Research map / Probabilistic and scientific modeling / Variational and generative sampling
PHOG: Probabilistic Model for Code
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 |
|---|---|---|---|
| replace PCFG conditioning with learned PHOG context | PCFG · JavaScript terminals | error rate: 49.9% -> 18.5% | Helped |
| replace PCFG with PHOG | PCFG · JavaScript non-terminals | error rate: 48.5% -> 25.9% | Helped |
| replace best n-gram (3-gram) with PHOG | 3-gram · JavaScript non-terminals | error rate: 30.8% -> 25.9% | Helped |
| replace best n-gram (3-gram) with PHOG | 3-gram · JavaScript terminals | error rate: 28.7% -> 18.5% | Helped |
| increase n-gram order from 3 to 10 (ablation) | n-gram · JavaScript non-terminals | error rate: 30.8% -> 35.6% | Hurt |
| increase n-gram order from 3 to 10 (ablation) | n-gram · JavaScript terminals | error rate: 28.7% -> 29.0% | Hurt |
| replace SVM with PHOG | SVM linear classifier · JavaScript terminals | error rate: 29.5% -> 18.5% | Helped |
| replace SVM with PHOG | SVM linear classifier · JavaScript non-terminals | error rate: 32.5% -> 25.9% | Helped |
| use SVM instead of Naive Bayes | Naive Bayes · JavaScript terminals | error rate: 45.8% -> 29.5% | Helped |
| replace lexicalized n-gram with PHOG | lexicalized n-gram · syntactic prediction | error rate: 40.8% -> 33.5% | Helped |
| add learned TCOND function search to PCFG-like model | PCFG · training time | training time: 1 MIN -> 162 + 3 MIN | Hurt |
| replace PCFG with PHOG | PCFG · query speed | queries per second: 71, 000 -> 50, 000 | Mixed |
Closest papers by meaning
- Syntactic and Semantic Control of Large Language Models via Sequential Monte Carlo (ICLR 2025 · similarity 0.84)
- Grammars of Formal Uncertainty: When to Trust LLMs in Automated Reasoning Tasks (NeurIPS 2025 · similarity 0.83)
- Emergent Representations of Program Semantics in Language Models Trained on Programs (ICML 2024 · similarity 0.83)
- AlchemistCoder: Harmonizing and Eliciting Code Capability by Hindsight Tuning on Multi-source Data (NeurIPS 2024 · similarity 0.83)
- DyCodeEval: Dynamic Benchmarking of Reasoning Capabilities in Code Large Language Models Under Data Contamination (ICML 2025 · similarity 0.83)
- CodeSteer: Symbolic-Augmented Language Models via Code/Text Guidance (ICML 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.