Research map / Graph learning / GNN architectures
Learning to Execute Programs with Instruction Pointer Attention Graph Neural Networks
David Bieber, Charles A. Sutton, Hugo Larochelle, Daniel Tarlow
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 |
|---|---|---|---|
| use IPA-GNN instead of NoExecute, retaining instruction-pointer attention and control flow while replacing execution components (ablation) | IPA-GNN · D_test full program execution | accuracy (%): 50.7 -> 62.1 | Improved |
| use IPA-GNN instead of NoControl, retaining interpreter-like execution while replacing control-flow components (ablation) | IPA-GNN · D_test full program execution | accuracy (%): 28.1 -> 62.1 | Improved |
| add IPA-GNN execution and control-flow components rather than use GGNN | GGNN · D_test full program execution | accuracy (%): 16.0 -> 62.1 | Improved |
| use IPA-GNN instead of Line-by-Line RNN | program execution models · D_test full program execution | accuracy (%): 32.0 -> 62.1 | Improved |
| use bounded IPA-GNN execution rather than the trace-oracle RNN | program execution models · D_test full program execution | accuracy (%): 66.4 -> 62.1 | Hurt |
| use IPA-GNN instead of NoExecute, retaining instruction-pointer attention and control flow while replacing execution components (ablation) | IPA-GNN · D_test partial program execution | accuracy (%): 20.7 -> 29.1 | Improved |
| use IPA-GNN instead of NoControl, retaining interpreter-like execution while replacing control-flow components (ablation) | IPA-GNN · D_test partial program execution | accuracy (%): 8.1 -> 29.1 | Improved |
| add IPA-GNN execution and control-flow components rather than use GGNN | GGNN · D_test partial program execution | accuracy (%): 5.7 -> 29.1 | Improved |
| use IPA-GNN instead of Line-by-Line RNN | program execution models · D_test partial program execution | accuracy (%): 11.5 -> 29.1 | Improved |
Closest papers by meaning
- LEARNING EXECUTION THROUGH NEURAL CODE FUSION (ICLR 2020 · similarity 0.88)
- Structure-Aware Graph Hypernetworks for Neural Program Synthesis (ICLR 2026 · similarity 0.86)
- Recurrent Graph Neural Networks and Their Connections to Bisimulation and Logic (AAAI 2024 · similarity 0.84)
- Exploring Rationale Learning for Continual Graph Learning (AAAI 2025 · similarity 0.83)
- On Vanishing Gradients, Over-Smoothing, and Over-Squashing in GNNs: Bridging Recurrent and Graph Learning (NeurIPS 2025 · similarity 0.83)
- Searching Latent Program Spaces (NeurIPS 2025 · similarity 0.83)
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.
