Woosung (Reiss) Koh
Woosung (Reiss) Koh

Research Intern

Trillion Labs

About Me

👋 Hi! My principal research interest is at the intersection of Foundation Models and Agentic Behaviour. [P1] I have recently worked on efficient and effective self-taught reasoning training. My past works make [C1] multi-agents more generalizable and significantly more reliable; and [W1] improves agents in noisy environments. I have also made an [C2] agentic framework for scalable LM-based chart generation (code and image), with robust improvements across 27B, 70B, and frontier-level LMs.

My current research topic revolves around: (1) LM pretraining, (2) LM reasoning, and (3) LM evaluation. Nevertheless, I am interested in a wide-range of open problems.

In the past, I worked on [J1] personalization and user-centric machine learning; and [W2] improving time-series representation learning under non-stationarity.

I am perfectly bilingual in English and Korean. I love to chat about research and downstream impact; feel free to reach out via email 📧.

Interests
  • Reasoning and Agents
  • Efficient Learning and Models
Education

  • Graduate Student

    KAIST AI

  • Bachelors Degree

    Yonsei University

  • Diploma Programme

    International Baccalaureate

📚 Publications

*First Author(s), ^Advisor(s)

[WIP] Woosung Koh*, ... (2025). rBridge: Efficient Evaluation Scale Transfer for Reasoning Pre-training. Work in Progress.
[P1] Woosung Koh*, Wonbeen Oh, Jaein Jang, MinHyung Lee, Hyeongjin Kim, Ah Yeon Kim, Joonkee Kim, Junghyun Lee, Taehyeon Kim, Se-Young Yun^ (2025). AdaSTaR: Adaptive Data Sampling for Training Self-Taught Reasoners. Preprint.
PDF
[C2] Woosung Koh*, Jang Han Yoon*, MinHyung Lee, Youngjin Song, Jaegwan Cho, Jaehyun Kang, Taehyeon Kim, Se-Young Yun^, Youngjae Yu^, Bongshin Lee^ (2024). C2^2: Scalable Auto-Feedback for LLM-based Chart Generation. NAACL 2025 Main (Oral, Long) (7.7% of Submissions).
PDF Code Project Video
[C1] Woosung Koh*, Wonbeen Oh, Siyeol Kim, Suhin Shin, Hyeongjin Kim, Jaein Jang, Junghyun Lee, Se-Young Yun^ (2024). FlickerFusion: Intra-trajectory Domain Generalizing Multi-Agent RL. ICLR 2025 (Prelim. NeurIPS 2024 Open-World Agent Workshop).
PDF Code Project
[W2] Insu Choi*, Woosung Koh*, Gimin Kang, Yuntae Jang, Woo Chang Kim^ (2024). Encoding Temporal Statistical-space Priors via Augmented Representation. IJCAI 2024 STRL Workshop (Oral).
PDF
See all publications
🤺 Mentees

I have been fortunate to mentor individuals listed below, just as I have benefited from the guidance of numerous mentors and advisors throughout my own journey.

  • MinHyung Lee ⟶ Research Intern, DAVIAN Lab, KAIST AI
  • Hyeongjin Kim ⟶ Intern (Advanced Research), Hyundai Mobis