Research map / Architectures and efficiency / Neural architecture search
GroSS: Group-Size Series Decomposition for Grouped Architecture Search
Henry Howard-Jenkins, Yiwen Li, Victor Adrian Prisacariu
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 GroSS searched configuration instead of uniform 32s baseline at equal MACs | GroSS search · CIFAR-10 4-layer | true accuracy: 83.70 -> 84.05 | Improved |
| use GroSS searched configuration instead of uniform 16s baseline, 4.25% fewer MACs | GroSS search · CIFAR-10 4-layer | true accuracy: 82.94 -> 83.88 | Improved |
| use GroSS searched configuration instead of VBMF rank selection | GroSS search · CIFAR-10 4-layer | true accuracy: 82.83 -> 83.88 | Improved |
| GroSS search below uniform 32s baseline on VGG-16 | GroSS search · CIFAR-10 VGG-16 | true accuracy: 91.57 -> 91.41 | Hurt |
| GroSS search below uniform 16s baseline on VGG-16 | GroSS search · CIFAR-10 VGG-16 | true accuracy: 91.19 -> 91.31 | Improved |
| GroSS search below uniform 4s baseline on VGG-16 | GroSS search · ImageNet VGG-16 | true accuracy: 70.51 -> 70.63 | Improved |
| GroSS search below uniform 4s baseline on ResNet-18 | GroSS search · ImageNet ResNet-18 | true accuracy: 65.46 -> 65.44 | No clear effect |
| GroSS search on high-compression structure of prior work | GroSS search · ImageNet VGG-16 | true accuracy: 66.85 -> 67.22 | Improved |
| use GroSS instead of partial fine-tuning for estimating configuration accuracy | GroSS · 4-layer CIFAR-10 | mean top-5 average precision: 52.0 -> 71.2 | Improved |
| use GroSS instead of no fine-tuning for estimating configuration accuracy | GroSS · 4-layer CIFAR-10 | mean top-5 average precision: 60.7 -> 71.2 | Improved |
| evaluate 45 configurations with GroSS dynamic group-size instead of separate networks | GroSS · 4-layer CIFAR-10 | inference time: 287s -> 7s | Improved |
Closest papers by meaning
- Differentiable Learning-to-Group Channels via Groupable Convolutional Neural Networks (ICCV 2019 · similarity 0.87)
- Fully Learnable Group Convolution for Acceleration of Deep Neural Networks (CVPR 2019 · similarity 0.87)
- On the VC dimension of deep group convolutional neural networks (NeurIPS 2025 · similarity 0.87)
- Group Downsampling with Equivariant Anti-aliasing (ICLR 2025 · similarity 0.85)
- Unitary Convolutions for Learning on Graphs and Groups (NeurIPS 2024 · similarity 0.85)
- Flexible Group Count Enables Hassle-Free Structured Pruning (CVPR 2025 · similarity 0.85)
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.
