Research map / Architectures and efficiency / Neural architecture search
FBNetV3: Joint Architecture-Recipe Search Using Predictor Pretraining
Check 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 |
|---|---|---|---|
| Search the training recipe jointly with the architecture | FBNetV3 / NARS · ImageNet | top-1 accuracy: 79.1 -> 79.9 | Helped |
| Apply the NARS-searched recipe | ResNet50 · ImageNet | top-1 accuracy: 76.1 -> 79.2 | Helped |
| Apply the NARS-searched recipe | ResNeXt101-32x8d · ImageNet | top-1 accuracy: 79.3 -> 82.6 | Helped |
| Use architecture-specific recipes rather than swapping the B and E recipes (ablation) | FBNetV3-B and FBNetV3-E · ImageNet | top-1 accuracy: 79.8% -> 78.5%; 81.3% -> 80.8% | Mixed |
| Add predictor-based evolutionary search after constrained iterative optimization (ablation) | FBNetV3 · ImageNet | top-1 accuracy: 79.0% -> 79.8%; 80.7% -> 81.3% | Helped |
| Add EMA (ablation) | FBNetV3-G · ImageNet | top-1 accuracy: 80.9% -> 82.3% | Helped |
| Add distillation (ablation) | FBNetV3-G · ImageNet | top-1 accuracy: 80.9% -> 82.2% | Helped |
| Combine EMA and distillation (ablation) | FBNetV3-G · ImageNet | top-1 accuracy: 80.9% -> 82.8% | Helped |
| Use a higher-accuracy RegNetY-32G teacher for distillation | FBNetV3 models · ImageNet | top-1 accuracy: 0.2% - 0.5% | Helped |
| Pretrain the predictor embedding on architecture statistics | NARS accuracy predictor · ImageNet predictor evaluation | samples needed to reach the same prediction MSE: 5× less samples | Helped |
| Use searched FBNetV3-A rather than EfficientNetB0 as Faster R-CNN backbone | Faster R-CNN · COCO detection | mAP: 30.2 -> 30.5 | Helped |
| Use searched FBNetV3-E rather than EfficientNetB1 as Faster R-CNN backbone | Faster R-CNN · COCO detection | mAP: 32.2 -> 33.0 | Helped |
Closest papers by meaning
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.91)
- FP-NAS: Fast Probabilistic Neural Architecture Search (CVPR 2021 · similarity 0.91)
- NetAdaptV2: Efficient Neural Architecture Search With Fast Super-Network Training and Architecture Optimization (CVPR 2021 · similarity 0.90)
- NAS-FCOS: Fast Neural Architecture Search for Object Detection (CVPR 2020 · similarity 0.90)
- Encodings for Prediction-based Neural Architecture Search (ICML 2024 · similarity 0.89)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.88)
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.