Research map / Architectures and efficiency / CNN design
Winograd Algorithm for AdderNet
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 Winograd transforms to AdderNet | AdderNet · FPGA simulation, single layer | total energy consuming (equivalent): 50.4M -> 24.0M | Helped |
| apply Winograd transforms to AdderNet | ResNet-18 · ImageNet | top-1 accuracy: 67.0% -> 66.2% | Hurt |
| add modified transform matrix A (ablation) | Winograd AdderNet · CIFAR-10 | accuracy: 83.87% -> 89.25% | Helped |
| add ℓ2-to-ℓ1 training (ablation) | Winograd AdderNet · CIFAR-10 | accuracy: 83.87% -> 88.25% | Helped |
| combine modified transform matrix A with ℓ2-to-ℓ1 training (ablation) | Winograd AdderNet · CIFAR-10 | accuracy: 83.87% -> 91.56% | Helped |
| combine modified transform matrix A with ℓ2-to-ℓ1 training (ablation) | Winograd AdderNet · CIFAR-100 | accuracy: 54.72% -> 67.96% | Helped |
| reduce p during convergence rather than after convergence (ablation) | ℓ2-to-ℓ1 Winograd AdderNet training · CIFAR-10 | accuracy: 89.24 -> 91.56 | Helped |
| use p=140 rather than p=35 during convergence (ablation) | ℓ2-to-ℓ1 Winograd AdderNet training · CIFAR-10 | accuracy: 91.56 -> 91.44 | Hurt |
| apply kernel transformation during training (ablation) | Winograd AdderNet · CIFAR-10 | accuracy: 91.56 -> 89.19 | Hurt |
| initialize the original Adder kernel and transform it instead of directly initializing the Winograd kernel (ablation) | Winograd AdderNet · CIFAR-10 | accuracy: 91.56 -> 91.28 | Hurt |
Closest papers by meaning
- AdderNet: Do We Really Need Multiplications in Deep Learning? (CVPR 2020 · similarity 0.88)
- Adder Attention for Vision Transformer (NeurIPS 2021 · similarity 0.86)
- Towards Stable and Robust AdderNets (NeurIPS 2021 · similarity 0.86)
- Redistribution of Weights and Activations for AdderNet Quantization (NeurIPS 2022 · similarity 0.85)
- SFC: Achieve Accurate Fast Convolution under Low-precision Arithmetic (ICML 2024 · similarity 0.81)
- InceptionNeXt: When Inception Meets ConvNeXt (CVPR 2024 · similarity 0.80)
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.