Research map / Architectures and efficiency / Attention and transformers
Fast Monte-Carlo Approximation of the Attention Mechanism
Hyunjun Kim, JeongGil Ko
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 |
|---|---|---|---|
| Use attention-weighted Monte-Carlo sampling with α=0.2 | BERT · SST-2 | accuracy: 92.43 -> 92.26 | No clear effect |
| Use attention-weighted Monte-Carlo sampling with α=0.2 | BERT · GLUE (CoLA) | FLOPS reduction: 11.44$\times$ | Improved |
| Increase the error coefficient α from 0.2 to 1.0 (ablation) | MCA-BERT · SST-2 | FLOPS reduction: 5.58$\times$ -> 12.34$\times$ | Improved |
| Increase the error coefficient α from 0.2 to 1.0 (ablation) | MCA-BERT · SST-2 | accuracy: 92.26$\pm$0.0 -> 80.66$\pm$0.1 | Hurt |
| Increase the error coefficient α from 0.2 to 1.0 (ablation) | MCA-BERT · CoLA | Matthews correlation: 53.74$\pm$0.1 -> 50.95$\pm$0.7 | Hurt |
| Apply Monte-Carlo Attention with α=0.2 | DistilBERT · SST-2 | accuracy: 91.51 -> 91.50 | No clear effect |
| Apply Monte-Carlo Attention with α=0.2 | Longformer · AAPD | accuracy: 74.81 -> 74.70 | No clear effect |
| Apply Monte-Carlo Attention with α=0.2 | Longformer · HND | FLOPS reduction: 5.32$\times$ | Improved |
Closest papers by meaning
- Understanding the Expressive Power and Mechanisms of Transformer for Sequence Modeling (NeurIPS 2024 · similarity 0.88)
- Spectral Conditioning of Attention Improves Transformer Performance (NeurIPS 2025 · similarity 0.88)
- Token Statistics Transformer: Linear-Time Attention via Variational Rate Reduction (ICLR 2025 · similarity 0.87)
- kNN Attention Demystified: A Theoretical Exploration for Scalable Transformers (ICLR 2025 · similarity 0.87)
- Are Transformers with One Layer Self-Attention Using Low-Rank Weight Matrices Universal Approximators? (ICLR 2024 · similarity 0.87)
- Fast attention mechanisms: a tale of parallelism (NeurIPS 2025 · similarity 0.87)
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.
