Research
My research is centered on building trustworthy AI systems whose decision making is robust and interpretable, encompassing the fields of representation learning, reinforcement learning, and causal inference.
In particular, my recent works involve developing robust and efficient algorithms for causal inference and causal discovery, with their application for building reliable machine learning models.
Additionally, I am interested in discovering and utilizing useful inductive biases to better align model decisions with human reasoning.
|
Publications
(* equal contribution, † equal advising)
|
|
Locality-aware Concept Bottleneck Model
Sujin Jeon*,
Inwoo Hwang*,
Sanghack Lee†,
Byoung-Tak Zhang†
NeurIPS Workshop on Unifying Representations in Neural Models, 2024
[PDF]
|
|
On Positivity Condition for Causal Inference
Inwoo Hwang*,
Yesong Choe*,
Yeahoon Kwon,
Sanghack Lee
International Conference on Machine Learning (ICML), 2024
UAI Workshop on Causal Inference, 2024
We establish rigorous foundations for licensing the use of identification formulas without strict positivity, a long-standing critical assumption in causal inference.
[PDF]
|
|
Fine-Grained Causal Dynamics Learning with Quantization for Improving Robustness in
Reinforcement Learning
Inwoo Hwang,
Yunhyeok Kwak,
Suhyung Choi,
Byoung-Tak Zhang†,
Sanghack Lee†
International Conference on Machine Learning (ICML), 2024
NeurIPS Workshop on Generalization in Planning, 2023
Finalist, Qualcomm Innovation Fellowship Korea, 2024
We propose a principled and practical approach to discovering fine-grained causal relationships with identifiability guarantees for robust decision-making.
[PDF]
[Code]
|
|
Efficient Monte Carlo Tree Search via On-the-Fly State-Conditioned Action Abstraction
Yunhyeok Kwak*,
Inwoo Hwang*,
Dooyoung Kim,
Sanghack Lee†,
Byoung-Tak Zhang†
Uncertainty in Artificial Intelligence (UAI), 2024   (Oral, 28/744=3.8%)
We propose state-conditioned action abstraction that effectively reduces the search space of MCTS under vast combinatorial action space by harnessing compositional relationships between state and sub-actions.
[PDF]
[Code]
|
|
Causal Discovery with Deductive Reasoning: One Less Problem
Jonghwan Kim,
Inwoo Hwang,
Sanghack Lee
Uncertainty in Artificial Intelligence (UAI), 2024
We propose a simple yet effective plug-in module that corrects unreliable CI statements through deductive reasoning using graphoid axioms,
thereby improving the robustness of constraint-based causal discovery methods.
[PDF]
[Code]
|
|
Learning Geometry-aware Representations by Sketching
Hyundo Lee,
Inwoo Hwang,
Hyunsung Go,
Won-Seok Choi,
Kibeom Kim,
Byoung-Tak Zhang
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),
2023
Inspired by human behavior that depicts an image by sketching, we propose a novel representation learning framework
that captures geometric information of the scene, such as distance or shape.
[PDF]
[Code]
|
|
On Discovery of Local Independence over Continuous Variables via Neural Contextual
Decomposition
Inwoo Hwang,
Yunhyeok Kwak,
Yeon-Ji Song,
Byoung-Tak Zhang†,
Sanghack Lee†
Conference on Causal Learning and Reasoning (CLeaR), 2023
NeurIPS Workshop on Causal Inference Challenges in Sequential Decision Making: Bridging
Theory and Practice, 2021
Local independence (e.g., context-specific independence) provides a way to understand fine-grained causal relationships,
but it has mostly been studied for discrete variables. We define and characterize local independence for continuous variables,
provide its fundamental properties, and propose a differentiable method to discover it.
[PDF]
[Code]
|
|
SelecMix: Debiased Learning by Contradicting-pair Sampling
Inwoo Hwang, Sangjun Lee,
Yunhyeok Kwak,
Seong Joon Oh,
Damien Teney,
Jin-Hwa Kim†,
Byoung-Tak Zhang†
Neural Information Processing Systems (NeurIPS), 2022
ICML Workshop on Spurious Correlations, Invariance, and Stability, 2022
Neural networks trained with ERM often learn unintended decision rules when trained on a biased dataset where the labels are strongly correlated with undesirable features.
We propose a novel debiasing method that applies mixup to the selected pairs of examples, utilizing a contrastive loss designed to amplify reliance on biased features.
[PDF]
[Code]
|
|
On the Importance of Critical Period in Multi-stage Reinforcement Learning
Junseok Park,
Inwoo Hwang,
Min Whoo Lee, Hyunseok Oh, Minsu Lee, Youngki Lee, Byoung-Tak Zhang
ICML Workshop on Complex Feedback in Online Learning, 2022
[PDF]
|
|
Improving Robustness to Texture Bias via Shape-focused Augmentation
Sangjun Lee,
Inwoo Hwang,
Gi-Cheon Kang,
Byoung-Tak Zhang
CVPR Workshop on Human-centered Intelligent Services: Safety and Trustworthy, 2022
[PDF]
|
Education
- (2019.03 - current) Ph.D in Computer Science and Engineering, Seoul National University
- (2016.03 - 2018.02) MS in School of Computing, KAIST
- (2010.02 - 2016.02) BS in Mathematical Science, KAIST
- (2007.02 - 2010.02) Highschool, Korea Science Academy of KAIST
|
Work Experience
- (Oct 2024 - current) Visiting scholar, Columbia University (host: Elias Bareinboim)
- (Sep 2021 - May 2022) External collaborator, Naver AI (host: Jin-Hwa Kim)
- (Aug 2012 - May 2014) Mandatory military service, Korean Augmentation To the US Army (KATUSA)
|
Academic Services
- Conference Reviewer: NeurIPS (2023-2024), ICLR (2024-2025), ICML (2024), AAAI (2025), AISTATS (2024-2025), CLeaR (2024-2025),
CVPR (2023-2024), ICCV (2023), ECCV (2024), ICRA (2024-2025)
- Journal Reviewer: IEEE Trans. Multimedia
- Workshop Reviewer
- NeurIPS 2024 Workshop on Causality and Large Models (CaLM)
- NeurIPS 2024 Workshop on Unifying Representations in Neural Models (UniReps)
- RLC 2024 Workshop on Reinforcement Learning Beyond Rewards (RLBRew)
- NeurIPS 2023 Workshop on Causal Representation Learning (CRL)
- ICML 2023 Workshop on Spurious Correlations, Invariance, and Stability (SCIS)
|
Invited Talks
- (Jun 2024) IITP Workshop
- (Sep 2023) IITP Workshop
- (May 2023) SNU AIIS Retreat
- (Dec 2022) Korea Software Congress
- (Nov 2022) Kakao Enterprise TechTalk
- (Nov 2022) SNU AIIS Retreat
- (Oct 2022) NAVER TechTalk
|
Honors and Awards
- Finalist, Qualcomm Innovation Fellowship Korea, 2024
- Outstanding Reviewer, ECCV 2024
- Recipient, Youlchon AI Star Scholarship, 2024
- Recipient, UAI Scholarship, 2024
- Recipient, NAVER PhD Fellowship, 2022
- Recipient, NeurIPS Scholarship, 2022
- Recipient, BK21 Plus Scholarship, Republic of Korea
- Recipient, National Science and Technology Scholarship, Korea Student Aid Foundation
- Gold Award, The Korean Mathematical Olympiad (KMO)
|
Teaching Experience
- [CS204] Discrete Mathematics, KAIST, 2016S - 2017F
|
The source of this website is from here.
|
|