple binary classification (safe/unsafe) to precisely remove harmful information while preserving the utility of the response.
Interpretability: I've explored methods (e.g., Activation Steering, Sparse Autoencoder) to gain insights into the internal mechanisms of LLMs. Beyond simply controlling these models without understanding their inner workings, I aim to comprehend how LLMs make decisions and leverage world knowledge to perform complex tasks.
Retrieval-Augmented Generation: I’ve tackled RAG bottlenecks, focusing on improving retriever and reader components. For example, RetPO optimizes a query reformulation model to better align with retriever preferences, enhancing off-the-shelf retrievers in conversational search, while CompAct actively compresses retrieved documents to help models to handle long contexts more efficiently.

News

08.2025. 1 paper accepted to @ EMNLP'25
05.2025. 🎓 Selected as a recipient of a research fund supported by the Korea Institute for Advancement of Technology (KIAT), receiving over USD 20,000 in research support to conduct a research visit at Georgia Tech.
03.2025. 2 papers accepted to @ ACL'25: TemporalHead, Outlier-Safe Pre-Training (OSP).
01.2025. 3 papers accepted to @ NAACL'25: RetPO, ETHIC, RAG².
01.2025. Unveiling chronological knowledge of language models accepted to @ ICLR'25: ChroKnowledge.
09.2024. Novel framework for context compression accepted to @ EMNLP'24: CompAct.
07.2024. Language Model-Augmented Police Investigation System paper accepted to @ CIKM'24: LAPIS.

Selected Publications

Ask Optimal Questions: Aligning Large Language Models with Retriever's Preference in Conversational Search
Chanwoong Yoon*, Gangwoo Kim*, Byeongguk Jeon, Sungdong Kim, Yohan Jo, Jaewoo Kang
NAACL 2025 Findings.

Paper / Code

we present Retriever’s Preference Optimization (RetPO), which optimizes a language model (LM) for reformulating search queries in line with the preferences of the target retrieval systems.

CompAct: Compressing Retrieved Documents Actively for Question Answering
Chanwoong Yoon, Taewhoo Lee, Hyeon Hwang, Minbyul Jeong, Jaewoo Kang
EMNLP 2024 Main.

Paper / Code

We propose a novel framework that employs an active strategy for compressing extensive documents. CompAct dynamically preserves query-related contexts, focusing on the integration of information across documents.

Template based on Jon Barron's website.