Research map / Architectures and efficiency / RNNs and normalization
Deep Polynomial Neural Networks
Grigorios G. Chrysos, Stylianos Moschoglou, Giorgos Bouritsas, Jiankang Deng, Yannis Panagakis, Stefanos P. Zafeiriou
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| convert SNGAN generator to single NCP polynomial | SNGAN · CIFAR10 generation | FID: 19.06 -> 17.65 | Improved |
| use product of polynomials instead of single NCP (ablation) | SNGAN generator · CIFAR10 generation | FID: 17.65 -> 16.79 | Improved |
| convert StyleGAN mapping network to NCP polynomial | StyleGAN · FFHQ 256x256 | FID: 7.15 -> 6.82 | Improved |
| use second-order residual blocks to cut blocks of ResNet34 | ResNet34 · Speech Commands | accuracy: 0.951 -> 0.951, params 21.3 -> 13.2 | No clear effect |
| second-order residual blocks with fewer blocks | ResNet18 · CIFAR10 | accuracy: 0.945 -> 0.945, params 11.2 -> 6.0 | No clear effect |
| second-order residual blocks with fewer blocks | ResNet34 · CIFAR100 | accuracy: 0.769 -> 0.769, params 21.3 -> 14.7 | No clear effect |
| second-order residual blocks (NCP-Skip) in ResNet50 | ResNet50 · ImageNet | Top-1 error: 23.546 -> 22.827 | Improved |
| second-order residual blocks in ResNet50 | ResNet50 · ImageNet training | throughput: 1625 -> 1531 | Hurt |
| second-order residual blocks in ArcFace ResNet50 | ResNet50 · IJB-B 1:1 verification FAR=1e-6 | TAR: 37.28 -> 43.46 | Improved |
| second-order residual blocks in ArcFace ResNet50 | ResNet50 · CPLFW | verification accuracy: 92.433 -> 93.317 | Improved |
| second-order residual blocks in ArcFace ResNet50 | ResNet50 · MegaFace | Id / Ver: 98.28 -> 98.78 (Id); 98.64 -> 98.95 (Ver) | Improved |
| increase polynomial order of Spiral Pi-Net, simple model (ablation) | Spiral mesh autoencoder · COMA and DFAUST mesh reconstruction | reconstruction error | Mixed |
Closest papers by meaning
- P-nets: Deep Polynomial Neural Networks (CVPR 2020 · similarity 0.96)
- Regularization of Polynomial Networks for Image Recognition (CVPR 2023 · similarity 0.90)
- The Spectral Bias of Polynomial Neural Networks (ICLR 2022 · similarity 0.88)
- Multilinear Operator Networks (ICLR 2024 · similarity 0.87)
- Identifiability of Deep Polynomial Neural Networks (NeurIPS 2025 · similarity 0.87)
- Convolutions and More as Einsum: A Tensor Network Perspective with Advances for Second-Order Methods (NeurIPS 2024 · similarity 0.86)
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.
