GNN Inference
GNN inference focuses on efficiently executing trained Graph Neural Networks (GNNs), aiming to overcome computational bottlenecks and improve scalability for real-world applications. Current research emphasizes optimizing inference speed and energy efficiency through techniques like personalized scoping to adapt to graph heterophily, gradient-based approximations for node influence analysis, and hardware acceleration using FPGAs and specialized architectures tailored to exploit graph sparsity. These advancements are crucial for deploying GNNs in resource-constrained environments and handling massive graphs, impacting fields like social network analysis, fraud detection, and drug discovery.
16papers
Papers
March 8, 2025
March 13, 2024
August 4, 2023
July 1, 2023
November 28, 2022
April 27, 2022
March 7, 2022
January 23, 2022