Research map / Graph learning / Knowledge and temporal graphs
LEARNING EXECUTION THROUGH NEURAL CODE FUSION
Zhan Shi, Kevin Swersky, Daniel Tarlow, Parthasarathy Ranganathan, Milad Hashemi
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 |
|---|---|---|---|
| Fuse assembly structure with dynamic memory snapshots in a GNN | Neural Code Fusion · SPEC CPU2006 branch prediction | MPKI: 26% lower than the perceptron; 22% lower than the MLP | Improved |
| Fuse assembly structure with dynamic memory snapshots in a GNN | Neural Code Fusion · SPEC CPU2006 prefetching | prefetching performance: 45% | Improved |
| Pretrain NCF embeddings on branch prediction and use them with a linear SVM | Algorithm classification · Algorithm classification test | accuracy: 95.3% -> 96.0% | Improved |
| Remove pseudo nodes other than src and tgt, connecting variable nodes directly to src and tgt (ablation) | Neural Code Fusion · SPEC CPU2006 branch prediction | MPKI: MPKI increase of 0.26 | Hurt |
| Remove pseudo nodes other than src and tgt, connecting variable nodes directly to src and tgt (ablation) | Neural Code Fusion · SPEC CPU2006 prefetching | complete accuracy: accuracy goes down by 12.1% | Hurt |
| Use 5-8 GNN propagation steps (ablation) | Neural Code Fusion · omnetpp | prefetching accuracy | Improved |
| Represent dynamic values as binary rather than categorical (ablation) | MLP branch predictor in the loop generalization test · Loop generalization test, k values 1 to 80 | correct predictions across tested k values | Improved |
| Represent dynamic values as scalar or binary rather than categorical (ablation) | MLP branch predictor in the loop generalization test · Loop generalization test, unseen k values | generalization | Improved |
| Train a model jointly on branch prediction and prefetching (ablation) | Neural Code Fusion · Branch prediction and prefetching | task performance | No clear effect |
Closest papers by meaning
- Learning to Execute Programs with Instruction Pointer Attention Graph Neural Networks (NeurIPS 2020 · similarity 0.88)
- Neural Code Comprehension: A Learnable Representation of Code Semantics (NeurIPS 2018 · similarity 0.85)
- Global Relational Models of Source Code (ICLR 2020 · similarity 0.85)
- Learning to Represent Programs with Graphs (ICLR 2018 · similarity 0.84)
- Multi-View Graph Representation for Programming Language Processing: An Investigation into Algorithm Detection (AAAI 2022 · similarity 0.84)
- Static Prediction of Runtime Errors by Learning to Execute Programs with External Resource Descriptions (ICLR 2023 · similarity 0.84)
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.
