임성빈(林聖彬)
Lim, Sungbin

Research Area & Collaboration

My recent research focuses on a Computational Statistics approach to incomplete data problems in Machine Learning (including Deep Learning) from the Trustworthy & Causal AI perspective. I am recently interested in the following topics:

• • Probabilistic Machine Learning: 

    • Generative Models (NeurIPS '23, NeurIPS '23, NeurIPS '24)

    • Robust Learning (ICRA '18, CVPR '20, NeurIPS '21, EACL '23)

  • Stochastic Optimization

    • Bandit (NeurIPS '20 , TNNLS '22, NeurIPS `25) & Reinforcement Learning (RSS '20, TR Part C '23)

    • Black-box Optimization (AAAI '20) & AutoML (NeurIPS '19, MICCAI '19)

  • Causal Learning & Machine Reasoning

    • Generative Model based Causal Reasoning (ACML '17 Machine Learning for AI Platforms Workshop, NeurIPS `25)

    • Large Language Model based Causal Discovery (NeurIPS '24  Causality & Large Models Workshop)

  • AI for Scientific Computing and Discovery

    • Collaboration with LG AI Research (NeurIPS '25  AI for Science Workshop, ICML '26, COLM '26)

(Important Notice)

현재 고려대학교 인공지능학과 김성웅 교수님과 Artificial General Intelligence Lab 을 공동지도하고 있습니다. 연구실에서는 기계학습 이론 연구를 희망하는 석·박사과정 학생 중 수학, 통계학, 최적화, 알고리즘, 기계학습 및 프로그래밍 적성이 우수한 지원자를 인터뷰를 통해 선발하여 집중적인 연구 지도를 합니다. 연구실 소속 없이 지도교수를 희망하는 통계학과 대학원생 분들은 KUBIG 학회원에 한하여 선발하고 있으니 참고 바랍니다.

프로그래밍(C++, Python, Julia), 기계학습 모델링, 고성능 컴퓨팅에 숙달한 소프트웨어 개발자나 박사후 연구원은 기업 협업을 통해 공동연구 프로젝트에 참여할 수 있는 기회를 제공하고 있습니다.

Office Hour & Consulting Service

AMA(Ask Me Anything) 상담을 포함한 개인 면담은 당분간 진행하고 있지 않습니다. 

Selected Publications

See [Google Scholar] for other publications.

• (new) AnnFlux: object-conditioned neural stochastic differential equations for single-cell perturbation dynamics, 2026 [paper]

• (new) Plan-and-Verify Reward Reasoning with Spatio-Temporal Scene Graph Grounding, EMNLP, 2026 [paper]

• (new) RetroReasoner: A Reasoning LLM for Strategic Retrosynthesis Prediction, EMNLP Findings, 2026 [paper]

• (new) MolDesignBench: Evaluating LLM-based Agent for Scenario-grounded Molecular Design, COLM, 2026 [paper]

• (new) Machine Learning Hamiltonians are Accurate Energe-Force Predictors, ICML, 2026 [paper]

• (new) ORCA: ORchestrating Causal Agent, CHI (Poster), 2026 [paper]

• Pareto Optimal Risk-Agnostic Distributional Bandits with Heavy-Tail Rewards, NeurIPS, 2025 [paper]

• Score-informed Neural Operator for Enhancing Ordering-based Causal Discovery, NeurIPS, 2025 [paper]

• ReactionReasoner: Towards Reasoning LLM for Chemical Reaction Prediction, NeurIPS Workshop on AI for Science, 2025 [paper]

• Mol-LLM: Multimodal Generalist LLM with Improved Graph Utilization, NeurIPS Workshop on AI for Science, 2025 [paper]

• On Incorporating Prior Knowledge Extracted from Pre-trained Language Models into Causal Discovery, NeurIPS Workshop on Causality and Large Models, 2024 [paper]

• Stochastic Optimal Control for Diffusion Bridges in Function Spaces, NeurIPS, 2024 [paper]

• Filling in the Gaps: LLM-based Structured Data Generation from Semi-Structured Scientific Data, ICML Workshop on AI for Science, 2024 [paper]

• Scalable Multi-Task Transfer Learning for Molecular Property Prediction, ICML Workshop on AI for Science, 2025 [paper]

• Score-based Generative Modeling through Stochastic Evolution Equations in Hilbert Spaces, NeurIPS (Spotlight), 2023 [paper]

• Score-based Generative Models with Lévy Processes, NeurIPS (Spotlight), 2023 [paper]

• • Neural Bootstrapper, NeurIPS, 2021 [paper] [code]

  • Optimal Algorithms for Stochastic Multi-Armed Bandits with Heavy Tailed Rewards, NeurIPS, 2020 [paper]

  • Task Agnostic Robust Learning on Corrupt Outputs by Correlation-Guided Mixture Density Network, CVPR (Oral), 2020 [video] [paper] [code] 

  • Fast AutoAugment, NeurIPS, 2019 [paper] [code]

  • Scalable Neural Architecture Search for 3D Medical Image Segmentation, MICCAI, 2019 [paper]

  • A Sobolev space theory for stochastic partial differential equations with time-fractional derivatives, Ann. Probab., 2019 [paper]

  • An Lq(Lp)-theory for the time fractional evolution equations with variable coefficients, Adv. Math., 2017 [paper]

  • Neural Stain-Style Transfer Learning using GAN for Histopathological Images, ACML Workshop on Machine Learning for AI Platforms, 2017 [paper]

News

• • Our lab students ranked 2nd on the Global Leaderboard and 1st on the Student Leaderboard at ML4CO NeurIPS 2021 Competition. [Leaderboard] [News]

Open Source

• • AutoCLINT: The Winning Method in AutoCV Challenge 2019, 2020 [paper] [code]

  • torchgpipe: On-the-fly Pipeline Parallelism for Training Giant Models, 2020 [paper] [code]