Research map / Label-efficient and robust learning / Self-supervised and contrastive
DOBF: A Deobfuscation Pre-Training Objective for Programming Languages
Marie-anne Lachaux, Baptiste Roziere, Marc Szafraniec, Guillaume Lample
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 |
|---|---|---|---|
| Pre-train with DOBF+DAE instead of DOBF alone (ablation) | DOBF · CodeXGLUE Code Summarization Java | BLEU: 19.0546 -> 19.3646 | Improved |
| Pre-train with DOBF+DAE instead of DOBF alone (ablation) | DOBF · CodeXGLUE Code Summarization Python | BLEU: 18.2399 -> 18.5767 | Improved |
| Pre-train with DOBF+DAE instead of DOBF alone (ablation) | DOBF · CodeXGLUE NL Code Search Python | MRR: 0.383 -> 0.3966 | Improved |
| Pre-train with DOBF+DAE instead of DOBF alone (ablation) | DOBF · TransCoder Python→Java, beam size 10 | CA@1: 44.12 -> 47.27 | Improved |
| Pre-train with DOBF+DAE instead of DOBF alone (ablation) | DOBF · TransCoder Java→Python, beam size 10 | CA@1: 39.96 -> 42.36 | Improved |
| Pre-train with DOBF+DAE instead of DOBF alone (ablation) | DOBF · CodeXGLUE Clone Detection Java | F1: 95.87 -> 95.82 | Hurt |
| Initialize DOBF with MLM instead of training from scratch (ablation) | DOBF · Identifier Name Proposal, eval p_obf=0 | accuracy: 56.3272 -> 67.58465 | Improved |
| Train with p_obf=0.5 instead of p_obf=0 (ablation) | DOBF · Identifier Name Proposal, eval p_obf=0 | accuracy: 56.3272 -> 61.06085 | Improved |
| Train with p_obf=1 instead of p_obf=0.5 (ablation) | DOBF · Identifier Name Proposal, eval p_obf=0 | accuracy: 61.06085 -> 18.12775 | Hurt |
| Train with p_obf=1 instead of p_obf=0.5 (ablation) | DOBF · Full Deobfuscation, eval p_obf=1 | accuracy: 41.80665 -> 45.56205 | Improved |
| Initialize DOBF with MLM instead of training from scratch (ablation) | DOBF · Full Deobfuscation, eval p_obf=1 | accuracy: 45.56205 -> 49.7472 | Improved |
| Use DOBF pre-training rather than MLM pre-training | CodeXGLUE NL Code Search Python · CodeXGLUE NL Code Search Python | MRR: 0.3076 -> 0.383 | Improved |
Closest papers by meaning
- ObscuraCoder: Powering Efficient Code LM Pre-Training Via Obfuscation Grounding (ICLR 2025 · similarity 0.90)
- CODE REPRESENTATION LEARNING AT SCALE (ICLR 2024 · similarity 0.87)
- LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code (ICLR 2025 · similarity 0.86)
- Teaching Large Language Models to Self-Debug (ICLR 2024 · similarity 0.85)
- LILO: Learning Interpretable Libraries by Compressing and Documenting Code (ICLR 2024 · similarity 0.85)
- AlchemistCoder: Harmonizing and Eliciting Code Capability by Hindsight Tuning on Multi-source Data (NeurIPS 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.
