Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Uniform Semantic Features for Natural Language and Programming Language Globally, Locally and Sequentially
Yudong Zhang, Wenhao Zheng, 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 |
|---|---|---|---|
| remove global information (ablation) | UniEmbed · dup-question | AUC: 95.44% -> 60.75% | Hurt |
| remove global information (ablation) | UniEmbed · code-clone | AUC: 87.69% -> 50.00%∗ | Hurt |
| remove local information (ablation) | UniEmbed · dup-question | AUC: 95.44% -> 64.81% | Hurt |
| remove local information (ablation) | UniEmbed · code-anno | F-measure: 71.85% -> 47.54% | Hurt |
| remove sequential information (ablation) | UniEmbed · dup-question | F-measure: 87.70% -> 63.88% | Hurt |
| remove sequential information (ablation) | UniEmbed · code-anno | AUC: 78.21% -> 49.95% | Hurt |
| remove attention layer (ablation) | UniEmbed · dup-question | AUC: 95.44% -> 82.66% | Hurt |
| remove attention layer (ablation) | UniEmbed · code-clone | F-measure: 80.31% -> 74.40% | Hurt |
| remove attention layer (ablation) | UniEmbed · code-anno | AUC: 78.21% -> 65.59% | Hurt |
| joint training on all three datasets instead of single-task training | UniEmbed · code-clone | AUC: 70.50% -> 87.69% | Improved |
| joint training instead of single-task training | UniEmbed · dup-question | AUC: 92.64% -> 95.44% | Improved |
| joint training instead of single-task training | UniEmbed · code-anno | AUC: 75.15% -> 78.21% | Improved |
Closest papers by meaning
- Language-Agnostic Representation Learning of Source Code from Structure and Context (ICLR 2021 · similarity 0.86)
- Learning and Evaluating Contextual Embedding of Source Code (ICML 2020 · similarity 0.85)
- Neural Code Comprehension: A Learnable Representation of Code Semantics (NeurIPS 2018 · similarity 0.84)
- Emergent Representations of Program Semantics in Language Models Trained on Programs (ICML 2024 · similarity 0.83)
- DiscoveryBench: Towards Data-Driven Discovery with Large Language Models (ICLR 2025 · similarity 0.83)
- CODE REPRESENTATION LEARNING AT SCALE (ICLR 2024 · 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.
