Vs-graph Achieves 5% Higher Accuracy in Graph Classification Using Hyperdimensional Computing and Scalable Vector-symbolic Representations

Graph classification represents a critical challenge across diverse scientific fields, including drug discovery and materials science, yet current methods often struggle with both accuracy and computational cost. Hamed Poursiami, Shay Snyder, and colleagues from George Mason University and Oak Ridge National Laboratory, including Guojing Cong, Thomas Potok, and Maryam Parsa, present a new approach, VS-Graph, that significantly improves both speed and performance. This innovative framework leverages hyperdimensional computing, a brain-inspired technique, to create a system that rivals the accuracy of complex graph neural networks while dramatically reducing computational demands. VS-Graph employs a novel mechanism for identifying key nodes and efficiently aggregating information from across the graph, achieving up to 450times faster training and maintaining high accuracy even with limited computational resources, thereby opening new possibilities for graph analysis on resource-constrained devices.

Hyperdimensional Computing for Robust Graph Learning

This research explores the application of Hyperdimensional Computing (HDC), also known as Vector Symbolic Architectures (VSAs), to graph machine learning. HDC is a biologically inspired computing method that uses high-dimensional random vectors to represent data, leveraging properties like distance preservation and robustness to noise for efficient computations. Key features include distributed representation, the holographic principle, similarity-based computation, and fast operations within these high-dimensional spaces. The study introduces GraphHD, an implementation of HDC designed for classifying graphs by encoding both graph structure and node features as high-dimensional vectors. Experiments on standard graph datasets demonstrate that GraphHD achieves competitive or state-of-the-art classification accuracy, with potential for faster computation and reduced memory usage, particularly for large graphs. This work highlights the potential to implement HDC algorithms on neuromorphic hardware, potentially leading to substantial energy savings and enabling real-time graph processing for applications like anomaly detection and robotics.

Spike Diffusion and Associative Message Passing

Scientists have developed VS-Graph, a new framework that combines the efficiency of hyperdimensional computing (HDC) with the expressive power of graph neural networks (GNNs). This innovative method performs graph learning entirely within a high-dimensional vector space, eliminating the need for computationally expensive gradient-based optimization. Researchers engineered a Spike Diffusion mechanism to identify key nodes based on the graph’s topology, guiding the construction of node representations that accurately reflect their position and importance. The team then implemented an Associative Message Passing scheme to aggregate information from multi-hop neighborhoods, enabling rapid and efficient information exchange. Experiments demonstrate that VS-Graph achieves competitive accuracy with modern GNNs on benchmarks such as MUTAG and DD, surpassing prior HDC approaches by 4-5%, and accelerating training by up to 450x.

VS-Graph Achieves Fast, Accurate Graph Classification

The research team developed VS-Graph, a novel framework for graph classification that achieves high performance with significantly improved computational efficiency. Experiments demonstrate that VS-Graph attains comparable or superior accuracy to existing graph neural networks on benchmark datasets, while drastically reducing both training and inference times. On the NCI1 dataset, VS-Graph completed training in under one second, compared to over five minutes for a comparable graph neural network. VS-Graph maintains high accuracy even with significant reductions in hypervector dimensionality, unlike comparison approaches which suffered accuracy loss, making it suitable for resource-constrained environments.

Vector Symbolic Graphs for Fast Classification

VS-Graph represents a significant advance in graph classification, introducing a vector-symbolic framework that bridges the gap between the computational efficiency of hyperdimensional computing and the expressive power of modern graph neural networks. The method utilizes novel techniques, including Spike Diffusion and Associative Message Passing, to construct robust graph-level representations in a single pass, achieving competitive accuracy on benchmarks such as MUTAG and DD. VS-Graph accomplishes this performance without relying on gradient-based optimization, resulting in faster training times and ultra-low inference latency. Future work will focus on exploring neuromorphic co-design and specialized hardware accelerators to further enhance the efficiency and scalability of VS-Graph’s brain-inspired hyperdimensional computations.