Research map / Architectures and efficiency / Neural architecture search
AutoAttend: Automated Attention Representation Search
Chaoyu Guan, Xin Wang, Wenwu Zhu
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 attention layer (attention representation search) to the search space (ablation) | NAS search space for SST · SST validation | accuracy [%]: 81.15 -> 81.68 | Improved |
| add attention layer to the search space (ablation) | NAS search space for Cora · Cora validation | accuracy [%]: 81.80 -> 82.96 | Improved |
| add attention layer to the search space (ablation) | NAS search space for CiteSeer · CiteSeer validation | accuracy [%]: 72.18 -> 72.90 | Improved |
| add attention layer to the search space (ablation) | NAS search space for PubMed · PubMed validation | accuracy [%]: 81.04 -> 81.04 | No clear effect |
| full context-aware parameter sharing (source and target context) vs no context (ablation) | supernet parameter sharing for SST · SST validation | accuracy [%]: 68.68 -> 69.40 | Improved |
| full context-aware parameter sharing vs no context (ablation) | supernet parameter sharing for Cora · Cora validation | accuracy [%]: 77.09 -> 78.61 | Improved |
| full context-aware parameter sharing vs no context (ablation) | supernet parameter sharing for PubMed · PubMed validation | accuracy [%]: 72.72 -> 77.72 | Improved |
| source-only context (SC) vs no context (ablation) | supernet parameter sharing · SST/Cora/CiteSeer/PubMed validation | accuracy [%]: SST 68.68->68.96, Cora 77.09->77.81, CiteSeer 63.68->63.62, PubMed 72.72->73.31 | Mixed |
| searched attention representation (AutoAttend) vs previous NAS on text classification | transferred sentence encoder · SST and transferred NLP datasets | accuracy [%]: SST +1.20 over TextNAS (53.71 vs 52.51); transfer gains under 1.00% | Improved |
| searched attention representation vs hand-crafted SOTA on graph node classification | GRL transductive and inductive tasks · Cora, CiteSeer, PubMed, PPI | accuracy / F1: Cora +0.3, PubMed +0.4, PPI +0.1 over best baseline; CiteSeer 73.0 vs AGNN 73.8 | Mixed |
| searched attention representation vs hand-crafted SOTA on text classification | transferred sentence encoder · YELP and AG | accuracy [%]: AG 93.53 vs 93.20 (GELE); YELP 66.82 vs 67.00 (GELE) | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TEXTNAS — TextNAS: A Neural Architecture Search Space Tailored for Text Representation (AAAI 2020) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- On the Dynamics of Training Attention Models (ICLR 2021 · similarity 0.85)
- TextNAS: A Neural Architecture Search Space Tailored for Text Representation (AAAI 2020 · similarity 0.85)
- SAS: Simple Attention Sparsification via End-to-End Optimization of Context Ranking (NeurIPS 2026 · similarity 0.85)
- Selective Attention: Enhancing Transformer through Principled Context Control (NeurIPS 2024 · similarity 0.85)
- Neural Attention Search (NeurIPS 2025 · similarity 0.84)
- Attention with Trained Embeddings Provably Selects Important Tokens (NeurIPS 2025 · similarity 0.84)
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.
