Research map / Architectures and efficiency / Neural architecture search
Search What You Want: Barrier Panelty NAS for Mixed Precision Quantization
Haibao Yu, Qi Han, Jianbo Li, Jianping Shi, Guangliang Cheng, Bin Fan
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 BP-NAS instead of HAWQ at a 4-bit budget | HAWQ · CIFAR-10 (ResNet20) | Top-1 accuracy: 92.22 -> 92.30 | Improved |
| use BP-NAS instead of HAWQ at a 4-bit budget | HAWQ · CIFAR-10 (ResNet20) | BOPs compression ratio: 64.49 -> 81.53 | Improved |
| use BP-NAS (3-bit budget) instead of fixed 3-bit PACT | PACT · CIFAR-10 (ResNet20) | Top-1 accuracy: 91.10 -> 92.04 | Improved |
| use BP-NAS (3-bit budget) instead of fixed 3-bit LQ-Nets | LQ-Nets · CIFAR-10 (ResNet20) | Top-1 accuracy: 91.60 -> 92.04 | Improved |
| use mixed precision from BP-NAS instead of fixed (4,4)-bit | Baseline (4,4)-bit · ImageNet (ResNet50) | Top-1 accuracy: 76.02 -> 76.67 | Improved |
| use BP-NAS (3-bit budget) instead of fixed 3-bit PACT | PACT · ImageNet (ResNet50) | Top-1 accuracy: 75.30 -> 75.71 | Improved |
| use BP-NAS instead of HAWQ | HAWQ · ImageNet (ResNet50) | Top-1 accuracy: 75.30 -> 75.71 | Improved |
| use mixed 4-bit BP-NAS instead of uniform 4-bit | Baseline 4-bit · COCO (Faster R-CNN ResNet50) | AP: 0.343 -> 0.358 | Improved |
| use own 4-bit baseline instead of FQN | FQN · COCO (Faster R-CNN ResNet50) | AP: 0.331 -> 0.343 | Improved |
| add barrier penalty regularizer to the search (ablation) | BP-NAS · CIFAR-10 (ResNet20) | Top-1 accuracy: 92.6 -> 92.04 | Mixed |
| add distribution reshaping (uniformization) in training (ablation) | mixed-precision ResNet20 · CIFAR-10 (ResNet20) | Top-1 accuracy: 60 -> 89 | Improved |
| add Prob-1 regularizer to the supernet (ablation) | BP-NAS · CIFAR-10 (ResNet20) | importance factor selectivity | Improved |
Closest papers by meaning
- BSQ: Exploring Bit-Level Sparsity for Mixed-Precision Neural Network Quantization (ICLR 2021 · similarity 0.87)
- AMPA: Adaptive Mixed Precision Allocation for Low-Bit Integer Training (ICML 2024 · similarity 0.84)
- Learning from Loss Landscape: Generalizable Mixed-Precision Quantization via Adaptive Sharpness-Aware Gradient Aligning (ICML 2025 · similarity 0.84)
- MetaMix: Meta-State Precision Searcher for Mixed-Precision Activation Quantization (AAAI 2024 · similarity 0.84)
- Efficient Multi-bit Quantization Network Training via Weight Bias Correction and Bit-wise Coreset Sampling (NeurIPS 2025 · similarity 0.83)
- Retraining-Free Model Quantization via One-Shot Weight-Coupling Learning (CVPR 2024 · similarity 0.83)
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.
