Changhun Lee

Postdoctoral Researcher, Graduate School of Artificial Intelligence, UNIST

changhun@unist.ac.kr, clubing92@gmail.com

Bio

I am a postdoctoral associate at the graduate school of artificial intelligence in Ulsan National Institute of Science and Technology (UNIST). I received my Ph.D. degree in February 2023 and won the best researcher award for my thesis titled "Tackling Three Problems in Controlled Sequence Generations: Bridging Reinforcement Learning with Language Models.". Since March 2023, I am working as a post-doc associate at the Artificial Intelligence Graduate School (AIGS) in UNIST.

My research interest lies in the development of controllable/reliable large language models (LLMs) and their application in AI4Healthcare & AI4Science. In particular, the use of reinforcement learning to align LLMs with human preferences is of the most interest that I am currently focusing on. In this regard, my research topics range from fundamental to applied NLP+RL research, including but not limited to:

  • Developing fundamental frameworks, algorithms, and building blocks for controllable/reliable LLMs.
  • Designing alignment mechanisms to reinforce the interaction between LLMs and humans.
  • Deploying an LLM-based service, e.g., chatbot, to solve real-world problems in healthcare and science.

    • Publications

    Most recent publications on Google Scholar.
    ‡ indicates equal contribution.

    Reward Dropout Improves Control: Bi-objective Perspective on Reinforced LM

    Changhun Lee, Chiehyeon Lim

    arXiv:2310.04483

    Paper Code

    Recommendation in Offline Stores: A Gamification Approach for Learning the Spatiotemporal Representation of Indoor Shopping

    Jongkyung Shin‡, Changhun Lee‡, Chiehyeon Lim, Yunmo Shin, Junseok Lim

    KDD '22: Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining.

    Paper Code Data

    MIND dataset for diet planning and dietary healthcare with machine learning: Dataset creation using combinatorial optimization and controllable generation with domain experts

    Changhun Lee‡, Soohyeok Kim‡, Sehwa Jeong, Jayun Kim, Yeji Kim, Chiehyeon Lim, Minyoung Jung

    Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks (NeurIPS Datasets and Benchmarks 2021)

    Paper Code Data

    Diet Planning with Machine Learning: Teacher-forced REINFORCE for Composition Compliance with Nutrition Enhancement

    Changhun Lee, Soohyeok Kim, Chiehyeon Lim, Jayun Kim, Yeji Kim, Minyoung Jung

    KDD '21: Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining.

    Paper Code

    From technological development to social advance: A review of Industry 4.0 through machine learning

    Changhun Lee, Chiehyeon Lim

    Technological Forecasting and Social Change, 167, 120653.

    Paper

    Recommendation in Offline Stores: A Gamification Approach for Learning the Spatiotemporal Representation of Indoor Shopping

    Jongkyung Shin‡, Changhun Lee‡, Chiehyeon Lim, Yunmo Shin, Junseok Lim

    KDD '22: Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining.

    Paper Code Data

    Challenges of Diet Planning for Children using Artificial Intelligence

    Changhun Lee‡, Soohyeok Kim, Jayun Kim, Chiehyeon Lim, Minyoung Jung

    Nutrition Research and Practice. 2022; 16.

    Paper

    MIND dataset for diet planning and dietary healthcare with machine learning: Dataset creation using combinatorial optimization and controllable generation with domain experts

    Changhun Lee‡, Soohyeok Kim‡, Sehwa Jeong, Jayun Kim, Yeji Kim, Chiehyeon Lim, Minyoung Jung

    Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks (NeurIPS Datasets and Benchmarks 2021)

    Paper Code Data

    Diet Planning with Machine Learning: Teacher-forced REINFORCE for Composition Compliance with Nutrition Enhancement

    Changhun Lee, Soohyeok Kim, Chiehyeon Lim, Jayun Kim, Yeji Kim, Minyoung Jung

    KDD '21: Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining.

    Paper Code

    Human dietitians vs. Artificial intelligence: Which diet design do you prefer for your children?

    Min Jung, Chiehyeon Lim, Changhun Lee, Soohyeok Kim, Jayun Kim

    Journal of Allergy and Clinical Immunology, 147(2), AB117.

    Paper

    Reward Dropout Improves Control: Bi-objective Perspective on Reinforced LM

    Changhun Lee, Chiehyeon Lim

    arXiv:2310.04483

    Paper Code

    Reward Dropout Improves Control: Bi-objective Perspective on Reinforced LM

    Changhun Lee, Chiehyeon Lim

    arXiv:2310.04483

    Paper Code

    Recommendation in Offline Stores: A Gamification Approach for Learning the Spatiotemporal Representation of Indoor Shopping

    Jongkyung Shin‡, Changhun Lee‡, Chiehyeon Lim, Yunmo Shin, Junseok Lim

    KDD '22: Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining.

    Paper Code Data

    Challenges of Diet Planning for Children using Artificial Intelligence

    Changhun Lee‡, Soohyeok Kim, Jayun Kim, Chiehyeon Lim, Minyoung Jung

    Nutrition Research and Practice. 2022; 16.

    Paper

    MIND dataset for diet planning and dietary healthcare with machine learning: Dataset creation using combinatorial optimization and controllable generation with domain experts

    Changhun Lee‡, Soohyeok Kim‡, Sehwa Jeong, Jayun Kim, Yeji Kim, Chiehyeon Lim, Minyoung Jung

    Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks (NeurIPS Datasets and Benchmarks 2021)

    Paper Code Data

    Diet Planning with Machine Learning: Teacher-forced REINFORCE for Composition Compliance with Nutrition Enhancement

    Changhun Lee, Soohyeok Kim, Chiehyeon Lim, Jayun Kim, Yeji Kim, Minyoung Jung

    KDD '21: Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining.

    Paper Code

    From technological development to social advance: A review of Industry 4.0 through machine learning

    Changhun Lee, Chiehyeon Lim

    Technological Forecasting and Social Change, 167, 120653.

    Paper

    Human dietitians vs. Artificial intelligence: Which diet design do you prefer for your children?

    Min Jung, Chiehyeon Lim, Changhun Lee, Soohyeok Kim, Jayun Kim

    Journal of Allergy and Clinical Immunology, 147(2), AB117.

    Paper

    Fundings

    As of September 1, 2023, I have a research grant from the National Research Foundation of Korea for the next 2 years, with permission to use this grant to stay abroad for up to 6 months. This means that I can support myself for 6 months and conduct research as a visiting researcher with the status of a self-funded resident fellow before the official postdoctoral associate period begins. Please feel free to contact me at changhun@unist.ac.kr.

    Co-work Opportunities

    I am currently looking for a post-doctoral position related to NLP and RL research, with reponsibilities including:

    1. studying theoretical foundations of controllable/reliable LLMs,
    2. and applying them to solve real-world problems in AI4Healthcare and AI4Science contexts.
    Of course, I am also open to any other interesting topics as long as they are relevant to NLP and RL research. If you are interested, please contact me at changhun@unist.ac.kr.

    History

    In February 2015, I got my B.Sc in Economics from Ajou University as an outstanding student in the top 10% of the major GPA., and I enrolled in the master's program in the Department of Industrial Engineering at Ulsan National Institute of Science and Technology (UNIST) in February 2016.

    For the first one and a half years (Jan 2016 - June 2017), I focused on studying quantitative research methodologies (e.g., advanced statistics) and sequential modeling (e.g., time-series analysis). Then, I spent half a year as a data analyst intern at Hyundai Mipo Dockyard (July 2017 - Dec 2017), developing language models (LMs) and text-mining algorithms. Completing the internship, I returned to UNIST and transferred to the combined Master-Ph.D program.

    I spent the first two years of Ph.D. course (Jan 2018 - Jan 2020) in analyzing real-world problems with advanced text-mining techniques. For the last three years (Feb 2020 - Dec 2022), I focused on solving real-world problems using controllable language models (CLMs).

    In February 2023, I wrote my Ph.D thesis on controllable sequence generation, titled "Tackling Three Problems in Controlled Sequence Generations: Bridging Reinforcement Learning with Language Models." and won the best researcher award in UNIST.

    From March 2023, I'm joining the Graduate School of Artificial Intelligence at UNIST as a post-doc associate. Currently, I belong to the Service Intelligence Lab that is led by Prof.Chiehyeon Lim, my advisor whom I truly respect and appreciate.

    I dream of becoming a researcher who is willing to stand up against prejudice and inequality in the world with logical thinking and a scientific attitude. I am looking forward to a heart-pounding journey of research that will create a better world and promote the future with great colleagues. I will end this article by introducing a quote from one of my favorite scholars, Alfred Marshall:

    "Cool head, but warm heart!"

    Vitæ

    You can download my Curriculum Vitae in PDF file.

    Acknowledgement

    This website uses the website design and template by Martin Saveski

    Changhun Lee
    changhun@unist.ac.kr, clubing92@gmail.com