Research map / Architectures and efficiency / CNN design
Stochastic Downsampling for Cost-Adjustable Inference and Improved Regularization in Convolutional Networks
Jason Kuen, Xiangfei Kong, Zhe Lin, Gang Wang, Jianxiong Yin, Simon See, Yap-Peng Tan
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 |
|---|---|---|---|
| Train WRN-d40-w4 with SDPoint | WRN-d40-w4 · CIFAR-10 | validation error (%): 4.29 -> 3.73 | Improved |
| Train WRN-d28-w10 with SDPoint | WRN-d28-w10 · CIFAR-10 | validation error (%): 3.84 -> 3.35 | Improved |
| Train DenseNetBC-d40-g60 with SDPoint | DenseNetBC-d40-g60 · CIFAR-10 | validation error (%): 3.99 -> 3.39 | Improved |
| Train WRN-d40-w4 with SDPoint | WRN-d40-w4 · CIFAR-100 | validation error (%): 20.78 -> 19.55 | Improved |
| Train WRN-d28-w10 with SDPoint | WRN-d28-w10 · CIFAR-100 | validation error (%): 18.51 -> 17.53 | Improved |
| Train DenseNetBC-d40-g60 with SDPoint | DenseNetBC-d40-g60 · CIFAR-100 | validation error (%): 20.00 -> 19.25 | Improved |
| Train ResNeXt-d101-c32 with SDPoint | ResNeXt-d101-c32 · ImageNet-1k val | Top-1 error (%): 21.2 -> 20.4 | Improved |
| Train ResNeXt-d101-c32 with SDPoint | ResNeXt-d101-c32 · ImageNet-1k val | Top-5 error (%): 5.6 -> 5.3 | Improved |
| Train PreResNet-d101 with SDPoint | PreResNet-d101 · ImageNet-1k val | Top-1 error (%): 22.0 -> 21.4 | Improved |
| Train PreResNet-d101 with SDPoint | PreResNet-d101 · ImageNet-1k val | Top-5 error (%): 6.1 -> 5.6 | Improved |
| Use instance-specific rather than uniform BatchNorm statistics (ablation) | SDPoint · CIFAR cost-error evaluation | classification error | Improved |
| Add the 0.5 downsampling ratio alongside 0.75 (ablation) | ResNeXt-d101-c32 with SDPoint · ImageNet cost-error evaluation | classification error | Improved |
Closest papers by meaning
- S3Pool: Pooling With Stochastic Spatial Sampling (CVPR 2017 · similarity 0.85)
- Spatially Adaptive Inference with Stochastic Feature Sampling and Interpolation (ECCV 2020 · similarity 0.85)
- Learning to Downsample for Segmentation of Ultra-High Resolution Images (ICLR 2022 · similarity 0.84)
- Learning to Zoom: a Saliency-Based Sampling Layer for Neural Networks (ECCV 2018 · similarity 0.84)
- Repeated Random Sampling for Minimizing the Time-to-Accuracy of Learning (ICLR 2024 · similarity 0.84)
- Pruning Convolutional Neural Networks for Resource Efficient Inference (ICLR 2017 · similarity 0.84)
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.
