Research map / Graph learning / GNN architectures
gLSTM: Mitigating Over-Squashing by Increasing Storage Capacity
Hugh Blayney, Alvaro Arroyo, Xiaowen Dong, Michael M. Bronstein
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add K-hop aggregation (ablation) | gLSTM · Peptides-Func | AP: 0.6030 -> 0.7250 | Improved |
| add K-hop aggregation (ablation) | gLSTM · GPP Eccentricity | log10(MSE): 0.673 -> -4.036 | Improved |
| add K-hop aggregation (ablation) | gLSTM · GPP Diameter | log10(MSE): 0.042 -> -0.715 | Improved |
| add K-hop aggregation (ablation) | gLSTM · GPP SSSP | log10(MSE): -2.836 vs -3.377 without | Hurt |
| add K-hop aggregation (ablation) | gLSTM · Peptides-Struct | MAE: 0.2638 -> 0.2527 | Improved |
| add output gate (ablation) | gLSTM · Peptides-Func | AP: 0.7086 -> 0.7250 | Improved |
| add input gate (ablation) | gLSTM · GPP Eccentricity | log10(MSE): -4.036 vs -4.72 without | Hurt |
| add input gate (ablation) | gLSTM · GPP SSSP | log10(MSE): -2.836 vs -3.27 without | Hurt |
| add forget gate (ablation) | gLSTM · Peptides-Func | AP: 0.7236 -> 0.7250 | No clear effect |
| add positional encoding (ablation) | gLSTM · Peptides-Struct | MAE: 0.2539 -> 0.2527 | No clear effect |
| use gLSTM (associative memory) instead of GCN with K-hop on NAR | GCN · Neighbor Associative Recall | accuracy | Improved |
| gLSTM vs GCN on Peptides-Struct | GCN · Peptides-Struct | MAE: 0.2460 -> 0.2527 | Hurt |
Closest papers by meaning
- Over-squashing in Spatiotemporal Graph Neural Networks (NeurIPS 2025 · similarity 0.90)
- Mitigating Over-squashing without Rewiring: A Sheaf Effective Resistance Perspective (NeurIPS 2026 · similarity 0.89)
- On Vanishing Gradients, Over-Smoothing, and Over-Squashing in GNNs: Bridging Recurrent and Graph Learning (NeurIPS 2025 · similarity 0.89)
- PANDA: Expanded Width-Aware Message Passing Beyond Rewiring (ICML 2024 · similarity 0.88)
- Get a GRIP, this will be a long TRIP: A Quantifiable Long-Range Framework for Verifying Over-squashing (NeurIPS 2026 · similarity 0.88)
- Graph Neural Networks Do Not Always Oversmooth (NeurIPS 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
