SLeDGe: Semi-Supervised Learning on Data Streams with Graph Structure Learning

2026-06-19

AI Summary

This work addresses the challenges of semi-supervised learning in evolving data streams, where labels are extremely scarce and existing approaches rely on static graph structures. To overcome these limitations under strict memory and labeling constraints, the authors propose a dynamic sparse graph neural network framework that jointly optimizes the prediction model and an adaptive graph structure. The method employs a differentiated update strategy to maintain a compact memory buffer and introduces a dynamic graph learning mechanism that automatically refines sample relationships and prunes spurious connections, thereby enabling efficient label propagation and balanced integration of old and new features. Evaluated on twelve benchmark datasets, the approach achieves average relative accuracy improvements of 31.7% and 14.8% using only 0.1% and 1% labeled data, respectively, significantly outperforming current state-of-the-art methods.

Abstract

Semi-supervised learning (SSL) on data streams is challenging due to the continuous evolution of high-volume data and the scarcity of labels. Existing methods are limited in leveraging the intrinsic relationships among samples because they typically rely on fixed similarity measures or static graph structures, which cannot capture how relationships evolve over time. We propose SLeDGe, an SSL method for data streams that jointly learns a predictive model and an adaptive graph structure under strict memory and label constraints. SLeDGe maintains compact labeled and unlabeled memories using distinct update strategies, balancing rapid adaptation to novel features with the retention of historical consistency. In addition, by encouraging sparsity in the relational graph, SLeDGe filters out spurious connections and enables effective propagation of label supervision. Across 12 datasets, SLeDGe outperforms state-of-the-art competitors, achieving average relative accuracy gains of 31.7% with 0.1% labels and 14.8% with 1% labels.

Problem

• semi-supervised learning
• data streams
• graph structure learning
• label scarcity
• dynamic relationships

Innovation

• semi-supervised learning
• data streams
• adaptive graph learning
• sparse graph
• memory-efficient learning