Faster Inference Time for GNNs using coarsening

2024-10-19

arXiv.org

AI Summary

To address the low inference efficiency and high memory overhead of large-scale Graph Neural Networks (GNNs), this paper proposes an end-to-end acceleration framework based on graph coarsening. It is the first work to systematically adapt coarsening techniques to graph-level classification and regression tasks. The framework introduces a dual-path lightweight inference mechanism—comprising Extra-Nodes and Cluster-Nodes—that significantly reduces per-node computational load. Crucially, cost reduction is achieved jointly during both training and inference phases. On benchmark datasets with up to 100K nodes, the method retains accuracy comparable to full-scale GNNs while accelerating inference by 10–100× and substantially reducing GPU memory consumption—enabling deployment on resource-constrained devices. Key contributions include: (i) a novel coarsening adaptation mechanism tailored for graph-level tasks, and (ii) the dual-path inference architecture that balances expressivity and efficiency.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data
• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies

Abstract

Scalability of Graph Neural Networks (GNNs) remains a significant challenge, particularly when dealing with large-scale graphs. To tackle this, coarsening-based methods are used to reduce the graph into a smaller graph, resulting in faster computation. Nonetheless, prior research has not adequately addressed the computational costs during the inference phase. This paper presents a novel approach to improve the scalability of GNNs by reducing computational burden during both training and inference phases. We demonstrate two different methods (Extra-Nodes and Cluster-Nodes). Our study also proposes a unique application of the coarsening algorithm for graph-level tasks, including graph classification and graph regression, which have not yet been explored. We conduct extensive experiments on multiple benchmark datasets in the order of $100K$ nodes to evaluate the performance of our approach. The results demonstrate that our method achieves competitive performance in tasks involving classification and regression on nodes and graphs, compared to traditional GNNs, while having single-node inference times that are orders of magnitude faster. Furthermore, our approach significantly reduces memory consumption, allowing training and inference on low-resource devices where traditional methods struggle.

Problem

• Graph Neural Networks
• Computational Efficiency
• Resource-constrained Devices

Innovation

• FIT-GNN
• Extra-Nodes and Cluster-Nodes
• Efficiency and Memory Reduction