Research map / Architectures and efficiency / Attention and transformers
Krause Synchronization Transformers
Jingkun Liu, Yisong Yue, Max Welling, Yue Song
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 |
|---|---|---|---|
| Replace self-attention with Krause Attention in ViT | ViT-B · CIFAR-100 | accuracy: 72.28 -> 78.03 | Improved |
| Replace self-attention with Krause Attention in ViT-T | ViT-T · CIFAR-100 | accuracy: 66.07 -> 74.34 | Improved |
| Replace self-attention with Krause Attention in ViT-B | ViT-B · CIFAR-10 | accuracy: 92.45 -> 95.35 | Improved |
| Replace self-attention with Krause Attention in ViT-B-16 | ViT-B-16 · ImageNet-1K | accuracy: 75.85 -> 76.75 | Improved |
| Replace self-attention with Krause Attention in ViT-B | ViT-B · CIFAR-10 | FLOPs: 5.61G -> 3.77G | Improved |
| Add RoPE to Krause ViT-B | KViT-B · CIFAR-100 | accuracy: 78.03 -> 79.65 | Improved |
| Add RoPE to Krause ViT-B-16 | KViT-B-16 · ImageNet-1K | accuracy: 76.75 -> 78.61 | Improved |
| Replace dot-product similarity with RBF Krause kernel in Swin-S | Swin-S · CIFAR-10 | accuracy: 90.21 -> 91.13 | Improved |
| Replace dot-product similarity with RBF Krause kernel in Swin-T | Swin-T · CIFAR-10 | accuracy: 90.73 -> 91.09 | Improved |
| Replace full attention with Krause Attention in autoregressive model | ARM · MNIST | BPD: 0.5685 -> 0.5652 | Improved |
| Replace full attention with Krause Attention in autoregressive model | ARM · CIFAR-10 (autoregressive generation) | BPD: 3.0224 -> 3.0032 | Improved |
| Replace full attention with Krause Attention in autoregressive model | ARM · CIFAR-10 (autoregressive generation) | Images/sec: 1.8933 -> 4.5240 | Improved |
Closest papers by meaning
- Mixture-of-Top-$k$ Attention: Efficient Attention as Scalable Fast Weights (NeurIPS 2026 · similarity 0.87)
- kNN Attention Demystified: A Theoretical Exploration for Scalable Transformers (ICLR 2025 · similarity 0.87)
- Clustering in Causal Attention Masking (NeurIPS 2024 · similarity 0.87)
- On the Role of Attention Masks and LayerNorm in Transformers (NeurIPS 2024 · similarity 0.87)
- JoMA: Demystifying Multilayer Transformers via Joint Dynamics of MLP and Attention (ICLR 2024 · similarity 0.86)
- Training-Induced Escape from Token Clustering in a Mean-Field Formulation of Transformers (NeurIPS 2026 · similarity 0.86)
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.
