Research map / Large language models / Language, safety and interpretability
Leveraging Automated Unit Tests for Unsupervised Code Translation
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 |
|---|---|---|---|
| Fine-tune the unsupervised model on translations that pass automatically generated unit tests | TransCoder and DOBF · TransCoder test set, average across language directions | CA@1: 52.24833333 -> 64.68 | Helped |
| Continue offline self-training from iteration 1 to iteration 3 | Offline ST · TransCoder test set, average across language directions | CA@1: 60.75833333 -> 61.698 | Helped |
| Continue offline self-training from iteration 3 to iteration 4 | Offline ST · TransCoder test set, average across language directions | CA@1: 61.698 -> 58.50666667 | Hurt |
| Use online rather than offline self-training | Unit-test-filtered self-training · TransCoder test set, average across language directions | CA@1: 61.698 -> 64.68 | Helped |
| Add a cache for previously accepted training examples (ablation) | Online ST · TransCoder test set, average across language directions | CA@1: 59.35833333 -> 64.68 | Helped |
| Initialize the cache with offline parallel examples rather than starting empty (ablation) | Online ST · TransCoder test set, average across language directions | CA@1: 60.07833333 -> 64.68 | Helped |
| Require a minimum mutation score when selecting tests (ablation) | Online ST with no initial cache · TransCoder test set, average across language directions | CA@1: 58.68 -> 60.07833333 | Helped |
| Generate new training examples with beam search rather than greedy decoding (ablation) | Online ST · TransCoder test set, average across language directions | CA@1: 62.66333333 -> 64.68 | Helped |
| Evaluate with beam size 10 instead of greedy decoding (ablation) | Online ST with beam 20 example generation · TransCoder test set, average across language directions | CA@1: 64.68 -> 66.27166667 | Helped |
Closest papers by meaning
- Automatically Testing Functional Properties of Code Translation Models (AAAI 2024 · similarity 0.88)
- QiMeng-MuPa: Mutual-Supervised Learning for Sequential-to-Parallel Code Translation (NeurIPS 2025 · similarity 0.86)
- MT-Ranker: Reference-free machine translation evaluation by inter-system ranking (ICLR 2024 · similarity 0.85)
- ObscuraCoder: Powering Efficient Code LM Pre-Training Via Obfuscation Grounding (ICLR 2025 · similarity 0.85)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.85)
- Teaching Large Language Models to Self-Debug (ICLR 2024 · similarity 0.85)
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.