Research map / Graph learning / Knowledge and temporal graphs
Neural Relational Inference with Fast Modular Meta-learning
Ferran Alet, Erica Weng, Tomás Lozano-Pérez, Leslie Pack Kaelbling
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 |
|---|---|---|---|
| Replace variational independent edge prediction with joint simulated-annealing structure search | NRI · Charged particles edge type prediction | accuracy: 82.1 -> 88.4 | Improved |
| Replace variational encoder with modular meta-learning | NRI · Springs edge type prediction | accuracy: 99.9 -> 99.9 | No clear effect |
| Replace variational NRI with modular meta-learning | NRI · Charged particles, 10-step prediction | MSE: 3.21e-3 -> 3.11e-3 | Improved |
| Replace variational NRI with modular meta-learning | NRI · Springs, 10-step prediction | MSE: 3.29e-6 -> 3.25e-6 | No clear effect |
| Use model-based inference instead of variational approach under small training sets | NRI · Edge accuracy vs training set size | accuracy | Improved |
| Use modular meta-learning edge inference instead of correlation of raw states | Correlation(data) · Charged edge type prediction | accuracy: 55.8 -> 88.4 | Improved |
| Use LSTM hidden-state correlation instead of raw-state correlation | Correlation baseline · Springs edge type prediction | accuracy: 52.4 -> 52.7 | No clear effect |
| Use LSTM joint instead of single trajectory | LSTM · Springs 10-step prediction | MSE: 4.69e-4 -> 2.19e-5 | Improved |
| Add a meta-learned proposal function to simulated annealing (ablation) | Modular meta-learning · Search speed | scalability | Improved |
| Batch computation over many tasks using a super-graph (ablation) | Modular meta-learning · Training and evaluation time | speed | Improved |
| Add a latent unseen node optimized in the inner loop | Modular meta-learning with learned model · Springs with one particle hidden | MSE: 1.09e-3 | Improved |
| Use true graph instead of inferred graph | NRI · Charged particles prediction | MSE: 3.11e-3 -> 3.03e-3 | No clear effect |
Closest papers by meaning
- A Graph Dynamics Prior for Relational Inference (AAAI 2024 · similarity 0.91)
- Graph Metanetworks for Processing Diverse Neural Architectures (ICLR 2024 · similarity 0.86)
- MoEMeta: Mixture-of-Experts Meta Learning for Few-Shot Relational Learning (NeurIPS 2025 · similarity 0.86)
- Online Relational Inference for Evolving Multi-agent Interacting Systems (NeurIPS 2024 · similarity 0.86)
- Interpretable Relational Inference with LLM-Guided Symbolic Dynamics Modeling (NeurIPS 2026 · similarity 0.86)
- Relational Learning in Pre-Trained Models: A Theory from Hypergraph Recovery Perspective (ICML 2024 · similarity 0.86)
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.
