Seohong Park

[Pronunciation: "suh-hong" ("Seo" as in "Seoul")]

Hey! I'm a Ph.D. student at UC Berkeley advised by Sergey Levine. I'm broadly interested in developing better reinforcement learning (RL) algorithms and understanding their mathematical properties. I believe RL can be maximally effective and scalable when we can leverage large-scale unsupervised pre-training, potentially from unlabeled data. To this end, I'm working on unsupervised RL, goal-conditioned RL, and offline RL.

CV | Email | Twitter | Google Scholar | Github

Publications
(*: equal contribution)

OGBench: Benchmarking Offline Goal-Conditioned RL

Seohong Park, Kevin Frans, Benjamin Eysenbach, Sergey Levine
Preprint
paper | code | project page

Unsupervised-to-Online Reinforcement Learning

Junsu Kim*, Seohong Park*, Sergey Levine
Preprint
paper | thread

Is Value Learning Really the Main Bottleneck in Offline RL?

Seohong Park, Kevin Frans, Sergey Levine, Aviral Kumar
NeurIPS 2024
ICML 2024 ARLET Workshop (Oral)
paper | thread | talk (10 min) | blog post

Foundation Policies with Hilbert Representations

Seohong Park, Tobias Kreiman, Sergey Levine
ICML 2024
paper | code | thread | project page

Unsupervised Zero-Shot Reinforcement Learning via Functional Reward Encodings

Kevin Frans, Seohong Park, Pieter Abbeel, Sergey Levine
ICML 2024 (Spotlight)
paper | code | thread | project page

METRA: Scalable Unsupervised RL with Metric-Aware Abstraction

Seohong Park, Oleh Rybkin, Sergey Levine
ICLR 2024 (Oral)
paper | code | thread | talk (15 min) | project page

HIQL: Offline Goal-Conditioned RL with Latent States as Actions

Seohong Park, Dibya Ghosh, Benjamin Eysenbach, Sergey Levine
NeurIPS 2023 (Spotlight)
paper | code | thread | talk (5 min) | project page

Controllability-Aware Unsupervised Skill Discovery

Seohong Park, Kimin Lee, Youngwoon Lee, Pieter Abbeel
ICML 2023
paper | code1 | code2 | thread | project page

Predictable MDP Abstraction for Unsupervised Model-Based RL

Seohong Park, Sergey Levine
ICML 2023
paper | code | thread | project page

Constrained GPI for Zero-Shot Transfer in Reinforcement Learning

Jaekyeom Kim, Seohong Park, Gunhee Kim
NeurIPS 2022
paper

Lipschitz-constrained Unsupervised Skill Discovery

Seohong Park, Jongwook Choi*, Jaekyeom Kim*, Honglak Lee, Gunhee Kim
ICLR 2022
paper | code | thread | project page

Time Discretization-Invariant Safe Action Repetition for Policy Gradient Methods

Seohong Park, Jaekyeom Kim, Gunhee Kim
NeurIPS 2021
paper | code

Unsupervised Skill Discovery with Bottleneck Option Learning

Jaekyeom Kim*, Seohong Park*, Gunhee Kim
ICML 2021
paper | code

Education

University of California, Berkeley (Aug 2022 - Present)

Ph.D. student in Computer Science

Seoul National University (Mar 2014 - Aug 2022)

B.S. in Computer Science and Engineering
Leave of absence for military service: Sep 2017 - Sep 2020 (3 years)

The University of Tokyo (Sep 2016 - Feb 2017)

Exchange student

Work Experience

Devsisters (Sep 2018 - Sep 2020)

Machine Learning Engineer
Worked as part of the mandatory military service in the Republic of Korea

Ace Project (Sep 2017 - Aug 2018)

Software Engineer
Worked as part of the mandatory military service in the Republic of Korea

Honors and Awards

Scholarships

KFAS Overseas PhD Scholarship (Aug 2022 - Present)

Korea Foundation for Advanced Studies (KFAS)
Full tuition, insurance, and living expenses support for graduate studies

Berkeley Fellowship (Aug 2022 - Aug 2023)

Presidential Science Scholarship (Mar 2014 - Aug 2022)

Korea Student Aid Foundation (KOSAF)
Full tuition and living expenses support for undergraduate studies

Awards

Gold Prize (1st Place in Signal Processing), Samsung Humantech Paper Award (Jan 2022)

Programming Contests (Selected)

2nd Place, ACM-ICPC Asia Daejeon Regional Contest (Nov 2016)

1st Place, Google Code Jam Round 1C (May 2016)

3rd Place, ACM-ICPC Asia Daejeon Regional Contest (Nov 2015)

1st Place, ACM-ICPC Asia Daejeon Regional Preliminary Contest (Oct 2015)

1st Place, Korea Olympiad in Informatics (KOI) (Jul 2012)

Codeforces: polequoll

Services

Reviews

Conferences: ICML (2023, 2024), NeurIPS (2023, 2024), ICLR (2023, 2024, 2025), IROS (2024)

Workshops: ICML Frontiers4LCD (2023), NeurIPS FMDM (2023), ICML ARLET (2024)

Template