Hierarchical Matrices in Graph Convolutional Deep Neural Network Context: Performance Evaluation in a Case Study

This paper explores the integration of Hierarchical Matrices (H-matrices) within Graph Convolutional Deep Neural Networks (GC-DNNs) laying the foundations to assess their performance in a specific case study. Hierarchical Matrices have been recognized for their efficiency in matrix polynomial evaluations, particularly in high-performance computing (HPC) environments. We analyze their potential to optimize computational workloads in GC-DNNs, emphasizing scalability and accuracy. The study provides preliminary insights and outlines future research directions.