Research map / Architectures and efficiency / Neural architecture search
ElasticViT: Conflict-aware Supernet Training for Deploying Fast Vision Transformer on Diverse Mobile Devices
Chen Tang, Li Lyna Zhang, Huiqiang Jiang, Jiahang Xu, Ting Cao, Quanlu Zhang, Yuqing Yang, Zhi Wang, Mao Yang
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 |
|---|---|---|---|
| add adjacent step (complexity-aware) sampling (ablation) | ElasticViT supernet · ImageNet supernet, 100 MFLOPs | top-1 accuracy: 69.9 -> 73.2 | Improved |
| add adjacent step sampling at large FLOPs (ablation) | ElasticViT supernet · ImageNet supernet, 800 MFLOPs | top-1 accuracy: 79.0 -> 79.0 | No clear effect |
| add multiple hierarchical smallest subnets (HSS) (ablation) | ElasticViT supernet · ImageNet supernet, 300 MFLOPs | top-1 accuracy: 76.7 -> 78.4 | Improved |
| add multiple hierarchical smallest subnets (HSS) (ablation) | ElasticViT supernet · ImageNet supernet, 100 MFLOPs | top-1 accuracy: 73.2 -> 72.8 | Hurt |
| add performance-aware sampling (ablation) | ElasticViT supernet · ImageNet supernet, 100 MFLOPs | top-1 accuracy: 72.8 -> 73.8 | Improved |
| add all ElasticViT sampling techniques versus sandwich rule (ablation) | ElasticViT supernet · ImageNet supernet, 200 MFLOPs | top-1 accuracy: 74.2 -> 77.2 | Improved |
| apply conflict-aware sampling to reduce gradient conflicts (ablation) | ElasticViT supernet · Gradient cosine similarity, 200 MFLOPs random subnets | cosine similarity: 0.32 -> 0.51 | Improved |
| inherit supernet weights instead of retraining from scratch (ablation) | searched ViTs · ImageNet, 200 MFLOPs | top-1 accuracy: 75.3 -> 77.2 | Improved |
| inherit supernet weights instead of retraining from scratch (ablation) | searched ViTs · ImageNet, 700 MFLOPs | top-1 accuracy: 79.9 (scratch) vs 79.8 | No clear effect |
| use FLOPs-focused hybrid ViTs such as MobileFormer instead of mobile-friendly ops | mobile ViTs · Pixel4 latency | latency: MobileFormer 415.1 ms vs ElasticViT-S1 21.0 ms | Hurt |
Closest papers by meaning
- RepViT: Revisiting Mobile CNN From ViT Perspective (CVPR 2024 · similarity 0.88)
- Elastic ViTs from Pretrained Models without Retraining (NeurIPS 2025 · similarity 0.87)
- HydraViT: Stacking Heads for a Scalable ViT (NeurIPS 2024 · similarity 0.87)
- Subnet-Aware Dynamic Supernet Training for Neural Architecture Search (CVPR 2025 · similarity 0.86)
- Vision Transformer Neural Architecture Search for Out-of-Distribution Generalization: Benchmark and Insights (NeurIPS 2024 · similarity 0.86)
- Real-time Core-Periphery Guided ViT with Smart Data Layout Selection on Mobile Devices (NeurIPS 2024 · 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.
