Research map / Architectures and efficiency / CNN design
Gaussian-Based Pooling for Convolutional Neural Networks
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 |
|---|---|---|---|
| make half-Gaussian pooling stochastic with sigma0 learned by GFGP vs fixed sigma0=1 (ablation) | half-Gaussian pooling · CIFAR-100 val | error rate (%): 24.25 -> 23.48 | Helped |
| add stochasticity to iSP-Gaussian pooling (sigma0 > 0) (ablation) | iSP-Gaussian pooling · CIFAR-100 val | error rate (%): 24.12 -> 23.52 | Helped |
| restrict output to at or above the mean (half-Gaussian) instead of Gaussian (ablation) | Gaussian pooling · CIFAR-100 val | error rate (%): 24.51 -> 24.25 | Helped |
| share random number across spatial positions (partial stochasticity) (ablation) | iSP-Gaussian pooling · CIFAR-100 val | error rate (%): 23.52 -> 23.68 | Hurt |
| share random number across spatial positions (partial stochasticity) (ablation) | half-Gaussian pooling · CIFAR-100 val | error rate (%): 23.48 -> 23.60 | Hurt |
| combine with S3 pooling (ablation) | iSP-Gaussian pooling · CIFAR-100 val | error rate (%): 23.52 -> 22.98 | Helped |
| combine with S3 pooling (ablation) | half-Gaussian pooling · CIFAR-100 val | error rate (%): 23.48 -> 23.12 | Helped |
| use iSP-Gaussian as global pooling in place of GAP (ablation) | 13-layer Net · CIFAR-100 val | error rate (%): 24.78 -> 23.83 | Helped |
| use half-Gaussian as global pooling in place of GAP (ablation) | 13-layer Net · CIFAR-100 val | error rate (%): 24.78 -> 24.54 | No clear effect |
| iSP-Gaussian vs half-Gaussian pooling on larger CNNs | iSP-Gaussian pooling · ImageNet val | top-1 error (%): 27.96 -> 27.33 | Helped |
| iSP-Gaussian pooling vs GFGP | ResNet-50 · ImageNet val | top-1 error (%): 21.79 -> 21.37 | Helped |
| iSP-Gaussian pooling vs average pooling | ResNeXt-50 · ImageNet val | top-1 error (%): 22.14 -> 20.66 | Helped |
Closest papers by meaning
- S3Pool: Pooling With Stochastic Spatial Sampling (CVPR 2017 · similarity 0.90)
- Wavelet Pooling for Convolutional Neural Networks (ICLR 2018 · similarity 0.89)
- Detail-Preserving Pooling in Deep Networks (CVPR 2018 · similarity 0.88)
- TI-Pooling: Transformation-Invariant Pooling for Feature Learning in Convolutional Neural Networks (CVPR 2016 · similarity 0.87)
- Kernel Pooling for Convolutional Neural Networks (CVPR 2017 · similarity 0.87)
- Learning with convolution and pooling operations in kernel methods (NeurIPS 2022 · 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.