Taehwan Lee (이태환)

Google Scholar Link, [CV]

Ph.D Student, Graduate School of Artificial Intelligence, UNIST

Address: 409-1, Building 106, 50 UNIST-gil, Ulju-gun, Ulsan 44919, Republic of Korea

Email: taehwan@unist.ac.kr

Lab: MIIT Lab

Research Interests

• • • Federated & decentralized learning

    • Model generalization

    • Multimodal learning

    • Generative models

Education

• • • Ph.D in Artificial Intelligence Graduate School, Ulsan National Institute of Science and Technology (UNIST), Ulsan, South Korea.

      1. • Advisor: Sung Whan Yoon (Lab: MIIT)

         • Mar. 2022 ~ Present

    • M.S. in Electrical Engineering at Ulsan National Institute of Science and Technology (UNIST), Ulsan, South Korea.

      1. • Advisor: Jinho Chung

         • Mar. 2017 ~ Aug. 2019

    • B.S. in Electronic Engineering, Hanbat National University, Daejeon, South Korea.

      1. • Advisor:   Kyoungjae Lee (Lab: WiSL)

         • Mar. 2011 ~ Feb. 2017.

Journals & Conferences

*Equal contribution.

• Taehwan Lee*, Kyeongkook Seo*, Jaejun Yoo, Sung Whan Yoon, "Understanding Flatness in Generative Models: Its Role and Benefits," in International Conference on Computer Vision (ICCV), Honolulu, HI, 2025. 

• [Paper] [Project]

• SeungBum Ha*, Taehwan Lee*, Jiyoun Lim, Sung Whan Yoon, "Benchmarking Federated Learning for Semantic Datasets: Federated Scene Graph Generation," Pattern Recognition Letters, vol. 197, Nov. 2025 

• [Paper] [GitHub]

• Taehwan Lee, Sung Whan Yoon, "Rethinking the Flat Minima Searching in Federated Learning," in the 41st International Conference on Machine Learning (ICML), Vienna, Austria, 2024. 

• [Paper][GitHub]

• Taehwan Lee*, Hee-Heon Jung*, and Jin-Ho Chung. "A new one-coincidence frequency-hopping sequence set of length p2-p." 2018 IEEE Information Theory Workshop (ITW). IEEE, 2018.

Projects

* All projects were supervised by the PI (Prof. Sung Whan Yoon).

• FLAME-ARK: Federated Learning Multimodal Foundation Models for Medical Institutes 

  • • Korean Ministry of Health and Welfare (Korea) | Apr. 2025 – Present

    • Develop privacy-preserving foundation-model federated learning techniques for multimodal medical data.

    • Investigate parameter-efficient adaptation in FL (e.g., LoRA-based tuning) and multimodal learning methods.

    • Academic outcome: Understanding Flatness in Generative Models: Its Role and Benefits (ICCV 2025).

• Korea–U.S. (NSF) International Collaborative Research Project

  • • IITP (Korea) | Jan. 2023 – Nov. 2024

    • Address non-IID data challenges in federated learning arising from heterogeneous data distributions across edge devices.

    • Academic outcome: Rethinking the Flat Minima Searching in Federated Learning (ICML 2024).

• Development of a Hierarchical FL Simulator and Contribution Measurement Toolkit

  • • ETRI (Korea) | Jun. 2024 – Nov. 2024

• Experimental Testbed for Asynchronous Consensus in Federated Learning (Multi-metric inference-driven)

  • • ETRI (Korea) | Jun. 2023 – Nov. 2023

• Feasibility Validation for Resource-Obfuscated Deep Learning Models

  • • ETRI (Korea) | Apr. 2023 – Nov. 2023

    • Academic outcome: Benchmarking Federated Learning for Semantic Datasets: Federated Scene Graph Generation (PRL).

Experiences

• "Rethinking the Flat Minima Searching in Federated Learning" is awarded as the finalist of Qualcomm Innovation Fellowship Korea (QIFK) 2024 [Link].

• Teaching Assistant: Electrical Engineering Programming

  • • Course: EE233, 2025 spring @ UNIST. 

    • Description: Basic programming tools for electrical engineering (C++).

• Teaching Assistant: Electrical Engineering Programming

  • • Course: EE233, 2024 spring @ UNIST. 

    • Description: Basic programming tools for electrical engineering (C++).