Research map / Architectures and efficiency / Neural architecture search
AutoShrink: A Topology-Aware NAS for Discovering Efficient Neural Architecture
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 |
|---|---|---|---|
| Use topology-aware edge shrinking instead of a complete-DAG cell topology (ablation) | ShrinkCNN · CIFAR-10 test | Accuracy: 91.56 -> 93.22 | Helped |
| Use topology-aware edge shrinking instead of an ER graph prior (ablation) | ShrinkCNN · CIFAR-10 test | Accuracy: 92.1±0.38 -> 93.22 | Helped |
| Use topology-aware edge shrinking instead of a WS graph prior (ablation) | ShrinkCNN · CIFAR-10 test | Accuracy: 90.5±0.41 -> 93.22 | Helped |
| Use topology-aware edge shrinking instead of a BA graph prior (ablation) | ShrinkCNN · CIFAR-10 test | Accuracy: 90.0±0.34 -> 93.22 | Helped |
| Shrink the RNN cell topology from 15 edges to 8 (ablation) | ShrinkRNN · Penn Treebank test | Perplexity: 63.1 -> 56.5 | Helped |
| Increase CNN search candidates K from 10 to 15 (ablation) | AutoShrink · CNN architecture search, N=8 | Validation accuracy | No clear effect |
| Increase CNN search candidates K from 10 to 15 (ablation) | AutoShrink · CNN architecture search, N=8 | Search time: 1.5 hours -> 1.875 hours | Hurt |
| Use swish activation rather than ReLU (ablation) | ShrinkCNN · ImageNet-1K val | Top-1 Error: 26.1% -> 24.9% | Helped |
| Use topology-aware shrink search rather than a complete-DAG cell (ablation) | ShrinkCNN cell · CNN cell | MACs: 6.12M -> 2.60M | Helped |
| Use ShrinkRNN rather than DARTS | RNN architecture search · Penn Treebank | Search cost: 24 -> 1.5 | Helped |
Closest papers by meaning
- Auto-DeepLab: Hierarchical Neural Architecture Search for Semantic Image Segmentation (CVPR 2019 · similarity 0.88)
- Understanding Architectures Learnt by Cell-based Neural Architecture Search (ICLR 2020 · similarity 0.88)
- Towards Neural Architecture Search through Hierarchical Generative Modeling (ICML 2024 · similarity 0.87)
- einspace: Searching for Neural Architectures from Fundamental Operations (NeurIPS 2024 · similarity 0.86)
- HEP-NAS: Towards Efficient Few-shot Neural Architecture Search via Hierarchical Edge Partitioning (AAAI 2025 · similarity 0.86)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · 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.