Research map / Graph learning / Graph self-supervised learning
Structure-Aware Graph Hypernetworks for Neural Program Synthesis
Wenhao Li, Yudong Xu, Elias Boutros Khalil, Scott Sanner
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 structure-aware message passing (Meta-GNN vs Meta-MLP) | hypernetwork · SUMFIRST-n OOD test | accuracy: 0.145 -> 0.866 | Improved |
| add group-aware parameter tying (Meta-gMLP vs Meta-MLP) | hypernetwork · SUMFIRST-n OOD test | accuracy: 0.145 -> 0.344 | Improved |
| synthesize weights with meta-learner instead of direct prediction | MLP · SUMFIRST-n OOD test | accuracy: 0.140 -> 0.866 | Improved |
| synthesize weights with Meta-MLP instead of direct Transformer | Transformer · ADDMOD-p OOD test | accuracy: 0.682 -> 0.947 | Improved |
| use Meta-GNN instead of Meta-MLP | hypernetwork · ADDMOD-p OOD test | accuracy: 0.947 -> 0.943 | No clear effect |
| use Meta-GNN instead of direct Transformer | Transformer · 1D-ARC mixture OOD test | accuracy: 0.603 -> 0.713 | Improved |
| use Meta-GNN instead of Meta-MLP | hypernetwork · 1D-ARC Flip OOD | accuracy: 0.127 -> 0.898 | Improved |
| use meta-learners instead of direct Transformer | Transformer · 1D-ARC MoveRight OOD | accuracy: 0.000 -> 0.996 | Improved |
| use meta-learners instead of direct Transformer | Transformer · 1D-ARC RecolorBar OOD | accuracy: 0.016 -> 0.000 | Hurt |
| use attention aggregation instead of average (ablation) | Meta-GNN · ADDMOD-p small ablation | accuracy: 0.949 -> 0.976 | Improved |
| use attention aggregation instead of sum (ablation) | Meta-GNN · ADDMOD-p small ablation | accuracy: 0.947 -> 0.976 | Improved |
| use separate parameters for reverse edges instead of shared (ablation) | Meta-GNN · ADDMOD-p small ablation | accuracy: 0.976 -> 0.979 | Improved |
Closest papers by meaning
- Graph Metanetworks for Processing Diverse Neural Architectures (ICLR 2024 · similarity 0.88)
- Scaling Optimization-Oriented Hypernetworks for Implicit Neural Representations (NeurIPS 2026 · similarity 0.87)
- Substructure Aware Graph Neural Networks (AAAI 2023 · similarity 0.87)
- HYSYNTH: Context-Free LLM Approximation for Guiding Program Synthesis (NeurIPS 2024 · similarity 0.86)
- Beyond Weisfeiler-Lehman: A Quantitative Framework for GNN Expressiveness (ICLR 2024 · similarity 0.85)
- Searching Latent Program Spaces (NeurIPS 2025 · similarity 0.85)
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.
