Research map / Architectures and efficiency / Attention and transformers
Flowformer: Linearizing Transformers with Conservation Flows
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 source competition (ablation) | Flowformer · LRA | accuracy: 55.25 -> 56.48 | Helped |
| Add sink allocation (ablation) | Flowformer · LRA | accuracy: 55.58 -> 56.48 | Helped |
| Add source competition (ablation) | Flowformer · WikiText-103 | perplexity: 31.2 -> 30.8 | Helped |
| Add sink allocation (ablation) | Flowformer · WikiText-103 | perplexity: 32.2 -> 30.8 | Helped |
| Use Sigmoid instead of ELU+1 for the non-negative projection (ablation) | Flowformer · LRA | accuracy: 56.01 -> 56.48 | Helped |
| Use Sigmoid instead of ReLU for the non-negative projection (ablation) | Flowformer · LRA | accuracy: 56.07 -> 56.48 | Helped |
| Use Softmax for competition instead of Sigmoid (ablation) | Flowformer · LRA | accuracy: 52.36 -> 56.48 | Helped |
| Use Sigmoid for allocation instead of Softmax (ablation) | Flowformer · LRA | accuracy: 55.41 -> 56.48 | Helped |
| Replace DeiT-S attention with Flow-Attention | DeiT-S · ImageNet-1K | Top-1 accuracy: 79.8 -> 80.0 | Helped |
| Replace Decision Transformer attention with Flow-Attention | Decision Transformer · D4RL | average reward: 72.2±2.6 -> 73.5±2.9 | Helped |
Closest papers by meaning
- Token Statistics Transformer: Linear-Time Attention via Variational Rate Reduction (ICLR 2025 · similarity 0.88)
- Rethinking Attention: Exploring Shallow Feed-Forward Neural Networks as an Alternative to Attention Layers in Transformers (Student Abstract) (AAAI 2024 · similarity 0.88)
- Mixture-of-Top-$k$ Attention: Efficient Attention as Scalable Fast Weights (NeurIPS 2026 · similarity 0.87)
- Pool Me Wisely: On the Effect of Pooling in Transformer-Based Models (NeurIPS 2025 · similarity 0.87)
- On the Role of Attention Masks and LayerNorm in Transformers (NeurIPS 2024 · similarity 0.87)
- Learning Linear Attention in Polynomial Time (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.