Mitigating Bias in Graph Hyperdimensional Computing

TL;DR

This work addresses bias in graph hyperdimensional computing by introducing Fair-GHDC, a fairness-aware training framework that debiases updates to class hypervectors via a scalar fairness factor derived from a demographic-parity regularizer. The method operates entirely in the hypervector space, avoiding encoder modification or backpropagation, and yields substantial reductions in demographic-parity and equal-opportunity gaps while preserving competitive node classification accuracy. Empirical results on six benchmark fairness datasets show Fair-GHDC delivering favorable fairness-utility trade-offs and up to ~10x training speedups on large graphs compared to GNN baselines. Overall, the approach demonstrates scalable, efficient debiasing for graph-structured data within the hyperdimensional computing paradigm, with strong potential for on-device and large-scale applications.

Abstract

Graph hyperdimensional computing (HDC) has emerged as a promising paradigm for cognitive tasks, emulating brain-like computation with high-dimensional vectors known as hypervectors. While HDC offers robustness and efficiency on graph-structured data, its fairness implications remain largely unexplored. In this paper, we study fairness in graph HDC, where biases in data representation and decision rules can lead to unequal treatment of different groups. We show how hypervector encoding and similarity-based classification can propagate or even amplify such biases, and we propose a fairness-aware training framework, FairGHDC, to mitigate them. FairGHDC introduces a bias correction term, derived from a gap-based demographic-parity regularizer, and converts it into a scalar fairness factor that scales the update of the class hypervector for the ground-truth label. This enables debiasing directly in the hypervector space without modifying the graph encoder or requiring backpropagation. Experimental results on six benchmark datasets demonstrate that FairGHDC substantially reduces demographic-parity and equal-opportunity gaps while maintaining accuracy comparable to standard GNNs and fairness-aware GNNs. At the same time, FairGHDC preserves the computational advantages of HDC, achieving up to about one order of magnitude ($\approx 10\times$) speedup in training time on GPU compared to GNN and fairness-aware GNN baselines.

Figures (4)

• Figure 1: The overall framework of graph HDC in Fair-GHDC, including four steps: (1) Feature Hypervector Encoding, (2) Edge Hypervector Encoding, (3) Node Hypervector Representation, (4) Training, and (5) Inference.
• Figure 2: Trade-off between fairness and node classification accuracy across six datasets. Results in the upper left corner, which exhibit lower bias and higher accuracy, represent the ideal balance.
• Figure 3: Runtime performance improvement of Fair-GHDC over baseline methods on GPU. Bars show the speedup (runtime(baseline)/runtime(Fair-GHDC)) on each dataset. For the Pokec-n and Pokec-z datasets, EDITS encountered an out-of-memory (OOM) issue. The corresponding wall-clock training times (in seconds) are reported in \ref{['tab:hdc_runtime_seconds']}. Slashed bars $\vcenter{}$ represent general GNN methods, while blank bars $\vcenter{}$ represent fairness-aware GNN methods.
• Figure 4: Parameter sensitivity analysis on Pokec-n (first row) and Credit (second row).