Jeongwhan Choi 최정환
 |
Jeongwhan Choi, Ph.D. jeongwhan.choi [at] kaist [dot] ac [dot] kr On the academic job market, 2026–2027 |
About
I am a Jang Young-Sil Postdoctoral Fellow at the Information & Electronics Research Institute, KAIST, working with Noseong Park. I received my Ph.D. in Artificial Intelligence from Yonsei University in August 2025, co-advised by Noseong Park and Sung-Bae Cho.
My research develops principled deep learning methods for graphs and sequences, built on two foundations: differential equations, which model learning as continuous-time dynamics, and graph signal processing, which views message passing and self-attention as graph filters. With these tools, I have addressed over-smoothing and over-squashing in graph neural networks, improved self-attention in Transformers, and designed models for recommender systems and spatiotemporal forecasting. More recently, I have been working on foundation models built on prior-data fitted networks, which make predictions on unseen graphs and recommendation datasets without retraining. My work also extends to scientific machine learning and to applications in finance, biomedicine, and climate.
My work has appeared at ICML, NeurIPS, ICLR, KDD, AAAI, IJCAI, SIGIR, WSDM, and CIKM, including oral presentations at AAAI 2022, AAAI 2026, and KDD 2026. My most influential work, STG-NCDE (AAAI 2022), introduced neural controlled differential equations to spatiotemporal graphs and was ranked 13th among the most influential AAAI 2022 papers. I received the Qualcomm Innovation Fellowship (2024) and the Best Paper Runner-up Award from KAIST–Samsung Industry-Academic Collaboration, and was selected for the Jang Young-Sil Fellow Program (Excellence Track) at KAIST. I serve as General Co-Chair of LoG 2026 and Area Chair for ICLR 2027, and previously served as Web Chair for CIKM 2025 and Website Chair for LoG 2025. I have been recognized as a top reviewer at NeurIPS 2024, LoG 2024, KDD 2025 (both cycles), ICML 2026, and KDD 2026.
What's New all news
Sep 25, 2026 “Learning Hierarchical Forward Processes For Discrete Diffusion Language Models” was accepted in NeurIPS 2026!
Sep 17, 2026 Invited to serve as an Area Chair for ICLR 2027.
Aug 8, 2026 Recognized as the Best Reviewer for KDD 2026.
Jun 24, 2026 I gave two talks at Korea Computer Congress (KCC 2026): “Toward Graph Foundation Models via Synthetic Graph Priors” at the Early-Career Researcher Session (slides), and our ICLR 2026 paper “Learning Posterior Predictive Distributions for Node Classification from Synthetic Graph Priors” at the Top-Conference Session (slides).
Jun 23, 2026 “One Sequential Recommendation Model Pretrained from Synthetic Priors Predicts Multiple Datasets” selected for oral presentation at KDD 2026
May 26, 2026 “How Much Memory Do We Need? Adaptive Memory Gate for Neural Operators” was accepted for an AI4Physics workshop at ICML 2026.
May 16, 2026 A paper I co-authored, “One Sequential Recommendation Model Pretrained from Synthetic Priors Predicts Multiple Datasets,” was accepted to KDD 2026 (Research Track).
May 14, 2026 Received the Gold Reviewer Award at ICML 2026.
Apr 28, 2026 Our AAAI 2026 paper was featured in the NAIRL Tech Horizon report on NeurIPS 2025 and AAAI 2026 as one of the representative research outcomes from the National AI Research Lab.
Apr 9, 2026 Invited to serve as an Area Chair for the GFM workshop at ICML 2026.
Mar 16, 2026 Selected as the 2026 Jang Young-Sil Fellow Program (Excellence Track), awarded by KAIST's Office of Research Affairs. The fellowship supports outstanding postdoctoral researchers throughout 2026.
Mar 1, 2026 Invited to serve as General Co-Chair of LoG 2026.
Jan 26, 2026 “Learning Posterior Predictive Distributions for Node Classification from Synthetic Graph Priors” was accepted in ICLR 2026!
Nov 19, 2025 I presented “Tackling Over-smoothing and Over-squashing in Deep Learning on Graphs” at the “Making Decisions with AI” seminar, Pusan National University.
Nov 11, 2025 I received the Best Paper Runner-up Award from KAIST-Samsung Electronics DS Division Industry-Academic Collaboration Program for our paper “Graph Convolutions Enrich the Self-Attention in Transformers!”
Oct 29, 2025 Our paper “Are Graph Transformers Necessary? Efficient Long-Range Message Passing with Fractal Nodes in MPNNs” has been accepted as an oral presentation at AAAI 2026!
Sep 19, 2025 TV-Rec: Time-Variant Convolutional Filter for Sequential Recommendation was accepted in NeurIPS 2025!
Sep 1, 2025 I have joined the Information & Electronics Research Institute at KAIST as a Postdoctoral Researcher.
Aug 29, 2025 I successfully defended my dissertation and received my Ph.D. in Artificial Intelligence from Yonsei University under the supervision of Prof. Noseong Park and Sung-Bae Cho.
Jun 11, 2025 I have been recognized as an Outstanding Reviewer (Top 10%) for KDD 2025 (February Cycle).
May 29, 2025 “Predicting Drug-Drug Interactions: A Deep Learning Approach with GCN-based Collaborative Filtering” was accepted in Artificial Intelligence in Medicine (IF 7.5 at submission, IF 6.1 at publication)!
May 5, 2025 I will serve as a Website Chair for the Learning on Graphs (LoG) conference!
Apr 29, 2025 “Learning Advanced Self-Attention for Linear Transformers in the Singular Value Domain” was accepted in IJCAI 2025!
Mar 3, 2025 Invited to be a Web Chair for CIKM 2025!
Mar 2, 2025 Paper Digest announced that “Graph Neural Controlled Differential Equations for Traffic Forecasting” ranked 13th among the most influential papers at AAAI 2022.
Feb 27, 2025 As an expert research personnel (전문연구요원), I completed a military training program in February!
Jan 23, 2025 “PIORF: Physics-Informed Ollivier-Ricci Flow for Long-Range Interactions in Mesh Graph Neural Networks” was accepted in ICLR 2025!
Jan 13, 2025 “FraudCenGCL: Enhancing Fraud Detection via Dual-View Graph Contrastive Learning with Account Centrality Features” was accepted for a workshop at AAAI 2025.
Jan 8, 2025 I received the “Merit Academic Paper Award” from Yonsei University for PANDA (ICML 2024).
Dec 18, 2024 I have been recognized as an Outstanding Reviewer (Top 10%) for KDD 2025 (August Cycle).
Dec 3, 2024 I won a 2024 Qualcomm Innovation Fellowship!
Nov 30, 2024 I was recognized as the “Top Reviewer” from LoG 2024.
Nov 5, 2024 I am honored to receive a “Top (8%) Reviewer” award from NeurIPS 2024.
Oct 29, 2024 I am honored to receive the First Prize (Paper Track on Business Idea Competition) by the Korea Financial Telecommunications & Clearings Institute (금융결제원).
Oct 24, 2024 “SCONE: A Novel Stochastic Sampling to Generate Contrastive Views and Hard Negative Samples for Recommendation” was accepted in WSDM 2025.
Oct 23, 2024 I was qualified as a Qualcomm Innovation Fellowship Finalist in the field of AI/ML.
Oct 12, 2024 “Enhanced Fraud Detection in Bank Transfers via Augmented Account Features” was accepted in ICAIF Workshop on Foundation Models for Time Series: Exploring New Frontiers (FMTS) as an oral presentation!
Oct 4, 2024 “Graph-Based Representation Approach for Deep Learning of Organic Light-Emitting Diode Devices” was accepted in Advanced Intelligent Systems.
Sep 26, 2024 “Graph Convolutions Enrich the Self-Attention in Transformers!” was accepted in NeurIPS 2024!
Aug 12, 2024 Paper Digest announced that our paper, “Graph Neural Controlled Differential Equations for Traffic Forecasting”, ranked 14th among the most influential papers at AAAI 2022.
Jul 17, 2024 “Bridging Dynamic Factor Models and Neural Controlled Differential Equations for Nowcasting GDP” was accepted in CIKM 2024!
Jul 12, 2024 I was awarded the “Merit Academic Paper Award” by Yonsei University for my research “Blurring-Sharpening Process Models for Collaborative Filtering”, which was presented at SIGIR 2023.
May 2, 2024 PANDA: Expanded Width-Aware Message Passing Beyond Rewiring was accepted in ICML 2024!
Apr 17, 2024 SVD-AE: Simple Autoencoders for Collaborative Filtering was accepted in IJCAI 2024!
Jan 16, 2024 Learning Flexible Body Collision Dynamics with Hierarchical Contact Mesh Transformer was accepted in ICLR 2024!
Jan 15, 2024 I was awarded the “Excellent Academic Paper Award” (학과우수논문상) by Yonsei University for my research “GREAD: Graph Neural Reaction-Diffusion Networks”, which was presented at ICML 2023.
Dec 9, 2023 An Attentive Inductive Bias for Sequential Recommendation Beyond the Self-Attention was accepted in AAAI 2024!
Dec 7, 2023 The preprint of our work, Graph Convolutions Enrich the Self-Attention in Transformers!, is released in arXiv.
Nov 24, 2023 GREAD: Graph Neural Reaction-Diffusion Networks was invited to the “Top Conference Session” at KSC 2023.
Oct 27, 2023 Long-term Time Series Forecasting based on Decomposition and Neural Ordinary Differential Equations was accepted in IEEE International Conference on Big Data (Big Data 2023).
Oct 27, 2023 HyperNetwork Approximating Future Parameters for Time Series Forecasting under Temporal Drifts was accepted in NeurIPS 2023 Workshop on Distribution Shifts (DistShift).
Sep 1, 2023 I started my alternative military service as an expert research personnel (전문연구요원).
Aug 23, 2023 QoS-Aware Graph Contrastive Learning for Web Service Recommendation was accepted in APSEC 2023.
Aug 10, 2023 Blurring-Sharpening Process Models for Collaborative Filtering was invited to the “Top Conference Session” at KCC 2023.
Jun 22, 2023 I joined the Graph User Group (GUG) community as a contributor.
Jun 13, 2023 GREAD was invited for SEA-CROGS, where Sandia National Labs and Pacific Northwest National Labs are collaborating to solve scientific problems with deep learning. Prof. Noseong Park, who is my research advisor, will present the paper on the 27th of June via weekly webinars.
May 29, 2023 Graph Neural Rough Differential Equations for Traffic Forecasting was accepted in ACM Transactions on Intelligent Systems and Technology (TIST) (IF 10.489 at submission, IF 5.0 at publication).
May 12, 2023 My research work (LT-OCF) is featured in the “Top 23 Python recommender-system Projects” and “Top 14 collaborative-filtering Open-Source Projects” on LibHunt.
Apr 25, 2023 GREAD: Graph Neural Reaction-Diffusion Networks was accepted in ICML 2023.
Apr 6, 2023 Blurring-Sharpening Process Models for Collaborative Filtering was accepted in SIGIR 2023.
Jan 6, 2023 Climate Modeling with Neural Advection-Diffusion Equation was accepted in Knowledge and Information Systems (IF 2.822 at submission, IF 2.531 at publication).
Sep 28, 2022 Prediction-based One-shot Dynamic Parking Pricing was accepted in CIKM 2022.
Jun 12, 2022 Graph Neural Controlled Differential Equations for Traffic Forecasting was selected for the Innovation Awards at Yonsei University.
Apr 7, 2022 Graph Neural Controlled Differential Equations for Traffic Forecasting was invited to the “Top Conference Session” at KCC 2022.
Feb 18, 2022 Graph Neural Controlled Differential Equations for Traffic Forecasting was selected for oral presentation in AAAI 2022. The top 4.21% (380/9,020) of papers were selected for oral presentation in AAAI 2022.
Dec 13, 2021 Our IEEE ICDM 2021 paper “Climate Modeling with Neural Diffusion Equations” has been invited for journal extension in Knowledge and Information Systems (KAIS).
Dec 8, 2021 Graph Neural Controlled Differential Equations for Traffic Forecasting was accepted in AAAI 2022.
Nov 23, 2021 Linear, or Non-Linear, That is the Question! was accepted in WSDM 2022.
Aug 31, 2021 Climate Modeling with Neural Diffusion Equations was accepted in ICDM 2021.
Aug 9, 2021 LT-OCF: Learnable-Time ODE-based Collaborative Filtering was accepted in CIKM 2021.
Feb 2, 2021 Awarded the Outstanding Paper Award (우수논문상) by the 23rd Korea Conference on Software Engineering (KCSE 2021).
Research and Selected Publications
Graph Neural Networks
Graph neural networks learn from connected data such as social, molecular, and transport networks, but deep models blur node features together (over-smoothing) and squeeze long-range signals through structural bottlenecks (over-squashing). I design message passing that mitigates both problems: reaction-diffusion dynamics (GREAD, ICML 2023), wider bottleneck nodes instead of graph rewiring (PANDA, ICML 2024), and fractal nodes that summarize subgraphs for long-range reach (Fractal Nodes, AAAI 2026, Oral).Transformers and Self-Attention
Self-attention, the core of Transformers, acts like a low-pass filter: as models get deeper, token representations lose fine detail and become alike. Viewing attention as a graph filter over tokens, I restore the missing high-frequency signal at little extra cost, improving Transformers in language, vision, speech, code, and graph tasks (GFSA, NeurIPS 2024) and in sequential recommendation (BSARec, AAAI 2024).Recommender Systems
Recommendation means uncovering preferences that users have not yet shown; graph-based collaborative filtering does this by spreading known preferences over a user-item graph, but how far and how this information spreads is usually fixed by hand. I model this spread as continuous dynamics: neural ODEs with learnable time points (LT-OCF, CIKM 2021), and a training-free blurring-sharpening process inspired by score-based generative models (BSPM, SIGIR 2023).Foundation Models with Prior-Data Fitted Networks
Most models must be retrained for every new dataset. Using prior-data fitted networks pretrained only on synthetic data, I built a model that classifies nodes on unseen graphs in a single forward pass (NodePFN, ICLR 2026), and with graduate students in our lab extended the idea to sequential recommendation (SRPFN, KDD 2026, Oral).Spatiotemporal Forecasting
Traffic and sensor networks change continuously over time and space, and their readings are often irregular or missing. I forecast them with neural differential equations that treat time as continuous: controlled differential equations that process time and space together (STG-NCDE, AAAI 2022, Oral; among the most influential AAAI 2022 papers), and rough differential equations for long and irregular sequences (STG-NRDE, TIST 2023).Scientific Machine Learning
Neural simulators can make physics simulations much faster, but they struggle with collisions and with effects that reach across the whole system. We address this with a hierarchical mesh Transformer that passes contact effects quickly through the body (HCMT, ICLR 2024) and with physics-informed graph rewiring based on Ollivier-Ricci curvature for fluid simulation (PIORF, ICLR 2025).Applications
These methods also serve real-world problems: fraud detection (FraudCenGCL, IEEE Big Data 2025) and GDP nowcasting (NCDENow, CIKM 2024) in finance, drug-drug interaction prediction in biomedicine (AI in Medicine 2025), OLED device modeling in materials (Advanced Intelligent Systems 2024), climate modeling (ICDM 2021), and dynamic parking pricing in urban mobility (CIKM 2022).