Research map / Architectures and efficiency / RNNs and normalization
Regularization of Polynomial Networks for Image Recognition
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 |
|---|---|---|---|
| Apply the proposed regularization scheme to Π-Nets | Polynomial networks · STL-10 | accuracy: 56.3% to 82.8% | Helped |
| Add zero-mean initialization (ablation) | R-PolyNets · CIFAR-100 | accuracy: 0.765 ± 0.002 -> 0.769 ± 0.002 | Helped |
| Add strong regularization, including normalization and label smoothing (ablation) | D-PolyNets · CIFAR-100 | accuracy: 0.726 ± 0.006 -> 0.767 ± 0.003 | Helped |
| Replace label smoothing with CutMix (ablation) | R-PolyNets · CIFAR-100 | accuracy: 0.766 ± 0.002 -> 0.771 ± 0.002 | Helped |
| Replace DropBlock with stochastic depth (ablation) | R-PolyNets · CIFAR-100 | accuracy: 0.766 ± 0.002 -> 0.769 ± 0.002 | Helped |
| Increase polynomial expansion degree (ablation) | R-PolyNets · CIFAR-10 | accuracy: 0.880 ± 0.003 -> 0.950 ± 0.002 | Helped |
| Increase polynomial expansion degree (ablation) | R-PolyNets · CIFAR-100 | accuracy: 0.671 ± 0.003 -> 0.775 ± 0.002 | Helped |
| Use dense connections across sequential polynomials | D-PolyNets versus R-PolyNets · STL-10 | accuracy: 0.828 ± 0.003 -> 0.834 ± 0.006 | Helped |
| Use dense connections across sequential polynomials | D-PolyNets versus R-PolyNets · CIFAR-100 | accuracy: 0.769 ± 0.002 -> 0.767 ± 0.003 | No clear effect |
| Use dense connections across sequential polynomials | D-PolyNets versus R-PolyNets · CIFAR-10 | number of parameters: 11.9M -> 7.1M | Helped |
| Increase weight decay from 5e^{-4} to 7e^{-4} (ablation) | R-PolyNets · CIFAR-100 | accuracy: 0.766 ± 0.002 -> 0.768 ± 0.002 | No clear effect |
| Apply the regularized polynomial method instead of Π-Nets | R-PolyNets · CIFAR-10 with 50 training samples per class | accuracy: 0.314 ± 0.005 -> 0.484 ± 0.004 | Helped |
Closest papers by meaning
- P-nets: Deep Polynomial Neural Networks (CVPR 2020 · similarity 0.90)
- Identifiability of Deep Polynomial Neural Networks (NeurIPS 2025 · similarity 0.88)
- Sound and Complete Verification of Polynomial Networks (NeurIPS 2022 · similarity 0.87)
- PolyNet: A Pursuit of Structural Diversity in Very Deep Networks (CVPR 2017 · similarity 0.87)
- Multilinear Operator Networks (ICLR 2024 · similarity 0.85)
- Characterizing ResNet's Universal Approximation Capability (ICML 2024 · similarity 0.85)
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.