Research map / Architectures and efficiency / Attention and transformers
Conditioned Initialization for Attention
Hemanth Saratchandran, Simon Lucey
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 |
|---|---|---|---|
| use conditioned instead of default init | ViT-B · ImageNet-1k | top-1 accuracy: 80.3 -> 81.5 | Improved |
| use conditioned instead of default init | Swin-B · ImageNet-1k | top-1 accuracy: 83.4 -> 84.6 | Improved |
| use conditioned instead of default init | DaViT-B · ImageNet-1k | top-1 accuracy: 84.3 -> 85.3 | Improved |
| use mimetic instead of default init | DeiT-B · ImageNet-1k | top-1 accuracy: 81.6 -> 81.6 | No clear effect |
| use conditioned init, measured as epochs to match default accuracy | ViT-B, DeiT-B, Swin-B, XCiT-M · ImageNet-1k | epochs to target: 20-30% faster | Improved |
| use conditioned instead of default init | ViT-T · Pets | top-1 accuracy: 26.7 -> 47.7 | Improved |
| use conditioned instead of mimetic init | ViT-T · Pets | top-1 accuracy: 47.7 -> 47.7 | No clear effect |
| use conditioned instead of mimetic init | ViT-T · CIFAR-100 | top-1 accuracy: 75.0 -> 75.3 | Improved |
| use conditioned instead of default init | XCiT-S Mask R-CNN · COCO 2017 | AP^b: 44.9 -> 45.5 | Improved |
| use mimetic instead of default init | XCiT-S Mask R-CNN · COCO 2017 | AP^b: 44.9 -> 44.8 | No clear effect |
| use conditioned instead of default init | Nystromformer · LRA ListOps | accuracy: 37.1 -> 37.9 | Improved |
| use conditioned instead of default init | Crammed BERT · GLUE average | score: 78.6 -> 79.6 | Improved |
Closest papers by meaning
- From Condensation to Rank Collapse: A Two-Stage Analysis of Transformer Training Dynamics (NeurIPS 2025 · similarity 0.89)
- In-context Convergence of Transformers (ICML 2024 · similarity 0.88)
- Spectral Conditioning of Attention Improves Transformer Performance (NeurIPS 2025 · similarity 0.88)
- Unveiling Induction Heads: Provable Training Dynamics and Feature Learning in Transformers (NeurIPS 2024 · similarity 0.88)
- Fine-grained Analysis of In-context Linear Estimation: Data, Architecture, and Beyond (NeurIPS 2024 · similarity 0.88)
- Transformers Learn Faster with Semantic Focus (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.
