Reinforcement Learning
SimBa: Simplicity Bias for Scaling Up Parameters in Deep Reinforcement Learning
Hojoon Lee, Dongyoon Hwang, Donghu Kim, Hyunseung Kim, Jun Jet Tai, Kaushik Subramanian, Peter R. Wurman, Jaegul Choo, Peter Stone, Takuma Seno.
Preprint.
arXiv

Designed network architectures that steer convergence toward simple functions which allows to scale up parameters in RL.

Reinforcement Learning Skill Discovery
Do’s and Don’ts:Learning Desirable Skills with Instruction Videos
Hyunseung Kim, Byungkun Lee, Hojoon Lee, Dongyoon Hwang, Donghu Kim, Jaegul Choo,
NeurIPS'24.
project page / arXiv

We present DoDont, a skill discovery algorithm that learns diverse behaviors while following the behaviors in "do" videos while avoiding the behaviors in "don't" videos.

Reinforcement Learning Plasticity
Slow and Steady Wins the Race: Maintaining Plasticity with Hare and Tortoise Networks
Hojoon Lee, Hyeonseo Cho, Hyunseung Kim, Donghu Kim, Dugki Min, Jaegul Choo, Clare Lyle.
ICML'24.
arXiv / poster / Bibtex

To allow the network to continually adapt and generalize, we introduce Hare and Tortoise architecture, inspired by the complementary learning system of the human brain.

Reinforcement Learning Pre-training
ATARI-PB: Investigating Pre-Training Objectives for Generalization in Vision-Based RL
Donghu Kim*, Hojoon Lee*, Kyungmin Lee*, Dongyoon Hwang, Jaegul Choo.
ICML'24.
project page / arXiv / poster / Bibtex

We investigate which pre-training objectives are beneficial for in-distribution, near-out-of-distribution, and far-out-of-distribution generalization in visual reinforcement learning.

Reinforcement Learning Adaptation
Adapting Pretrained ViTs with Convolution Injector for Visuo-Motor Control
Dongyoon Hwang*, Byungkun Lee*, Hojoon Lee, Hyunseung Kim, Jaegul Choo.
ICML'24.
project page / arXiv / Bibtex

Introducing an add-on convolution module for pre-trained ViT models, enhancing adaptability for vision-based motor control with injecting locality and translation equivariant biases.

Reinforcement Learning Plasticity
PLASTIC: Improving Input and Label Plasticity for Sample Efficient Reinforcement Learning
Hojoon Lee*, Hanseul Cho*, Hyunseung Kim*, Daehoon Gwak, Joonkee Kim, Jaegul Choo, Se-Young Yun, Chulhee Yun.
NeurIPS'23.
arXiv / code / slide / poster / Bibtex

For sample-efficient RL, the agent needs to quickly adapt to various inputs (input plasticity) and outputs (label plasticity). We present PLASTIC, which maintains both input and label plasticity by identifying smooth local minima and preserving gradient flow.

Reinforcement Learning Skill Discovery
DISCO-DANCE: Learning to Discover Skills through Guidance
Hyunseung Kim*, Byungkun Lee*, Hojoon Lee, Dongyoon Hwang, Jaegul Choo.
NeurIPS'23.
project page / arXiv / code / Bibtex

We introduce DISCO-DANCE, a Skill Discovery algorithm focused on learning diverse, task-agnostic behaviors. DISCO-DANCE addresses the common limitation of exploration in skill discovery algorithms through explicit guidance.

Reinforcement Learning Pre-training
SimTPR: On the Importance of Feature Decorrelation for Unsupervised Representation Learning for Reinforcement Learning
Hojoon Lee, Koanho Lee, Dongyoon Hwang, Hyunho Lee, Byungkun Lee, Jaegul Choo.
ICML'23.
arXiv / code / poster / Bibtex

We present a visual pre-training algorithm grounded in self-predictive learning principles tailored for reinforcement learning.

Data Mining
ST-RAP: A Spatio-Temporal Framework for Real Estate Appraisal
Hojoon Lee*, Hawon Jung*, Byungkun Lee*, Jaegul Choo.
CIKM'23 (short).
arXiv / code / poster / Bibtex

We construct a real estate appraisal framework that integrates spatial and temporal aspects, validated using a dataset of 3.6M real estate transactions in South Korea from 2016 to 2020.

Data Mining Reinforcement Learning
Towards Validating Long-Term User Feedbacks in Interactive Recommender System
Hojoon Lee, Dongyoon Hwang, Kyushik Min, Jaegul Choo.
SIGIR'22 (short), Best Honorable Mention Award.
arXiv / poster / Bibtex

Through validating the long-term impact of user feedback in MovieLens and Amazon Review datasets, we've discovered that these datasets are inadequate for evaluating reinforcement learning-based interactive recommender systems.

Data Mining Reinforcement Learning Game
DraftRec: Personalized Draft Recommendation for Winning in MOBA Games
Hojoon Lee*, Dongyoon Hwang*, Hyunseung Kim, Byungkun Lee, Jaegul Choo.
WWW'22.
arXiv / code / poster / Bibtex

We gathered data from 280,000 matches played by the top 0.3% rank players in Korea for League of Legends. From this, we developed DraftRec, a personalized champion recommendation system aimed at maximizing players' win rates.

Data Mining Game
Enemy Spotted: In-game Gun Sound Dataset for Gunshot Classification and Localization
Junwoo Park, Youngwoo Cho, Gyuhyeon Sim, Hojoon Lee, Jaegul Choo.
COG'22.
arXiv / Bibtex

We collected a gunshot sound dataset from PlayerUnknown's Battlegrounds (PUBG). Using this data, we developed a gunshot localization model that can be applied to real-world scenarios.

Reinforcement Learning Game
Conquering a rule-changing game with action-releavance aware Alpha-Zero
Hojoon Lee, Dongyoon Hwang, Jeesoo Woo, Jaegul Choo.
Preprint'19.
poster

We've developed an AI agent for the turn-based strategy game BrownDust. To enhance learning efficiency, we've designed feature representations and specified architectures for embedding the game characters.