Zhewei Wei (魏哲巍)

Streaming Algorithms

1. Optimal Matrix Sketching over Sliding Windows

   Hanyan Yin, Dongxie Wen, Jiajun Li, Zhewei Wei^*, Xiao Zhang, Zengfeng Huang, and Feifei Li

   International Conference on Very Large Data Bases (VLDB), 2024. (Best Paper Nomination)

   [TLDR] [arXiv] [Code] [Poster] [Slides]

   We introduce the DS-FD algorithm, which achieves the optimal \(O\left(d/\varepsilon\right)\) space bound for matrix sketching over sliding windows.

2. Persistent Summaries

   Tianjing Zeng, Zhewei Wei^*, Ge Luo, Ke Yi, Xiaoyong Du, and Ji-Rong Wen

   ACM Transactions on Database Systems (TODS), 2022

   [TLDR]

   We propose the persistent summaries for streaming algorithms, enabling them to answer queries about the streaming data at any prior time.

3. **Mergeable summaries

   Pankaj K. Agarwal, Graham Cormode, Zengfeng Huang, Jeff Phillips, Zhewei Wei, and Ke Yi^*

   ACM Symposium on Principles of Database Systems (PODS), 2012. (Test of Time Award)

   [TLDR] [Poster] [Slides]

   We study the mergeability of data summaries and propose methods to make summaries mergeable while preserving error and size guarantees.

Graph Learning

1. PolyGCL: GRAPH CONTRASTIVE LEARNING via Learnable Spectral Polynomial Filters

   Jingyu Chen, Runlin Lei, and Zhewei Wei^*

   International Conference on Learning Representations (ICLR), 2024. (Spotlight)

   [TLDR] [Code]

   We propose POLYGCL, a GCL pipeline that utilizes polynomial filters to achieve contrastive learning between the low-pass and highpass views.

2. Convolutional Neural Networks On Graphs With Chebyshev Approximation, Revisited

   Mingguo He, Zhewei Wei^*, and Ji-Rong Wen

   Annual Conference on Neural Information Processing Systems (NeurIPS), 2022. (Oral)

   [TLDR] [arXiv] [Code]

   We revisit the problem of approximating the spectral graph convolutions with Chebyshev polynomials and then propose ChebNetII, a new GNN model based on Chebyshev interpolation, which enhances the original Chebyshev polynomial approximation while reducing the Runge phenomenon.

3. Bernnet: Learning Arbitrary Graph Spectral Filters via Bernstein Approximation

   Mingguo He, Zhewei Wei^*, Zengfeng Huang, and Hongteng Xu^*

   Annual Conference on Neural Information Processing Systems (NeurIPS), 2021

   [TLDR] [arXiv] [Code]

   We propose the BernNet, a graph neural network that provides a simple and intuitive mechanism for designing and learning an arbitrary spectral filter via Bernstein polynomial approximation.

4. Simple and Deep Graph Convolutional Networks

   Ming Chen, Zhewei Wei^*, Zengfeng Huang, Bolin Ding, and Yaliang Li

   International Conference on Machine Learning (ICML), 2020. (World Artificial Intelligence Conference Youth Outstanding Paper Nomination Award)

   [TLDR] [arXiv] [Code]

   We propose the GCNII, an extension of the vanilla GCN model with two simple yet effective techniques: Initial residual and Identity mapping.

Graph Algorithms

1. **Revisiting Local Computation of PageRank: Simple and Optimal

   Hanzhi Wang, Zhewei Wei^*, Ji-Rong Wen, and Mingji Yang

   Annual ACM Symposium on Theory of Computing (STOC), 2024

   [TLDR] [arXiv] [Video] [Slides]

   We use simple techniques and analyses to give matching upper and lower bounds for estimating PageRank contributions and single-node PageRank. Our results for the upper bounds are derived by revisiting the known algorithms of ApproxContributions (a.k.a. Backward Push) and BiPPR.

2. Personalized PageRank to a Target Node, Revisited

   Hanzhi Wang, Zhewei Wei^*, Junhao Gan, Sibo Wang, and Zengfeng Huang

   ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), 2020

   [TLDR] [arXiv] [Slides]

   We propose Randomized Backward Search (RBS), a novel algorithm that answers approximate single-target personalized PageRank queries (also known as PageRank contribution queries) with near optimal time complexity.

3. Exact Single-Source SimRank Computation on Large Graphs

   Hanzhi Wang, Zhewei Wei^*, Ye Yuan, Xiaoyong Du, and Ji-Rong Wen

   ACM Conference on Management of Data (SIGMOD), 2020

   [TLDR] [arXiv] [Code] [Slides]

   We propose ExactSim, the first algorithm that enables probabilistic exact single-source SimRank queries on large graphs. ExactSim can provide the ground truth with a precision up to 7 decimal places for single-source SimRank queries on large graphs within a reasonable query time.

Data Structures

1. Tight Space Bounds for Two-Dimensional Approximate Range Counting

   Zhewei Wei, and Ke Yi

   ACM Transactions on Algorithms (TALG), 2018

   [TLDR]

   We study the problem of two-dimensional orthogonal range counting with additive error and give tight space bounds.

2. Indexing for Summary Queries: Theory and Practice

   Ke Yi, Lu Wang, and Zhewei Wei^*

   ACM Transactions on Database Systems (TODS), 2014

   [TLDR]

   We design efficient indexing techniques that enable extracting statistical summaries, such as frequent items, quantiles, and sketches with linear space requirements and optimal or near-optimal query cost.

3. **Cache-Oblivious Hashing
   Rasmus Pagh, Zhewei Wei, Ke Yi, and Qin Zhang
   ACM Symposium on Principles of Database Systems (PODS), 2010
   [Slides]

AI for Science

1. EquiPocket: an E(3)-Equivariant Geometric Graph Neural Network for Ligand Binding Site Prediction

   Yang Zhang, Wenbing Huang^*, Zhewei Wei^*, Ye Yuan, and Chongxuan Li

   International Conference on Machine Learning (ICML), 2024. (Oral)

   [TLDR] [arXiv]

   We propose EquiPocket, the first equivariant GNN-based method for ligand binding site prediction.

2. Uni-Mol: A Universal 3D Molecular Representation Learning Framework

   Gengmo Zhou, Zhifeng Gao, Qiankun Ding, Hang Zheng, Hongteng Xu, Zhewei Wei, Linfeng Zhang, and Guolin Ke

   International Conference on Learning Representations (ICLR), 2023

   [TLDR]

   Uni-Mol is the first pure 3D molecular pretraining framework that can predict 3D positions, and the first molecular pretraining framework that can be directly used in 3D tasks in the field of drug design.