Research map / Graph learning / Graph self-supervised learning
PLUR: A Unifying, Graph-Based View of Program Learning, Understanding, and Repair
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 PLUR autoregressive ToCoPo decoder with relational encoder instead of GREAT's independent heads | variable misuse localization+repair · VarMisuseH | Loc.+Repair Acc.: 80.4% -> 82.8% (±0.10) | Helped |
| use PLUR unified model instead of Hoppity custom model | program repair · Hoppity | Sequence Acc.: 14.2% -> 26.6% (±0.32) | Helped |
| use PLUR instead of code2seq custom model | function naming · code2seq med | F1: 53.2 -> 54.2 (±0.13) | Helped |
| use PLUR instead of CuBERT no-pretraining Transformer | CuBERT classification · CuBERT-FD | Classification Acc.: 91.0% -> 86.6% (±0.15) | Hurt |
| use PLUR instead of Retrieve&Edit with retrieval | code completion · Retrieve&Edit | Avg length: 5.8/17.6 -> 2.5/10.3 | Hurt |
| add copy mechanism to decoder (ablation) | GREAT2ToCoPo · VarMisuseH | sequence accuracy: 79.1% -> 87.1% | Helped |
| add pointer mechanism to decoder (ablation) | GREAT2ToCoPo · VarMisuseH | sequence accuracy: 85.0% -> 87.1% | Helped |
| add copy mechanism to decoder (ablation) | GREAT2ToCoPo · CuBERT-VMR | sequence accuracy: 83.5% -> 85.7% | Helped |
| use GREAT relational encoder instead of vanilla Transformer (ablation) | PLUR · code2seq small | validation sequence accuracy: 23.8% -> 31.6% | Helped |
| use GGNN encoder instead of GREAT on sequence-only input (ablation) | PLUR · CuBERT-VMR | validation sequence accuracy: 85.7% -> 61.5% | Hurt |
| use set-of-paths graph instead of compact leaves-only representation (ablation) | PLUR GREAT · code2seq med | Test F1: 51.2 -> 56.4 | Helped |
| multi-task training over six CuBERT tasks (per task checkpoint) (ablation) | PLUR · CuBERT-EC | accuracy: 62.7% -> 64.8% | Helped |
Closest papers by meaning
- Global Relational Models of Source Code (ICLR 2020 · similarity 0.88)
- Dynamic Neural Program Embeddings for Program Repair (ICLR 2018 · similarity 0.87)
- CoRNStack: High-Quality Contrastive Data for Better Code Retrieval and Reranking (ICLR 2025 · similarity 0.85)
- Counterexample Guided Program Repair Using Zero-Shot Learning and MaxSAT-based Fault Localization (AAAI 2025 · similarity 0.85)
- LILO: Learning Interpretable Libraries by Compressing and Documenting Code (ICLR 2024 · similarity 0.85)
- Toward Interactive Understanding of Code APIs (NeurIPS 2026 · 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.