AI

EDiS: Edge Disjoint Subgraph Sparsification Framework for Graph Neural Networks

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 09/10/2026, 10:18:24 UTC.

Briefings

Edge Disjoint Subgraph Sparsification Framework (EDiS) separates one-time structural extraction from per-epoch graph composition to reduce training costs in Graph Neural Networks. The framework decomposes graphs into cacheable edge-disjoint subgraphs, enabling efficient recombination across epochs while maintaining edge-budget constraints. On 19 benchmarks against 17 baselines, EDiS achieved the highest mean accuracy/ROC-AUC and lowest average rank, with ablations confirming benefits of structural decomposition and epoch variation at tight edge budgets.

Key points

• EDiS decomposes graphs into cacheable edge-disjoint subgraphs, then recombines them into graphs with edge-budget constraints across epochs without re-extracting structure.
• Under the default covering-forest selector, the stored decomposition deterministically preserves high-score cut edges, and a selector-agnostic conditional bound on high-score cut survival was derived.

Context

Sparse GNN training reduces computation but deciding which edges to keep is costly. Reusing one sparse graph is cheap but locks training to a fixed topology, while varying it across epochs can require repeated sampling or recomputation. EDiS addresses this by separating structural extraction from per-epoch composition.

Limitations

Evidence covers 19 benchmarks across 17 baselines under the same edge budget. Results are limited to homophilic, heterophilic, and large-scale node classification tasks. The study does not report results on other graph types or specific edge selection rules beyond the default covering-forest selector.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 15/11/2026, 22:03:13 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeiamzsvkjxhjio67yyyz5cdotaav3tkgar45td3jgwz3wrwunrjcmy