Research map / Graph learning / GNN architectures
Agent-based Graph Neural Networks
Karolis Martinkus, Pál András Papp, Benedikt Schesch, Roger Wattenhofer
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 |
|---|---|---|---|
| replace learned agent transitions with uniformly random neighbor walks (ablation) | AgentNet · 2-WL synthetic (Rook 4x4 vs Shrikhande) | accuracy: 100.0 -> 50.5 | Hurt |
| restrict transitions to coarse choices (stay, previous, explored/unexplored) instead of full dot-product attention (ablation) | AgentNet · 2-WL synthetic | accuracy: 100.0 vs 100.0 | No clear effect |
| learned attention-based agent transitions (vs random walk) (ablation) | AgentNet · 4-cycles and Circular Skip Links | accuracy: Random walk 100.0 vs full 100.0 | No clear effect |
| use k≈n agents with l steps (full AgentNet) vs single agent (ablation) | AgentNet · RDT-B and DD | accuracy: RDT-B 77.9 -> 94.2; DD 67.4 -> 80.1 | Improved |
| full AgentNet (k≈n agents) vs GIN on TU datasets | AgentNet · MUTAG, PTC, PROTEINS, IMDB-B, IMDB-M, DD, RDT-B | accuracy: improves on MUTAG (+4.2), PTC (+2.8), DD (+3.2), RDT-B (+1.8), PROTEINS (+0.5), IMDB-B (+0.1); IMDB-M (-0.1) | Mixed |
| full AgentNet vs GIN+virtual node / ESAN etc. on OGB-MolHIV | AgentNet · OGB-MolHIV test | ROC-AUC: 78.33 vs ESAN 78.00 and CRaWl 77.78 | Improved |
| single agent using only one agent (1-AgentNet) on MUTAG and RDT-B (ablation) | AgentNet · MUTAG, RDT-B | accuracy: MUTAG 89.4 (vs full 93.6); RDT-B 77.9 (vs full 94.2) | Mixed |
Closest papers by meaning
- A New Perspective on "How Graph Neural Networks Go Beyond Weisfeiler-Lehman?" (ICLR 2022 · similarity 0.90)
- Cycle to Clique (Cy2C) Graph Neural Network: A Sight to See beyond Neighborhood Aggregation (ICLR 2023 · similarity 0.89)
- Graph Agent Network: Empowering Nodes with Inference Capabilities for Adversarial Resilience (AAAI 2025 · similarity 0.88)
- AutoSGNN: Automatic Propagation Mechanism Discovery for Spectral Graph Neural Networks (AAAI 2025 · similarity 0.87)
- What do Graph Neural Networks learn? Insights from Tropical Geometry (NeurIPS 2024 · similarity 0.87)
- Deep Homomorphism Networks (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.
