Pinhuan Wang

Pinhuan Wang

Focused and decisive.

Posts by Collection

portfolio

publications

Swift Unfolding of Communities: GPU-Accelerated Louvain Algorithm

Published in PPoPP 25, 2025

This paper proposes GALA, a GPU-accelerated Louvain algorithm featuring a novel modularity gain-based pruning strategy and workload-aware GPU kernels to address existing pruning inaccuracies and state management inefficiencies. Experimental results demonstrate that GALA significantly reduces computations and synchronization overhead, achieving on average a 6x performance improvement over state-of-the-art approaches.

Recommended citation: Wang, Zhibin, Xi Lin, Xue Li, Pinhuan Wang, Ziheng Meng, Hang Liu, Chen Tian, and Sheng Zhong. "Swift Unfolding of Communities: GPU-Accelerated Louvain Algorithm." In Proceedings of the 30th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming, pp. 441-454. 2025. https://dl.acm.org/doi/abs/10.1145/3710848.3710884

BINGO: Radix-based Bias Factorization for Random Walk on Dynamic Graphs

Published in EuroSys 25, 2025

This paper presents Bingo, a GPU-based random walk engine designed for dynamic graphs. Unlike existing systems that struggle with updates, Bingo supports both fast streaming updates and high-throughput batch updates. To achieve this, the authors propose a radix-based bias factorization for constant-time sampling, a group-adaption method to save space, and GPU-friendly designs for efficient updates. Experiments show that Bingo significantly outperforms previous methods, with up to 271× speedup on various tasks and datasets.

Recommended citation: Wang, Pinhuan and Huan, Chengying and Wang, Zhibin and Tian, Chen and Ji, Yuede and Liu, Hang. "BINGO: Radix-based Bias Factorization for Random Walk on Dynamic Graphs." In Proceedings of the Twentieth European Conference on Computer Systems. pp. 605-620. 2025. https://doi.org/10.1145/3689031.3717456

talks

teaching