Research map / Architectures and efficiency / CNN design
Recomputation of dense layers for the performance improvement of DCNN
Yimin Yang, Q. M. Jonathan Wu, Xiexing Feng, Thangarajah Akilan
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 |
|---|---|---|---|
| replace BP in FC layers with Moore-Penrose recomputation | ImageNet-pretrained AlexNet · CIFAR10 | top-1 accuracy: 90.0 -> 91.2 | Improved |
| replace BP in FC layers with Moore-Penrose recomputation | ImageNet-pretrained AlexNet · SUN397 | top-1 accuracy: 38.5 -> 40.5 | Improved |
| add recomputation | ImageNet-pretrained VGG-16 · CIFAR10 | top-1 accuracy: 95.2 -> 95.2 | No clear effect |
| add recomputation | ImageNet-pretrained VGG-16 · SUN397 | top-1 accuracy: 53.1 -> 55.6 | Improved |
| add recomputation | ImageNet-pretrained Google Inception · SUN397 | top-1 accuracy: 49.3 -> 48.9 | Hurt |
| add recomputation | ImageNet-pretrained Google Inception · CIFAR100 | top-1 accuracy: 77.1 -> 77.3 | Improved |
| add recomputation | 40-layer DenseNet from scratch · CIFAR10 | top-1 accuracy: 93.0 -> 93.7 | Improved |
| add recomputation | ImageNet-pretrained ResNet-50 · CIFAR100 | top-1 accuracy: 80.8 -> 81.9 | Improved |
| add recomputation | Places205-pretrained 16-layer VGG · Scene15 | accuracy: 92.0 -> 94.8 | Improved |
| add recomputation | 121-layer DenseNet from scratch · ImageNet | top-1 accuracy: 74.98 -> 75.91 | Improved |
| add recomputation | 121-layer DenseNet from scratch · ImageNet Mini | top-1 accuracy: 48.83 -> 51.45 | Improved |
| add recomputation | ImageNet-pretrained AlexNet · Places365 (1500/class) | top-1 accuracy: 40.13 -> 42.21 | Improved |
Closest papers by meaning
- Doubly Convolutional Neural Networks (NeurIPS 2016 · similarity 0.85)
- Convolutional Networks with Dense Connectivity (TPAMI 2019 · similarity 0.84)
- Densely Connected Convolutional Networks (CVPR 2017 · similarity 0.84)
- The Unreasonable Effectiveness of Fully-Connected Layers for Low-Data Regimes (NeurIPS 2022 · similarity 0.84)
- Deep Pyramidal Residual Networks (CVPR 2017 · similarity 0.84)
- Cross-Convolutional-Layer Pooling for Image Recognition (TPAMI 2016 · similarity 0.83)
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.
