Research map / Architectures and efficiency / CNN design
Network of Experts for Large-Scale Image Categorization
Karim Ahmed, Mohammad Haris Baig, Lorenzo Torresani
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 |
|---|---|---|---|
| use K=10 experts instead of flat base model | AlexNet-C100 · CIFAR100 | top-1 accuracy: 54.0 -> 56.2 | Improved |
| increase number of experts from K=10 to K=50 (ablation) | NofE fully-balanced · CIFAR100 | top-1 accuracy: 56.2 -> 55.33 | Hurt |
| use only K=2 experts (ablation) | NofE fully-balanced · CIFAR100 | top-1 accuracy: 53.3 vs base 54.0 | No clear effect |
| use elasso soft balancing instead of fully-balanced (ablation) | NofE K=10 · CIFAR100 | top-1 accuracy: 56.2 -> 55.59 | Hurt |
| learn specialties jointly with generalist instead of spectral clustering (ablation) | NofE K=10 · CIFAR100 | top-1 accuracy: 53.2 -> 56.2 | Improved |
| learn specialties jointly instead of random balanced partition (ablation) | NofE K=10 · CIFAR100 | top-1 accuracy: 53.7 -> 56.2 | Improved |
| initialise trunk from generalist instead of random weights (ablation) | NofE AlexNet-C100 · CIFAR100 | top-1 accuracy: 49.53 -> 56.2 | Improved |
| initialise trunk from generalist instead of random weights (ablation) | NofE ResNet56-C100 · CIFAR100 | top-1 accuracy: 73.95 -> 76.24 | Improved |
| train generalist from scratch instead of finetuning from pretrained base model (ablation) | NofE AlexNet-C100 · CIFAR100 | top-1 accuracy: 55.5 -> 56.2 | Improved |
| convert base CNN into NofE (AlexNet-Quick) | AlexNet-Quick-C100 · CIFAR100 | top-1 accuracy: 37.94 -> 45.58 | Improved |
| convert base CNN into NofE (VGG11) | VGG11-C100 · CIFAR100 | top-1 accuracy: 68.48 -> 69.27 | Improved |
| match base model parameters and depth of NofE instead of using NofE (ResNet56) (ablation) | ResNet56-C100 · CIFAR100 | top-1 accuracy: 73.88 -> 76.24 | Improved |
Closest papers by meaning
- Multi-Scale Dense Networks for Resource Efficient Image Classification (ICLR 2018 · similarity 0.89)
- MuCaLe-Net: Multi Categorical-Level Networks to Generate More Discriminating Features (CVPR 2017 · similarity 0.88)
- Contextual Convolutional Networks (ICLR 2023 · similarity 0.88)
- Attention Convolutional Binary Neural Tree for Fine-Grained Visual Categorization (CVPR 2020 · similarity 0.88)
- Deeply-Supervised Knowledge Synergy (CVPR 2019 · similarity 0.87)
- SELECT: A Large-Scale Benchmark of Data Curation Strategies for Image Classification (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.
