Fang Sun

Fang Sun

Ph.D. @ UCLA, advised by Prof. Yizhou Sun

I am Fang Sun (孙昉), a Ph.D. student in Computer Science at University of California, Los Angeles, advised by Prof. Yizhou Sun. I received my B.Sc. in Computer Science and Technology (Summa cum laude) from Peking University, and I also completed a B.Sc. in Chemistry as part of a double degree program. During my undergrad, I worked closely with Prof. Ming Zhang and Prof. Jian Tang on GNNs for drug discovery and recommender systems.

My research focus is Enacted Molecular Intelligence: Geometric Deep Learning for Accelerating Research in Quantum Chemistry Simulation and Chemical Dynamical Systems Modeling, with applications in Materials Design and Drug Discovery.

I am actively looking for talented undergraduate and master’s students interested in research opportunities. Students from CS, Chemistry, Applied Math, and related majors are welcome. If you’re interested, please send me an email with a brief introduction, your CV, and potential research directions you’d like to work on or propose.

🔥 News

  • 2026.03: Received my M.S. in Computer Science from UCLA.
  • 2025: Our paper on Controllable Generation of Drug-like Molecular Materials got accepted to LoG 2025.
  • 2025: Our paper on Symmetry-Preserving Conformer Ensemble Networks got accepted to NeurIPS 2025.
  • 2025: Our paper on Deep Knowledge Tracing for Explainable Problem Recommendations got accepted to IJCAI 2025.
  • 2025: GF-NODE got accepted to TMLR 2025.
  • 2025.03: 3 papers got accepted to ICLR 2025 MLMP (1 oral, 2 posters).
  • 2025.02: Our paper on Automated Molecular Concept Generation using LLMs got accepted to COLING 2025.
  • 2024.08: Our survey on GNNs for Molecular Generation got accepted to Neural Network.
  • 2023.09: Started my Ph.D. at UCLA.

📝 Publications

TMLR 2025
GF-NODE

Graph Fourier Neural ODEs: Modeling Spatial-temporal Multi-scales in Molecular Dynamics

Fang Sun, Zijie Huang, Haixin Wang, Huacong Tang, Xiao Luo, Wei Wang, Yizhou Sun

Abstract: We introduce Graph Fourier Neural ODEs (GF-NODE), a novel approach for spatial-temporal multi-scale modeling. By integrating Neural ODEs with graph Fourier transforms, GF-NODE effectively captures both local and global dynamics in complex molecular systems.

LoG 2025
Multi-modal Variational Flow

Controllable Generation of Drug-like Molecular Materials with Multi-modal Variational Flow

Fang Sun, Zhihao Zhan, Hongyu Guo, Ming Zhang, Jian Tang, Yizhou Sun

Abstract: We propose GraphVF, a controllable 3D molecule generation framework that integrates both 3D geometry and 2D structures for protein-specific drug design. Our approach uses a valency-aware E(3)-GNN and a junction-tree encoder, achieving a 3.9% gain in high-affinity generation while preserving realistic bond distributions and property-oriented generation.

ICLR 2025 MLMP (Oral)
DoMiNO

DoMiNO: Down-scaling Molecular Dynamics with Neural Graph Ordinary Differential Equations

Fang Sun, Zijie Huang, Yadi Cao, Xiao Luo, Wei Wang, Yizhou Sun

Abstract: DoMiNO addresses multi-scale molecular dynamics by down-scaling simulation steps with Neural Graph ODEs. It adaptively unifies different timescales, enabling accurate short- and long-range predictions in molecular modeling. Selected as an Oral Presentation at ICLR 2025 MLMP.

SIGIR 2022
DisenCTR

DisenCTR: Dynamic Graph-based Disentangled Representation for Click-Through Rate Prediction

Yifan Wang, Yifang Qin, Fang Sun, Bo Zhang, Xuyang Hou, Ke Hu, Jia Cheng, Jun Lei, Ming Zhang

Abstract: DisenCTR leverages dynamic routing to disentangle user interests, incorporating a multivariate Hawkes process for time-aware modeling. It outperforms previous CTR methods with significant gains in recommendation accuracy.

WSDM 2023
DisenPOI

DisenPOI: Disentangling Sequential and Geographical Influence for Point-of-Interest Recommendation

Yifang Qin*, Yifan Wang*, Fang Sun*, Wei Ju, Xuyang Hou, Zhe Wang, Jia Cheng, Jun Lei, Ming Zhang (*Equal contribution)

Abstract: DisenPOI proposes a dual-graph architecture to disentangle sequential and geographical factors in POI recommendation. Contrastive learning is used to differentiate user dynamics, enabling more precise and interpretable location suggestions.

Other Publications

🎖 Honors and Awards

  • 2022 John Hopcroft Scholarship, Peking University
  • 2021 Canon Scholarship (4/25), Peking University
  • 2021 Second Award, PKU-CPC Programming Contest
  • 2019 National Scholarship (2/155), Peking University

📖 Education

💻 Work Experience

Research in Industrial Projects for Students – Sponsored by IPAM & Relay Therapeutics
2024, UCLA / Relay Therapeutics

  • Served as academic mentor, focusing on generative molecular design, structural diversity, and protein-ligand interaction analysis.

Euler-Ads: Distributed GNN Framework for Online CTR Prediction
2021-2022, Meituan Inc.

  • Developed Euler-Ads, integrating large-scale graph data into CTR prediction.
  • Deployed via A/B testing on Meituan’s advertising platform.

🛠 Technical Strengths

  • Deep Learning: PyTorch, Huggingface Transformers, PyTorch Geometric
  • Quantum Chemistry Simulation: Q-Chem, Synopsys QuantumATK, RDKit
  • Scientific Computing: Python, C++, NumPy, SciPy, CUDA
  • Infrastructure: AWS, GCP, Docker, Git

🔍 Professional Service

  • Tutorials Chair, Learning on Graphs (LoG) 2025
  • Session Chair, AI and Predictive Modeling, IEEE SoutheastCon 2026
  • Reviewer: ICML, ICLR, NeurIPS, CIKM, LoG

🌴 Personal Interests

Avid hiker and beach lover. My bias is LE SSERAFIM. I enjoy deep conversations, mentally stimulating films, and would welcome discussions in automated memecoin trading strategies. Let’s explore LA’s trails or catch a show together!