Research map / Large language models / Language, safety and interpretability
Neural Code Comprehension: A Learnable Representation of Code Semantics
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 inst2vec instead of TBCNN | POJ-104 algorithm classification · POJ-104 test | test accuracy [%]: 94.0 -> 94.83 | Helped |
| Use XFG context size 2 rather than context size 1 (ablation) | inst2vec · inst2vec analogy and semantic tests | semantic distance test: 60.98\thinspace\% -> 79.12\thinspace\% | Helped |
| Use XFG context size 2 rather than context size 3 (ablation) | inst2vec · inst2vec analogy and semantic tests | semantic distance test: 62.56\thinspace\% -> 79.12\thinspace\% | Helped |
| Combine data and control flow in XFG rather than using data flow only (ablation) | inst2vec · inst2vec analogy and semantic tests | semantic distance test: 57.44\thinspace\% -> 79.12\thinspace\% | Helped |
| Combine data and control flow in XFG rather than using control flow only (ablation) | inst2vec · inst2vec analogy and semantic tests | semantic distance test: 50.47\thinspace\% -> 79.12\thinspace\% | Helped |
| Include immediate values using concat_naïve (ablation) | inst2vec · AMD Tahiti 7970 | prediction accuracy [%]: 82.79 -> 88.09 | Helped |
| Process immediate values separately and concatenate after program characterization (ablation) | inst2vec · AMD Tahiti 7970 | prediction accuracy [%]: 82.79 -> 76.18 | Hurt |
| Concatenate immediate values then project them to embedding dimension (ablation) | inst2vec · AMD Tahiti 7970 | prediction accuracy [%]: 82.79 -> 72.06 | Hurt |
| Include immediate values using concat_naïve (ablation) | inst2vec · AMD Radeon HD 5900 | speedup: 1.37 -> 1.21 | Hurt |
| Include immediate values using extract_concat (ablation) | inst2vec · AMD Tahiti 7970 | speedup: 1.10 -> 1.18 | Helped |
| Include immediate values using extract_concat (ablation) | inst2vec · NVIDIA GTX 480 | speedup: 1.07 -> 1.11 | Helped |
| Include immediate values using concat_embed (ablation) | inst2vec · NVIDIA Tesla K20c | speedup: 1.06 -> 0.99 | Hurt |
Closest papers by meaning
- CODE REPRESENTATION LEARNING AT SCALE (ICLR 2024 · similarity 0.90)
- Emergent Representations of Program Semantics in Language Models Trained on Programs (ICML 2024 · similarity 0.89)
- Towards A Generalist Code Embedding Model Based On Massive Data Synthesis (NeurIPS 2025 · similarity 0.88)
- JanusCoder: Towards a Foundational Visual-Programmatic Interface for Code Intelligence (ICLR 2026 · similarity 0.87)
- CoRNStack: High-Quality Contrastive Data for Better Code Retrieval and Reranking (ICLR 2025 · similarity 0.87)
- Chain of Execution Supervision Promotes General Reasoning in Large Language Models (NeurIPS 2025 · similarity 0.87)
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.