Research map / Architectures and efficiency / RNNs and normalization
P-nets: Deep Polynomial Neural Networks
Check 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 the SNGAN generator with a single NCP polynomial | SNGAN · CIFAR10 generation | IS: 8.06± 0.10 -> 8.30± 0.09 | Helped |
| Use a product of NCP polynomials instead of a single NCP polynomial | SNGAN generator · CIFAR10 generation | IS: 8.30± 0.09 -> 8.49± 0.11 | Helped |
| Convert the mapping network to an NCP polynomial | StyleGAN · FFHQ 256×256 generation | FID: 7.15 -> 6.82 | Helped |
| Replace ResNet34 with second-order polynomial residual blocks and fewer blocks | ResNet34 · Speech Commands | accuracy: 0.951 ± 0.002 -> 0.951 ± 0.002 | No clear effect |
| Replace ResNet34 with second-order polynomial residual blocks and fewer blocks | ResNet34 · Speech Commands | parameters (millions): 21.3 -> 13.2 | Helped |
| Add second-order polynomial terms to residual blocks | ResNet50 · ImageNet | Top-1 error (%): 23.570 -> 22.875 | Helped |
| Add second-order polynomial terms to residual blocks | ResNet50 · ImageNet | Top-5 error (%): 6.838 -> 6.358 | Helped |
| Extend spiral convolution with the simple polynomial operator | Spiral convolution · 3D mesh autoencoding | reconstruction error (mm): 0.635 -> 0.530 | Helped |
| Remove activation functions from ProdPoly simple (ablation) | ProdPoly simple · 3D mesh autoencoding | reconstruction error (mm): 0.530 -> 0.529 | No clear effect |
| Use the full polynomial operator instead of the simple operator | ProdPoly simple · 3D mesh autoencoding | reconstruction error (mm): 0.530 -> 0.476 | Helped |
| Remove activation functions from ProdPoly full (ablation) | ProdPoly full · 3D mesh autoencoding | reconstruction error (mm): 0.476 -> 0.474 | No clear effect |
Closest papers by meaning
- Regularization of Polynomial Networks for Image Recognition (CVPR 2023 · similarity 0.90)
- The Spectral Bias of Polynomial Neural Networks (ICLR 2022 · similarity 0.87)
- Doubly Convolutional Neural Networks (NeurIPS 2016 · similarity 0.86)
- Multilinear Operator Networks (ICLR 2024 · similarity 0.85)
- Identifiability of Deep Polynomial Neural Networks (NeurIPS 2025 · similarity 0.85)
- On the Hardness of Training Deep Neural Networks Discretely (AAAI 2025 · 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.