Research map / Architectures and efficiency / Pruning and efficient tuning
Stochastic Normalization
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 |
|---|---|---|---|
| replace BN with StochNorm | ResNet-50 fine-tuning · CUB-200-2011 15% | top-1 accuracy: 45.25 -> 50.14 | Helped |
| replace BN with StochNorm | ResNet-50 fine-tuning · NIH Chest X-ray 5% | avg AUC: 70.37 -> 72.50 | Helped |
| replace BN with StochNorm | ResNet-50 fine-tuning · Stanford Cars 100% | top-1 accuracy: 87.20 -> 87.35 | No clear effect |
| replace BN with StochNorm | ResNet-50 fine-tuning · CUB-200-2011 50% | top-1 accuracy: 70.12 -> 72.01 (BSS 72.56) | Mixed |
| add StochNorm to L2-SP | L2-SP · CUB-200-2011 15% | top-1 accuracy: 45.08 -> 49.92 | Helped |
| add StochNorm to DELTA | DELTA · FGVC Aircraft 15% | top-1 accuracy: 42.16 -> 44.10 | Helped |
| add StochNorm to BSS | BSS · Stanford Cars 15% | top-1 accuracy: 40.57 -> 44.04 | Helped |
| add StochNorm to BSS | BSS · Stanford Cars 100% | top-1 accuracy: 87.63 -> 87.85 | No clear effect |
| two-branch stochastic selection with (0,1) init instead of BN (ablation) | StochNorm · 3 fine-grained datasets 15% | improvement over vanilla | Helped |
| initialize moving statistics from pre-trained model instead of (0,1) (ablation) | StochNorm · 3 fine-grained datasets 15% | improvement over vanilla | Helped |
| initialize moving statistics from pre-trained rather than target-data statistics (ablation) | StochNorm · 3 fine-grained datasets 15% | improvement over vanilla | Helped |
| moving-average coefficient alpha 0.1 instead of 0.001 (ablation) | StochNorm · 3 fine-grained datasets 15% | improvement over vanilla | Helped |
Closest papers by meaning
- Deconstructing the Regularization of BatchNorm (ICLR 2021 · similarity 0.88)
- Inherent Weight Normalization in Stochastic Neural Networks (NeurIPS 2019 · similarity 0.88)
- How Does Batch Normalization Help Optimization? (NeurIPS 2018 · similarity 0.88)
- Stochastic Whitening Batch Normalization (CVPR 2021 · similarity 0.88)
- Four Things Everyone Should Know to Improve Batch Normalization (ICLR 2020 · similarity 0.88)
- Beyond BatchNorm: Towards a Unified Understanding of Normalization in Deep Learning (NeurIPS 2021 · similarity 0.88)
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.