Research map / Architectures and efficiency / Attention and transformers
Pyramid Attention For Source Code Summarization
Lei Chai, Ming Li
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 |
|---|---|---|---|
| add multi-scale pyramid processing instead of single concatenated sequence (ablation) | PA-former · EMSE-DeepCom test | BLEU: 37.329 -> 38.848 | Improved |
| add bottom-to-top aggregation (ablation) | PA-former · EMSE-DeepCom test | BLEU: 37.532 -> 38.848 | Improved |
| add cross-granularity interaction (ablation) | PA-former · EMSE-DeepCom test | BLEU: 37.133 -> 38.848 | Improved |
| add granular-aware relative embedding (ablation) | PA-former · EMSE-DeepCom test | BLEU: 36.905 -> 38.848 | Improved |
| decode from fine-grained representation only rather than serial use of all granularities (ablation) | PA-former · EMSE-DeepCom test | BLEU: 38.729 -> 38.848 | Improved |
| decode from fine-grained representation only rather than concatenating all granularities (ablation) | PA-former · EMSE-DeepCom test | BLEU: 38.556 -> 38.848 | Improved |
| add grammar-type content to semi- and coarse-grained pyramid input (ablation) | PA-former · EMSE-DeepCom test | BLEU: 38.394 -> 38.848 | Improved |
| add grammar-type content to pyramid input (ablation) | PA-former · EMSE-DeepCom test | Rouge-L: 55.744 -> 56.095 | Improved |
| add multi-scale vs single-scale input (ablation) | PA-former · EMSE-DeepCom test | Rouge-L: 55.032 -> 56.095 | Improved |
| replace TPTrans/CAST-style full-AST single-granularity models with PA-former | PA-former · EMSE-DeepCom test | BLEU: 37.248 -> 38.848 | Improved |
| replace CAST with PA-former on FunCom | PA-former · FunCom test | BLEU: 43.580 -> 44.649 | Improved |
| PA-former vs NeuralCodeSum in human evaluation | PA-former · Human eval | naturalness: 5.12 -> 6.09 | Improved |
Closest papers by meaning
- Language Models of Code Are Few-Shot Planners and Reasoners for Multi-Document Summarization with Attribution (AAAI 2025 · similarity 0.84)
- PinNet: Pinpoint Instructive Information for Retrieval Augmented Code-to-Text Generation (ICML 2024 · similarity 0.84)
- CoRNStack: High-Quality Contrastive Data for Better Code Retrieval and Reranking (ICLR 2025 · similarity 0.83)
- AlchemistCoder: Harmonizing and Eliciting Code Capability by Hindsight Tuning on Multi-source Data (NeurIPS 2024 · similarity 0.83)
- CodeChain: Towards Modular Code Generation Through Chain of Self-revisions with Representative Sub-modules (ICLR 2024 · similarity 0.82)
- EpiCoder: Encompassing Diversity and Complexity in Code Generation (ICML 2025 · similarity 0.82)
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.
