Hyperdimensional Representation Learning for Node Classification and Link Prediction

TL;DR

HDGL introduces a one-pass, hyperdimensional representation learning framework for graphs that maps node features into HD space and aggregates local neighborhood information with Bundling, Binding, and Rotation. This HD-based approach yields latent node embeddings that support both node classification and link prediction, achieving competitive accuracy with substantially lower computational cost than iterative GNNs. It also demonstrates robustness to feature types beyond binary and offers advantages for class-incremental learning due to the non-iterative training paradigm. The work combines theoretical HD principles with practical graph tasks, delivering a scalable, energy-efficient alternative to traditional graph representation learning methods.

Abstract

We introduce Hyperdimensional Graph Learner (HDGL), a novel method for node classification and link prediction in graphs. HDGL maps node features into a very high-dimensional space (\textit{hyperdimensional} or HD space for short) using the \emph{injectivity} property of node representations in a family of Graph Neural Networks (GNNs) and then uses HD operators such as \textit{bundling} and \textit{binding} to aggregate information from the local neighborhood of each node yielding latent node representations that can support both node classification and link prediction tasks. HDGL, unlike GNNs that rely on computationally expensive iterative optimization and hyperparameter tuning, requires only a single pass through the data set. We report results of experiments using widely used benchmark datasets which demonstrate that, on the node classification task, HDGL achieves accuracy that is competitive with that of the state-of-the-art GNN methods at substantially reduced computational cost; and on the link prediction task, HDGL matches the performance of DeepWalk and related methods, although it falls short of computationally demanding state-of-the-art GNNs.

Figures (3)

• Figure 1: Mapping node features to a binary space using Random Hyperplanes: points are assigned 1 or 0 based on their position relative to each hyperplane.
• Figure 2: Comparison of Learning Times (in seconds) i.e Across Datasets for HDGL and RelHD with GCN, GAT and SGC with Hyperparameter Tuning with Exploration of Search Space with 12 configurations. Runtime for HDGL and RelHD exclusively includes only time Elapsed for learning label representations using train/validation data.
• Figure 3: Link Prediction Performance of HDGL under Various Dimensionality Configurations.