Research map / Architectures and efficiency / Attention and transformers
TreeGen: A Tree-Based Transformer Architecture for Code Generation
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 TreeGen-A instead of ReCode | HearthStone with plain preprocessing · HearthStone (plain) | StrAcc: 19.6 -> 25.8 | Helped |
| Use TreeGen-B instead of SZM19 | HearthStone with structural preprocessing · HearthStone (structural) | StrAcc: 27.3 -> 31.8 | Helped |
| Use TreeGen instead of the strongest neural baseline, COARSE2FINE | ATIS · ATIS | accuracy: 87.7 -> 89.1 | Helped |
| Use TreeGen instead of the strongest neural baseline, Seq2Act | GEO · GEO | accuracy: 88.9 -> 89.6 | Helped |
| Add structural convolution only to the first several AST reader blocks rather than all blocks (ablation) | TreeGen-B · HearthStone (structural) | StrAcc: 25.8 -> 27.3 | Helped |
| Add structural convolution rather than none (ablation) | TreeGen-B · HearthStone (structural) | StrAcc: 21.2 -> 31.8 | Helped |
| Add AST structure modeling instead of a Transformer without effective structure modeling (ablation) | TreeGen-B · HearthStone (structural) | StrAcc: 10.6 -> 31.8 | Helped |
| Add tree convolution (ablation) | TreeGen-B · HearthStone (structural) | StrAcc: 27.3 -> 31.8 | Helped |
| Add rule definition encoding (ablation) | TreeGen-B · HearthStone (structural) | StrAcc: 27.3 -> 31.8 | Helped |
| Add character embedding (ablation) | TreeGen-B · HearthStone (structural) | StrAcc: 15.2 -> 31.8 | Helped |
| Add self-attention layers (ablation) | TreeGen-B · HearthStone (structural) | StrAcc: 28.8 -> 31.8 | Helped |
| Add rule definition encoding (ablation) | TreeGen-B · HearthStone (structural) | BLEU: 81.8 -> 80.8 | Hurt |
Closest papers by meaning
- CodeGen: An Open Large Language Model for Code with Multi-Turn Program Synthesis (ICLR 2023 · similarity 0.88)
- A Grammar-Based Structural CNN Decoder for Code Generation (AAAI 2019 · similarity 0.87)
- CodeSteer: Symbolic-Augmented Language Models via Code/Text Guidance (ICML 2025 · similarity 0.86)
- AST-T5: Structure-Aware Pretraining for Code Generation and Understanding (ICML 2024 · similarity 0.85)
- Execution Guided Line-by-Line Code Generation (NeurIPS 2025 · similarity 0.85)
- EffiCoder: Enhancing Code Generation in Large Language Models through Efficiency-Aware Fine-tuning (ICML 2025 · 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.