Qihao Cheng
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Qihao Cheng has a total of 17 co-authors. Here are the top co-authors by number of shared publications:
- Da Yan
Indiana University Bloomington, Bloomington, USA
3 shared publications - Tianhao Wu
Tsinghua University, Beijing, China
3 shared publications - Zhongyi Huang
Tsinghua University, Beijing, China
2 shared publications - Bo Liu1 shared publication
- Feng Yu
National University of Singapore, Singapore, Singapore
1 shared publication - Haoran Tang
Tsinghua University, Beijing, China
1 shared publication
Latest Publications
- Open Access
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 568-579
NodeBits: A Plug-and-Play Framework for Accelerating Graph Inference by Post-Hoc Binary Quantization
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 568-579
Open AccessNodeBits: A Plug-and-Play Framework for Accelerating Graph Inference by Post-Hoc Binary Quantization
- Open Access
- 8 September 2025
- Published by Association for Computing Machinery (ACM)
- p. 710-719
CompreGel: Efficient Distributed Graph Propagation via Error-Bounded Lossy Message Compression
- 8 September 2025
- Published by Association for Computing Machinery (ACM)
- p. 710-719
Open AccessCompreGel: Efficient Distributed Graph Propagation via Error-Bounded Lossy Message Compression
- Open Access
- 22 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-2
Reproducibility Report for ACM SIGMOD 2025 Paper: 'PLM4NDV: Minimizing Data Access for Number of Distinct Values Estimation with Pre-trained Language Models'
We successfully reproduced the key performance results that presented in Tables~3 and~4 of the original paper. The authors provided a GitHub repository. https://github.com/bytedance/plm4ndv that includes the implementation, detailed build and execution instructions, and links to the datasets used in the experiments. We appreciate the authors' assistance with environment setup and debugging.
- 22 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-2
Open AccessReproducibility Report for ACM SIGMOD 2025 Paper: 'PLM4NDV: Minimizing Data Access for Number of Distinct Values Estimation with Pre-trained Language Models'
We successfully reproduced the key performance results that presented in Tables~3 and~4 of the original paper. The authors provided a GitHub repository. https://github.com/bytedance/plm4ndv that includes the implementation, detailed build and execution instructions, and links to the datasets used in the experiments. We appreciate the authors' assistance with environment setup and debugging.
- 1 March 2025
- Published by AIP Publishing in Physics of Fluids
- Vol. 37 (3)
A high-resolution finite difference scheme with neural network optimized dispersion and localized adaptive dissipation
High-order accuracy and high-resolution numerical schemes have been extensively utilized in computational fluid dynamics (CFD) and computational aeroacoustics (CAA) for their ability to precisely capture wave shapes. This paper introduces a novel finite difference scheme, the Neural Network Optimized Dispersion and Localized Adaptive Dissipation (NNoD-LAD) scheme, which combines a dispersion model and a dissipation model to achieve fourth-order accuracy and exceptional resolution. The NNoD-LAD's dispersion model is designed to address different wavenumber ranges effectively. For the low-intermediate wavenumber range, the dispersion coefficient is set to ensure zero dispersion error, guaranteeing accuracy. In the high wavenumber range, a fully connected neural network optimizes the coefficient to minimize dispersion error while maintaining numerical stability. The dissipation model in the NNoD-LAD scheme adjusts dissipation flexibly based on the wavenumber and solution variation. This improves simulation capabilities in the low-intermediate wavenumber range and effectively suppresses spurious oscillations in the high wavenumber range. Additionally, a new scale information recognizer has been developed to accurately compute the local wavenumber, at half-point locations. Numerical tests demonstrate that the NNoD-LAD scheme offers higher resolution compared to previous spectrally optimized schemes, making it a promising tool for CFD and CAA applications.
- 1 March 2025
- Published by AIP Publishing in Physics of Fluids
- Vol. 37 (3)
A high-resolution finite difference scheme with neural network optimized dispersion and localized adaptive dissipation
High-order accuracy and high-resolution numerical schemes have been extensively utilized in computational fluid dynamics (CFD) and computational aeroacoustics (CAA) for their ability to precisely capture wave shapes. This paper introduces a novel finite difference scheme, the Neural Network Optimized Dispersion and Localized Adaptive Dissipation (NNoD-LAD) scheme, which combines a dispersion model and a dissipation model to achieve fourth-order accuracy and exceptional resolution. The NNoD-LAD's dispersion model is designed to address different wavenumber ranges effectively. For the low-intermediate wavenumber range, the dispersion coefficient is set to ensure zero dispersion error, guaranteeing accuracy. In the high wavenumber range, a fully connected neural network optimizes the coefficient to minimize dispersion error while maintaining numerical stability. The dissipation model in the NNoD-LAD scheme adjusts dissipation flexibly based on the wavenumber and solution variation. This improves simulation capabilities in the low-intermediate wavenumber range and effectively suppresses spurious oscillations in the high wavenumber range. Additionally, a new scale information recognizer has been developed to accurately compute the local wavenumber, at half-point locations. Numerical tests demonstrate that the NNoD-LAD scheme offers higher resolution compared to previous spectrally optimized schemes, making it a promising tool for CFD and CAA applications.
- Open Access
- 10 February 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Management of Data
- Vol. 3 (1) , 1-26
Computing Approximate Graph Edit Distance via Optimal Transport
Given a graph pair (G 1 , G 2 ), graph edit distance (GED) is defined as the minimum number of edit operations converting G 1 to G 2 . GED is a fundamental operation widely used in many applications, but its exact computation is NP-hard, so the approximation of GED has gained a lot of attention. Data-driven learning-based methods have been found to provide superior results compared to classical approximate algorithms, but they directly fit the coupling relationship between a pair of vertices from their vertex features. We argue that while pairwise vertex features can capture the coupling cost (discrepancy) of a pair of vertices, the vertex coupling matrix should be derived from the vertex-pair cost matrix through a more well-established method that is aware of the global context of the graph pair, such as optimal transport. In this paper, we propose an ensemble approach that integrates a supervised learning-based method and an unsupervised method, both based on optimal transport. Our learning method, GEDIOT, is based on inverse optimal transport that leverages a learnable Sinkhorn algorithm to generate the coupling matrix. Our unsupervised method, GEDGW, models GED computation as a linear combination of optimal transport and its variant, Gromov-Wasserstein discrepancy, for node and edge operations, respectively, which can be solved efficiently without needing the ground truth. Our ensemble method, GEDHOT, combines GEDIOT and GEDGW to further boost the performance. Extensive experiments demonstrate that our methods significantly outperform the existing methods in terms of the performance of GED computation, edit path generation, and model generalizability.
- 10 February 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Management of Data
- Vol. 3 (1) , 1-26
Open AccessComputing Approximate Graph Edit Distance via Optimal Transport
Given a graph pair (G 1 , G 2 ), graph edit distance (GED) is defined as the minimum number of edit operations converting G 1 to G 2 . GED is a fundamental operation widely used in many applications, but its exact computation is NP-hard, so the approximation of GED has gained a lot of attention. Data-driven learning-based methods have been found to provide superior results compared to classical approximate algorithms, but they directly fit the coupling relationship between a pair of vertices from their vertex features. We argue that while pairwise vertex features can capture the coupling cost (discrepancy) of a pair of vertices, the vertex coupling matrix should be derived from the vertex-pair cost matrix through a more well-established method that is aware of the global context of the graph pair, such as optimal transport. In this paper, we propose an ensemble approach that integrates a supervised learning-based method and an unsupervised method, both based on optimal transport. Our learning method, GEDIOT, is based on inverse optimal transport that leverages a learnable Sinkhorn algorithm to generate the coupling matrix. Our unsupervised method, GEDGW, models GED computation as a linear combination of optimal transport and its variant, Gromov-Wasserstein discrepancy, for node and edge operations, respectively, which can be solved efficiently without needing the ground truth. Our ensemble method, GEDHOT, combines GEDIOT and GEDGW to further boost the performance. Extensive experiments demonstrate that our methods significantly outperform the existing methods in terms of the performance of GED computation, edit path generation, and model generalizability.