Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | Department of Computing | en_US |
| dc.contributor.advisor | Li, Qing (COMP) | en_US |
| dc.contributor.advisor | Fan, Wenqi (COMP) | en_US |
| dc.creator | Wang, Lin | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/14552 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | en_US |
| dc.rights | All rights reserved | en_US |
| dc.title | Efficient graph condensation : a unified framework from kernel methods to gaussian processes | en_US |
| dcterms.abstract | The rapid development of Internet technology has given rise to a vast amount of graph-structured data. Graph Neural Networks (GNNs), as an effective method for various graph mining tasks, incur substantial computational resource costs when dealing with large-scale graph data. This computational bottleneck significantly limits the practical applications of GNNs in real-world scenarios where efficiency and scalability are crucial. To address this challenge, graph condensation has emerged as a promising data-centric solution that aims to condense large graph datasets into smaller, representative subsets without sacrificing the predictive performance of GNNs. However, existing graph condensation methods often rely on computationally intensive bi-level optimization architectures, which themselves suffer from massive computation costs and limited scalability. | en_US |
| dcterms.abstract | In this thesis, we propose a comprehensive framework that addresses the computational challenges of graph condensation through three interconnected contributions: | en_US |
| dcterms.abstract | 1. Graph Condensation with Structure-based Neural Tangent Kernel (GC-SNTK): We reformulate the graph condensation problem as a Kernel Ridge Regression (KRR) task instead of iteratively training GNNs in the inner loop of bi-level optimization. This approach utilizes a Structure-based Neural Tangent Kernel (SNTK) to capture the topology of graphs and serves as the kernel function in the KRR paradigm, significantly accelerating graph condensation while maintaining high prediction performance. | en_US |
| dcterms.abstract | 2. Simplified Graph Neural Tangent Kernel (SGTK) and Simplified Graph Neural Kernel (SGNK): We address the computational limitations of existing Graph Neural Tangent Kernel (GNTK) methods, which suffer from redundant computations due to layer-stacking strategies. SGTK replaces the traditional multi-layer stacking mechanism with a continuous K-step aggregation operation, streamlining the iterative kernel computation process while preserving expressiveness. SGNK models infinitely wide Graph Neural Networks as Gaussian Processes, allowing kernel values to be directly determined from the expected outputs of activation functions in the infinite-width regime, further reducing computational complexity. | en_US |
| dcterms.abstract | 3. Graph Condensation via Gaussian Process (GCGP): We demonstrate the synergistic potential of our kernel methods by proposing GCGP, which leverages the computational efficiency of our simplified kernels to achieve even faster graph condensation. GCGP utilizes a Gaussian Process with the condensed graph serving as observations to estimate the posterior distribution of predictions, eliminating the need for iterative and resource-intensive GNN training. We derive a specialized covariance function that incorporates structural information through local neighborhood aggregation and utilize Concrete random variables to approximate binary adjacency matrices in continuous counterparts, enabling gradient-based optimization for discrete graph structures. | en_US |
| dcterms.abstract | Our comprehensive experimental results demonstrate that this integrated approach achieves superior computational efficiency while maintaining high prediction performance across various graph mining tasks. The proposed methods collectively address the scalability challenges of Graph Neural Networks through a unified framework that combines the benefits of kernel methods with graph condensation techniques. | en_US |
| dcterms.extent | xvii, 126 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2026 | en_US |
| dcterms.educationalLevel | Ph.D. | en_US |
| dcterms.educationalLevel | All Doctorate | en_US |
| dcterms.LCSH | Neural networks (Computer science) | en_US |
| dcterms.LCSH | Kernel functions | en_US |
| dcterms.LCSH | Gaussian processes | en_US |
| dcterms.LCSH | Graph theory -- Data processing | en_US |
| dcterms.LCSH | Hong Kong Polytechnic University -- Dissertations | en_US |
| dcterms.accessRights | open access | en_US |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/14552