Research map / Architectures and efficiency / Attention and transformers
Universal Approximation with Softmax Attention
Jerry Yao-Chieh Hu, Hude Liu, Hong-Yu Chen, Weimin Wu, Han Liu
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 |
|---|---|---|---|
| increase the number of interpolation points p (ablation) | one-layer single-head softmax attention approximating truncated linear models · synthetic truncated linear model (d=10, n=50) | test MSE | Improved |
| increase the number of interpolation points p (ablation) | one-layer single-head softmax attention · synthetic truncated linear model (d=10, n=50) | training stability (std over 10 runs) | Improved |
| widen the truncation interval |b-a| (ablation) | one-layer softmax attention · synthetic truncated linear model (d=10, n=50) | test MSE | Hurt |
| increase the number of attention heads H (ablation) | multi-head softmax attention · synthetic truncated linear model (d=10, n=50) | test MSE | Improved |
| increase the number of interpolation points p (ablation) | two-layer attention-only network (H=2) · synthetic seq-to-seq task (d=5, N=50000) | test MSE | Improved |
| increase the number of attention heads H (ablation) | two-layer attention-only network · synthetic seq-to-seq task (d=5, n=20, N=50000) | test MSE | Improved |
Closest papers by meaning
- Are Transformers with One Layer Self-Attention Using Low-Rank Weight Matrices Universal Approximators? (ICLR 2024 · similarity 0.91)
- Attention Mechanism, Max-Affine Partition, and Universal Approximation (NeurIPS 2025 · similarity 0.91)
- Theory, Analysis, and Best Practices for Sigmoid Self-Attention (ICLR 2025 · similarity 0.90)
- Degrees of Freedom for Linear Attention: Distilling Softmax Attention with Optimal Feature Efficiency (NeurIPS 2025 · similarity 0.89)
- ZeroS: Zero‑Sum Linear Attention for Efficient Transformers (NeurIPS 2025 · similarity 0.88)
- kNN Attention Demystified: A Theoretical Exploration for Scalable Transformers (ICLR 2025 · similarity 0.88)
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.
