Research map / Architectures and efficiency / Neural architecture search
Binarized Neural Architecture Search for Efficient Object Recognition
Hanlin Chen, Li’an Zhuo, Baochang Zhang, Xiawu Zheng, Jianzhuang Liu, Rongrong Ji, David Doermann, Guodong Guo
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 performance-based search instead of PC-DARTS for binarized (PCNN) architecture | BNAS · CIFAR-10 W1A32 | test accuracy (%): 95.12 -> 96.06 | Improved |
| use performance-based search instead of PC-DARTS for 1-bit architecture | BNAS · CIFAR-10 W1A1 | test accuracy (%): 90.06 -> 93.28 | Improved |
| add UCB exploration term to 1-bit BNAS (ablation) | BNAS (1-bit) · CIFAR-10 W1A1 | test error (%): 8.3 -> 6.72 | Improved |
| use performance-based operation pruning instead of PC-DARTS search | BNAS · CIFAR-10 W1A32 | search cost (GPU days): 0.18 -> 0.09375 | Improved |
| use performance-based operation pruning for full-precision search | BNAS (full-precision) · CIFAR-10 | test error (%): 2.78 -> 2.84 | No clear effect |
| increase initial channels for XNOR-based BNAS (44 channels) | BNAS (XNOR) · CIFAR-10 W1A32 | test error (%): 5.71 -> 4.88 | Improved |
| increase initial channels for PCNN-based BNAS (48 channels) | BNAS · CIFAR-10 W1A32 | test error (%): 3.94 -> 3.47 | Improved |
| increase warm-up epochs L from 3 to 5 (ablation) | BNAS (PCNN) · CIFAR-10 | accuracy (%): 95.8 -> 96.06 | Improved |
| increase warm-up epochs L from 5 to 11 (ablation) | BNAS (PCNN) · CIFAR-10 | accuracy (%): 96.06 -> 96.03; search cost 0.09375 -> 0.1687 | No clear effect |
| search binarized architecture with BNAS instead of using hand-designed ResNet-18 (PCNN) | BNAS · ImageNet W1A32 | top-1 accuracy (%): 63.5 -> 71.3 | Improved |
| search 1-bit architecture with BNAS instead of ResNet-18 (BONN) | BNAS (1-bit) · ImageNet W1A1 | top-1 accuracy (%): 59.3 -> 64.3 | Improved |
| binarize weights of searched network versus full-precision PC-DARTS | BNAS · ImageNet | top-1 accuracy (%): 75.8 -> 71.3 | Hurt |
Closest papers by meaning
- Binarized Neural Architecture Search (AAAI 2020 · similarity 0.96)
- BATS: Binary ArchitecTure Search (ECCV 2020 · similarity 0.88)
- ISTA-NAS: Efficient and Consistent Neural Architecture Search by Sparse Coding (NeurIPS 2020 · similarity 0.88)
- MIGO-NAS: Towards Fast and Generalizable Neural Architecture Search (TPAMI 2021 · similarity 0.87)
- DDPNAS: Efficient Neural Architecture Search via Dynamic Distribution Pruning (IJCV 2023 · similarity 0.87)
- 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.
