Sangwoo Park

E-mail: swgger (at) kaist.ac.kr

I am a Ph.D. student at the KAIST Graduate School of AI, co-advised by Sung Ju Hwang and Seong Joon Oh. My research focuses on LLM reasoning, distillation, and AI safety research. Current interests & recent work include:

  • Reasoning: Reducing sensitivity to pre-generated conditions, exploration, and extending reasoning to domains without readily verifiable rewards.
  • Distillation: Learning from noisy supervision or natural language feedback.
  • AI Safety: Upholding safety or contextual integrity while preserving useful reasoning and agentic capabilities, and mitigating reward hacking.

Updates

  • 3 papers accepted to NeurIPS 2026 [C3], [C4], [C5] 🎉
  • T-MAP selected as an EMNLP 2026 Oral Presentation [C2] 🎉

Education

Mar 2026 - Present Ph.D. in Artificial Intelligence
Korea Advanced Institute of Science and Technology (KAIST), Seoul, Korea
Advisors: Prof. Sung Ju Hwang and Prof. Seong Joon Oh
Mar 2025 - Feb 2026 M.S. in Artificial Intelligence
Korea Advanced Institute of Science and Technology (KAIST), Seoul, Korea
Advisor: Prof. Sung Ju Hwang
Mar 2019 - Feb 2024 B.S. in Computer Science & Engineering
Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Korea

Work Experience

Jun 2026 - Aug 2026 Applied Scientist Intern
Amazon, Berlin, Germany
Mar 2024 - Sep 2024 Research Engineer Intern
LG AI Language Lab, Seoul, Korea
Jan 2024 - Feb 2024 Research Intern
NCSOFT Speech AI Lab, Bundang, South Korea
Feb 2022 - Aug 2022 Research Engineer Intern
NAVER CLOVA, Bundang, South Korea

Publications

*: equal contribution

Conference Publications

  1. [C5] It Takes Two: Complementary Self-Distillation for Contextual Integrity in LLMs
    Sangwoo Park*, Woongyeong Yeo*, Seanie Lee, Yumin Choi, Hyomin Lee, and 4 more authors
    Neural Information Processing Systems (NeurIPS) 2026
    arXiv CODE
  2. [C4] PREPING: Building Agent Memory without Tasks
    Yumin Choi, Sangwoo Park, Minki Kang, Jinheon Baek, and Sung Ju Hwang
    Neural Information Processing Systems (NeurIPS) 2026
    arXiv PROJECT
  3. [C3] ThinkSafe: Self-Generated Safety Alignment for Reasoning Models
    Seanie Lee*, Sangwoo Park*, Yumin Choi, Gyeongman Kim, Minki Kang, and 4 more authors
    Neural Information Processing Systems (NeurIPS) 2026
    arXiv CODE
  4. [C2] T-MAP: Red-Teaming LLM Agents with Trajectory-aware Evolutionary Search
    Hyomin Lee, Sangwoo Park, Yumin Choi, Sohyun An, Seanie Lee, and 1 more author
    Empirical Methods in Natural Language Processing (EMNLP) 2026
    Oral Presentation
    arXiv CODE
  5. [C1] FedSVD: Adaptive Orthogonalization for Private Federated Learning with LoRA
    Seanie Lee*, Sangwoo Park*, Dong Bok Lee*, Dominik Wagner, Haebin Seong, and 3 more authors
    Neural Information Processing Systems (NeurIPS) 2025
    arXiv CODE

Journal Publications

  1. [J1] Rethinking Reward Models for Multi-Domain Test-Time Scaling
    Dong Bok Lee*, Seanie Lee*, Sangwoo Park, Minki Kang, Jinheon Baek, and 10 more authors
    Transactions on Machine Learning Research (TMLR)
    arXiv CODE

Workshop Publications

  1. [W1] Nudging Beyond the Comfort Zone: Efficient Strategy-Guided Exploration for RLVR
    Chanuk Lee*, Sangwoo Park*, Minki Kang, and Sung Ju Hwang
    NeurIPS 2026 Workshop MATH-AI
    arXiv CODE

Preprints

  1. [P4] Knowing When Thinking Is Not Enough: Teaching Small Reasoning Models to Reason Beyond Their Parametric Knowledge
    Chanuk Lee, Minki Kang, Sangwoo Park, Woongyeong Yeo, Jinheon Baek, and 1 more author
    Preprint 2026
    arXiv CODE
  2. [P3] Surprising Success, Repeated Failure: Entropy-Guided Credit Assignment for Exploration in LLM Reasoning
    Woongyeong Yeo, Minki Kang, Chanuk Lee, Sangwoo Park, Jinheon Baek, and 1 more author
    Preprint 2026
    arXiv PROJECT
  3. [P2] OmniRetrieval: Unified Retrieval across Heterogeneous Knowledge Sources
    Jinheon Baek, Soyeong Jeong, Sangwoo Park, Woongyeong Yeo, Minki Kang, and 3 more authors
    Preprint (Under Review) 2026
    arXiv CODE
  4. [P1] FedRand: Enhancing Privacy in Federated Learning with Randomized LoRA Subparameter Updates
    Sangwoo Park*, Seanie Lee*, Byungjoo Kim, and Sung Ju Hwang
    Preprint (Under Review) 2025
    arXiv

Language

Korean Native
English Full academic proficiency
German "Ein Döner, bitte!"