Dongwook Lee

Hi! I’m a undergraduate student at Seoul National University advised by Gunhee Kim. I’m deeply interested in guiding agents to survive in uncertain environments and constructing RL foundation model. I am currently focusing on training or guiding agents with prior knowledge of existing foundation models, like VLMs.

Email  /  CV  /  Twitter  /  Github  /  LinkedIn

profile photo

Research

Overview of Compositional Conservatism Compositional Conservatism: A Transductive Approach in Offline Reinforcement Learning
Yeda Song*, Dongwook Lee*, Gunhee Kim
ICLR, 2024
Openreview / arXiv / CODE

COCOA is a novel approach in offline reinforcement learning (RL) that promotes compositional conservatism by breaking down the input state into an anchor and its deviation. It utilizes a learned reverse dynamics model to ensure both the anchors and differences align with in-distribution data, thereby mitigating OOD problem in the input space used by policies or value functions.

Education

Seoul National University | Seoul, Korea
B.S. in Naval Architecture and Ocean Engineering
B.S. in Computer Science
Mar. 2016 - Present
Peking University | Beijing, China
Exchange Student in Engineering College
Jan. 2022 - Jan. 2023

Experience

Seoul National University Vision & Learning Lab | Seoul, Korea
Intern Researcher in Reinforcement Learning RL foundation model and offline RL.
Jan. 2023 - Present
Peking University | Beijing, China
Intern Researcher in Multimodal Domain
Fine-grained classification on sound, image, and text modalities.
Oct. 2022 - Jan. 2023
Seoul National University Semi-Conductor for AI Laboratory | Seoul, Korea
Remote Exam Supervision Program Team Intern
An Android application for real-time exam supervision.
Sep. 2021 - Dec. 2021
Daewoo ShipBuilding & Marine Engineering | Siheung, Korea
Propulsion Design R&D Team Intern
Rendering a propeller model for a commercial container ship with C++.
Jul. 2020 - Aug. 2020
Military Service | Air-force, Korea
Jan. 2018 - Dec. 2019

Corporate Projects

Edutech Research Institute, Woongjin CoE | Seoul, Korea
Enhancing children's concentration by recognizing speech(TTS) from audiobooks and highlighting corresponding text on the book using optical character recognition (OCR).
Jan. 2023 - Dec. 2023

Feel free to clone the format from here.