Research map / Graph learning / GNN architectures
Deep Expander Networks: Efficient Deep Networks from Graph Theory
Ameya Prabhu, Girish Varma, Anoop Namboodiri
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 1x1 convs with random-expander X-Conv instead of grouped conv (2x compression) (ablation) | MobileNet-0.5 · ImageNet | accuracy: 57.45 -> 58.22 | Improved |
| replace grouped conv with X-Conv at 4x compression (ablation) | MobileNet-0.5 · ImageNet | accuracy: 49.41 -> 54.00 | Improved |
| replace grouped conv with X-Conv at 8x compression (ablation) | MobileNet-0.5 · ImageNet | accuracy: 45.13 -> 49.23 | Improved |
| replace grouped conv with X-Conv at 16x compression (ablation) | MobileNet-0.5 · ImageNet | accuracy: 39.03 -> 44.63 | Improved |
| replace dense layers with expander X-Conv (2x) in ResNet-34 vs X-ResNet-2-50 | ResNet · ImageNet | accuracy: 71.66 -> 72.85 | Improved |
| apply X-Conv (2x compression) to ResNet-50 | ResNet-50 · ImageNet | accuracy: 74.46 -> 72.85 | Mixed |
| apply X-Conv (2x) to DenseNet-BC-161 | DenseNet-BC-161 · ImageNet | accuracy: 76.3 -> 74.3 | Hurt |
| compress VGG16 with expander layers (X-VGG16-1, 9x fewer params) | VGG16 · CIFAR-10 | accuracy: 94.0 -> 93.4 | Hurt |
| compress DenseNet-BC-40-24 with X-Conv (2x) | DenseNet-BC-40-24 · CIFAR-10 | accuracy: 94.79 -> 94.83 | No clear effect |
| compress DenseNet-BC-40-24 with X-Conv (2x) | DenseNet-BC-40-24 · CIFAR-100 | accuracy: 76.05 -> 74.37 | Hurt |
| make the X-DenseNet deeper (70 vs 58 layers) at similar compression (ablation) | X-DenseNet-BC · CIFAR-100 | accuracy: 78.74 -> 79.56 | Improved |
| widen DenseNet-BC-40-60 to growth rate 200 with 30x compression (ablation) | X-DenseNet-BC · CIFAR-100 | accuracy: 81.62 -> 77.29 | Hurt |
Closest papers by meaning
- ExpandNets: Linear Over-parameterization to Train Compact Convolutional Networks (NeurIPS 2020 · similarity 0.89)
- Towards Deepening Graph Neural Networks: A GNTK-based Optimization Perspective (ICLR 2022 · similarity 0.88)
- Network Expansion for Practical Training Acceleration (CVPR 2023 · similarity 0.88)
- Bag of Tricks for Training Deeper Graph Neural Networks: A Comprehensive Benchmark Study (TPAMI 2022 · similarity 0.87)
- Universal Graph Convolutional Networks (NeurIPS 2021 · similarity 0.87)
- Random Sparse Lifts: Construction, Analysis and Convergence of finite sparse networks (ICLR 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.
