---
title: "FBNet — paper by Wu et al. (2018) — summary and ideas"
description: "The authors propose DNAS, a differentiable search framework. 50.2% accuracy. 31 references, cited by 144 papers here."
authors: ["Bichen Wu", "Xiaoliang Dai", "Peizhao Zhang", "Yanghan Wang", "Fei Sun", "Yiming Wu", "Yuandong Tian", "Peter Vajda", "Yangqing Jia", "Kurt Keutzer"]
year: 2018
venue: "arXiv:1812.03443"
source: "https://arxiv.org/abs/1812.03443"
doi: "10.48550/arxiv.1812.03443"
arxiv: "1812.03443"
licence: "arXiv — non-exclusive license to distribute"
licence_url: "http://arxiv.org/licenses/nonexclusive-distrib/1.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/wu-2018-fbnet-hardware-aware-efficient-convnet/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/wu-2018-fbnet-hardware-aware-efficient-convnet/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2018](https://md.research.elamur.ai/free/en/papers/2018/) / FBNet: Hardware-Aware Efficient ConvNet Design via Differentiable Neural Architecture Search

# FBNet: Hardware-Aware Efficient ConvNet Design via Differentiable Neural Architecture Search

**Authors:** Bichen Wu, Xiaoliang Dai, Peizhao Zhang, Yanghan Wang, [Fei Sun](https://md.research.elamur.ai/free/en/authors/fei-sun/), Yiming Wu, [Yuandong Tian](https://md.research.elamur.ai/free/en/authors/yuandong-tian/), Peter Vajda, [Yangqing Jia](https://md.research.elamur.ai/free/en/authors/yangqing-jia/) and [Kurt Keutzer](https://md.research.elamur.ai/free/en/authors/kurt-keutzer/) · **Year:** 2018 · **Venue:** [arXiv:1812.03443](https://md.research.elamur.ai/free/en/venues/ieee-computer-society-conference-on-computer-vision-and-pattern-recognition/) ([IEEE Computer Society Conference on Computer Vision and Pattern Recognition 2018](https://md.research.elamur.ai/free/en/venues/ieee-computer-society-conference-on-computer-vision-and-pattern-recognition/2018/)) · **Publisher:** [Institute of Electrical and Electronics Engineers](https://md.research.elamur.ai/free/en/publishers/institute-of-electrical-and-electronics-engineers/) · **Licence:** [arXiv — non-exclusive license to distribute](http://arxiv.org/licenses/nonexclusive-distrib/1.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/1812.03443) · [PDF](https://arxiv.org/pdf/1812.03443) · [DOI 10.48550/arxiv.1812.03443](https://doi.org/10.48550/arxiv.1812.03443) · [arXiv 1812.03443](https://arxiv.org/abs/1812.03443) · [OpenAlex](https://openalex.org/W2967733054)

**Citations:** 1,261 citations (OpenAlex, 2026-10-14) · [144](https://md.research.elamur.ai/free/en/papers/wu-2018-fbnet-hardware-aware-efficient-convnet/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
Building ConvNets that are both accurate and fast on phones is hard. The space of possible architectures is enormous, earlier reinforcement-learning NAS methods cost thousands of GPU-hours, and the usual efficiency proxy (FLOPs) often fails to predict real latency. The authors propose DNAS, a differentiable search framework. It models a layer-wise search space, where every layer may pick its own block, as a stochastic super net. Using Gumbel Softmax, the distribution over architectures is trained directly with SGD rather than training many candidate networks one by one. The training loss combines cross-entropy with a latency penalty. Latency comes from a lookup table of operator timings measured on the actual target device, which makes the penalty differentiable with respect to block choices. The resulting FBNet family, evaluated on ImageNet, beats MobileNetV2 and MnasNet in the accuracy–latency trade-off at a fraction of the search cost. This makes it practical to redesign a network for each resolution, width, or device instead of using one model for everything.

## Key points
- **FBNet-B results:** Reaches 74.1% top-1 ImageNet accuracy while needing only 295M FLOPs and running in 23.1 ms on a Samsung S8. This is 2.4x smaller and 1.5x faster than MobileNetV2-1.3 at similar accuracy.
- **Search cost:** 216 GPU-hours, which the authors estimate is about 420x cheaper than MnasNet, while also getting better accuracy and latency.
- **Resolution and channel scaling:** Models searched for different input resolutions and channel scalings gain 1.5% to 6.4% top-1 over the corresponding MobileNetV2 models.
  - The smallest model, FBNet-96-0.35-1, reaches 50.2% accuracy at 2.9 ms (345 fps, batch size 1) on a Samsung S8.
  - It skips many layers, which the authors attribute to smaller inputs needing a smaller receptive field.
- **Device-specific search:** Using the same search space with different latency tables gives FBNet-iPhoneX and FBNet-S8. Their accuracy is nearly the same (73.20% vs. 73.27%).
  - On an iPhone X, the iPhone-optimized model runs in 19.84 ms, versus 27.53 ms for the S8-optimized one (about 1.4x faster).
  - On the S8 the ranking reverses: 22.12 ms for FBNet-S8 vs. 23.33 ms for FBNet-iPhoneX.
- **Release:** Models are released in the facebookresearch/mobile-vision repository.

## Interesting ideas
- **Additive latency lookup table:** The authors time each candidate operator once on the device and sum those timings to estimate a whole network's latency. This gives a cheap and differentiable cost estimate, and the same trick could work for any resource budget (memory, energy, inference cost) that is roughly additive over components.
- **Relaxing a discrete choice into a learnable distribution:** Instead of enumerating architectures, they learn probabilities over block choices with Gumbel Softmax and sample the final design. This applies wherever a discrete configuration choice (such as picking modules or layer types) would otherwise need expensive trial-and-error.
- **Fixed macro-structure with per-layer freedom:** Layer count and dimensions are fixed by hand, and only each layer's block type is searched. This keeps the search tractable and avoids the fragmented, hardware-unfriendly cells of earlier NAS work, which makes it a sensible template for constrained automated design.
- **Optimize the real metric, not a proxy:** The paper shows that FLOPs can mislead and measures latency on the actual target hardware. This is a reminder to validate efficiency claims on the deployment platform itself.
- **Cross-deployment evaluation matrix:** Each specialized model is run on every target device, not just its own. This cleanly shows that specialization matters and is easy to reuse when checking whether per-environment tuning pays off.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [Meta](https://md.research.elamur.ai/free/en/companies/meta/) — Company, Menlo Park, United States · 214 works · 241,492 citations
- [University of California, Berkeley](https://md.research.elamur.ai/free/en/universities/university-of-california-berkeley/) — University, Berkeley, United States · 209 works · 244,782 citations
- [Princeton University](https://md.research.elamur.ai/free/en/universities/princeton-university/) — University, Princeton, United States · 113 works · 48,612 citations

📚 Shelves (3): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [👁️ Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) · 3,486 works

🧭 Topics (1): [⚡ GPU & parallel computing](https://md.research.elamur.ai/free/en/topics/gpu-parallel-computing/) · 41 works

## Cite

Cite as: Bichen Wu, Xiaoliang Dai, Peizhao Zhang, Yanghan Wang, Fei Sun, Yiming Wu, Yuandong Tian, Peter Vajda, Yangqing Jia and Kurt Keutzer. 2018. FBNet: Hardware-Aware Efficient ConvNet Design via Differentiable Neural Architecture Search. arXiv:1812.03443.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@misc{wu2018fbnet,
  title = {{FBNet: Hardware-Aware Efficient ConvNet Design via Differentiable Neural Architecture Search}},
  author = {Bichen Wu, Xiaoliang Dai, Peizhao Zhang, Yanghan Wang, Fei Sun, Yiming Wu, Yuandong Tian, Peter Vajda, Yangqing Jia and Kurt Keutzer},
  year = {2018},
  doi = {10.48550/arxiv.1812.03443},
  eprint = {1812.03443},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.1812.03443}
}
```

- [← previous AMC: AutoML for Model Compression and Acceleration on Mobile Devices](https://md.research.elamur.ai/free/en/papers/he-2018-amc-automl-model-compression-acceleration/)
- [next → Contextual String Embeddings for Sequence Labeling](https://md.research.elamur.ai/free/en/papers/akbik-2018-contextual-string-embeddings-sequence-labeling/)
- [↑ back to papers of 2018](https://md.research.elamur.ai/free/en/papers/2018/)

## References

31 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Anonymous. Snas: stochastic neural architecture search. In Submitted to International Conference on Learning Representations, 2019. under review. [find](https://scholar.google.com/scholar?q=Anonymous%2E+Snas%3A+stochastic+neural+architecture+search%2E+In+Submitted+to+International+Conference+on+Learning+Representations%2C+2019%2E+under+review%2E)
2. J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei. Imagenet: A large-scale hierarchical image database. In Computer Vision and Pattern Recognition, 2009. CVPR 2009. IEEE Conference on, pages 248–255. Ieee, 2009. [find](https://scholar.google.com/scholar?q=J%2E+Deng%2C+W%2E+Dong%2C+R%2E+Socher%2C+L%2E%2DJ%2E+Li%2C+K%2E+Li%2C+and+L%2E+Fei%2DFei%2E+Imagenet%3A+A+large%2Dscale+hierarchical+image+database%2E+In+Computer+Vision+and+Pattern+Recognition%2C+2009%2E+CVPR+2009%2E+IEEE+Conference+on%2C+page)
3. A. Gholami, K. Kwon, B. Wu, Z. Tai, X. Yue, P. Jin, S. Zhao, and K. Keutzer. Squeezenext: Hardware-aware neural network design. arXiv preprint arXiv:1803.10615, 2018. — [arXiv:1803.10615](https://arxiv.org/abs/1803.10615) [find](https://scholar.google.com/scholar?q=A%2E+Gholami%2C+K%2E+Kwon%2C+B%2E+Wu%2C+Z%2E+Tai%2C+X%2E+Yue%2C+P%2E+Jin%2C+S%2E+Zhao%2C+and+K%2E+Keutzer%2E+Squeezenext%3A+Hardware%2Daware+neural+network+design%2E+arXiv+preprint+arXiv%3A1803%2E10615%2C+2018%2E+%E2%80%94+%5BarXiv%3A1803%2E10615%5D%28https%3A%2F%2Farxi)
4. K. He, X. Zhang, S. Ren, and J. Sun. [Deep residual learning for image recognition](https://md.research.elamur.ai/free/en/papers/he-2015-deep-residual-learning-image-recognition/). In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016.
5. Y. He, J. Lin, Z. Liu, H. Wang, L.-J. Li, and S. Han. [Amc: Automl for model compression and acceleration on mobile devices](https://md.research.elamur.ai/free/en/papers/he-2018-amc-automl-model-compression-acceleration/). In Proceedings of the European Conference on Computer Vision (ECCV), pages 784–800, 2018.
6. A. G. Howard, M. Zhu, B. Chen, D. Kalenichenko, W. Wang, T. Weyand, M. Andreetto, and H. Adam. [Mobilenets: Efficient convolutional neural networks for mobile vision applications](https://md.research.elamur.ai/free/en/papers/howard-2017-mobilenets-efficient-convolutional-neural-networks/). arXiv preprint arXiv:1704.04861, 2017.
7. G. Huang, S. Liu, L. van der Maaten, and K. Q. Weinberger. Condensenet: An efficient densenet using learned group convolutions. group, 3(12):11, 2017. [find](https://scholar.google.com/scholar?q=G%2E+Huang%2C+S%2E+Liu%2C+L%2E+van+der+Maaten%2C+and+K%2E+Q%2E+Weinberger%2E+Condensenet%3A+An+efficient+densenet+using+learned+group+convolutions%2E+group%2C+3%2812%29%3A11%2C+2017%2E)
8. F. N. Iandola, S. Han, M. W. Moskewicz, K. Ashraf, W. J. Dally, and K. Keutzer. Squeezenet: Alexnet-level accuracy with 50x fewer parameters and¡ 0.5 mb model size. arXiv preprint arXiv:1602.07360, 2016. [find](https://scholar.google.com/scholar?q=F%2E+N%2E+Iandola%2C+S%2E+Han%2C+M%2E+W%2E+Moskewicz%2C+K%2E+Ashraf%2C+W%2E+J%2E+Dally%2C+and+K%2E+Keutzer%2E+Squeezenet%3A+Alexnet%2Dlevel+accuracy+with+50x+fewer+parameters+and%C2%A1+0%2E5+mb+model+size%2E+arXiv+preprint+arXiv%3A1602%2E07360%2C+20)
9. E. Jang, S. Gu, and B. Poole. [Categorical reparameterization with gumbel-softmax](https://md.research.elamur.ai/free/en/papers/jang-2016-categorical-reparameterization-gumbel-softmax/). arXiv preprint arXiv:1611.01144, 2016.
10. D. P. Kingma and J. Ba. [Adam: A method for stochastic optimization](https://md.research.elamur.ai/free/en/papers/kingma-2014-adam-method-stochastic-optimization/). arXiv preprint arXiv:1412.6980, 2014.
11. C. Liu, B. Zoph, J. Shlens, W. Hua, L.-J. Li, L. Fei-Fei, A. Yuille, J. Huang, and K. Murphy. Progressive neural architecture search. arXiv preprint arXiv:1712.00559, 2017. [find](https://scholar.google.com/scholar?q=C%2E+Liu%2C+B%2E+Zoph%2C+J%2E+Shlens%2C+W%2E+Hua%2C+L%2E%2DJ%2E+Li%2C+L%2E+Fei%2DFei%2C+A%2E+Yuille%2C+J%2E+Huang%2C+and+K%2E+Murphy%2E+Progressive+neural+architecture+search%2E+arXiv+preprint+arXiv%3A1712%2E00559%2C+2017%2E)
12. H. Liu, K. Simonyan, and Y. Yang. Darts: Differentiable architecture search. arXiv preprint arXiv:1806.09055, 2018. [find](https://scholar.google.com/scholar?q=H%2E+Liu%2C+K%2E+Simonyan%2C+and+Y%2E+Yang%2E+Darts%3A+Differentiable+architecture+search%2E+arXiv+preprint+arXiv%3A1806%2E09055%2C+2018%2E)
13. N. Ma, X. Zhang, H.-T. Zheng, and J. Sun. Shufflenet v2: Practical guidelines for efficient cnn architecture design. arXiv preprint arXiv:1807.11164, 2018. — [arXiv:1807.11164](https://arxiv.org/abs/1807.11164) [find](https://scholar.google.com/scholar?q=N%2E+Ma%2C+X%2E+Zhang%2C+H%2E%2DT%2E+Zheng%2C+and+J%2E+Sun%2E+Shufflenet+v2%3A+Practical+guidelines+for+efficient+cnn+architecture+design%2E+arXiv+preprint+arXiv%3A1807%2E11164%2C+2018%2E+%E2%80%94+%5BarXiv%3A1807%2E11164%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F1)
14. C. J. Maddison, A. Mnih, and Y. W. Teh. [The concrete distribution: A continuous relaxation of discrete random variables](https://md.research.elamur.ai/free/en/papers/maddison-2016-concrete-distribution-continuous-relaxation-discrete/). arXiv preprint arXiv:1611.00712, 2016.
15. A. Paszke, S. Gross, S. Chintala, G. Chanan, E. Yang, Z. De-Vito, Z. Lin, A. Desmaison, L. Antiga, and A. Lerer. Automatic differentiation in pytorch. 2017. [find](https://scholar.google.com/scholar?q=A%2E+Paszke%2C+S%2E+Gross%2C+S%2E+Chintala%2C+G%2E+Chanan%2C+E%2E+Yang%2C+Z%2E+De%2DVito%2C+Z%2E+Lin%2C+A%2E+Desmaison%2C+L%2E+Antiga%2C+and+A%2E+Lerer%2E+Automatic+differentiation+in+pytorch%2E+2017%2E)
16. H. Pham, M. Y. Guan, B. Zoph, Q. V. Le, and J. Dean. [Efficient neural architecture search via parameter sharing](https://md.research.elamur.ai/free/en/papers/pham-2018-efficient-neural-architecture-search-parameter/). arXiv preprint arXiv:1802.03268, 2018.
17. M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen. Mobilenetv2: Inverted residuals and linear bottlenecks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4510–4520, 2018. [find](https://scholar.google.com/scholar?q=M%2E+Sandler%2C+A%2E+Howard%2C+M%2E+Zhu%2C+A%2E+Zhmoginov%2C+and+L%2E%2DC%2E+Chen%2E+Mobilenetv2%3A+Inverted+residuals+and+linear+bottlenecks%2E+In+Proceedings+of+the+IEEE+Conference+on+Computer+Vision+and+Pattern+Recognition%2C+p)
18. K. Simonyan and A. Zisserman. [Very deep convolutional networks for large-scale image recognition](https://md.research.elamur.ai/free/en/papers/simonyan-2014-very-deep-convolutional-networks-large/). arXiv preprint arXiv:1409.1556, 2014.
19. C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and A. Rabinovich. Going deeper with convolutions. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1–9, 2015. [find](https://scholar.google.com/scholar?q=C%2E+Szegedy%2C+W%2E+Liu%2C+Y%2E+Jia%2C+P%2E+Sermanet%2C+S%2E+Reed%2C+D%2E+Anguelov%2C+D%2E+Erhan%2C+V%2E+Vanhoucke%2C+and+A%2E+Rabinovich%2E+Going+deeper+with+convolutions%2E+In+Proceedings+of+the+IEEE+conference+on+computer+vision+and+p)
20. M. Tan, B. Chen, R. Pang, V. Vasudevan, and Q. V. Le. Mnasnet: Platform-aware neural architecture search for mobile. arXiv preprint arXiv:1807.11626, 2018. [find](https://scholar.google.com/scholar?q=M%2E+Tan%2C+B%2E+Chen%2C+R%2E+Pang%2C+V%2E+Vasudevan%2C+and+Q%2E+V%2E+Le%2E+Mnasnet%3A+Platform%2Daware+neural+architecture+search+for+mobile%2E+arXiv+preprint+arXiv%3A1807%2E11626%2C+2018%2E)
21. T. Veniat and L. Denoyer. Learning time/memory-efficient deep architectures with budgeted super networks. arXiv preprint arXiv:1706.00046, 2017. — [arXiv:1706.00046](https://arxiv.org/abs/1706.00046) [find](https://scholar.google.com/scholar?q=T%2E+Veniat+and+L%2E+Denoyer%2E+Learning+time%2Fmemory%2Defficient+deep+architectures+with+budgeted+super+networks%2E+arXiv+preprint+arXiv%3A1706%2E00046%2C+2017%2E+%E2%80%94+%5BarXiv%3A1706%2E00046%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F1706%2E00046%29)
22. B. Wu, F. N. Iandola, P. H. Jin, and K. Keutzer. Squeezedet: Unified, small, low power fully convolutional neural networks for real-time object detection for autonomous driving. In CVPR Workshops, pages 446–454, 2017. [find](https://scholar.google.com/scholar?q=B%2E+Wu%2C+F%2E+N%2E+Iandola%2C+P%2E+H%2E+Jin%2C+and+K%2E+Keutzer%2E+Squeezedet%3A+Unified%2C+small%2C+low+power+fully+convolutional+neural+networks+for+real%2Dtime+object+detection+for+autonomous+driving%2E+In+CVPR+Workshops%2C+pag)
23. B. Wu, A. Wan, X. Yue, P. Jin, S. Zhao, N. Golmant, A. Gholaminejad, J. Gonzalez, and K. Keutzer. Shift: A zero flop, zero parameter alternative to spatial convolutions. arXiv preprint arXiv:1711.08141, 2017. — [arXiv:1711.08141](https://arxiv.org/abs/1711.08141) [find](https://scholar.google.com/scholar?q=B%2E+Wu%2C+A%2E+Wan%2C+X%2E+Yue%2C+P%2E+Jin%2C+S%2E+Zhao%2C+N%2E+Golmant%2C+A%2E+Gholaminejad%2C+J%2E+Gonzalez%2C+and+K%2E+Keutzer%2E+Shift%3A+A+zero+flop%2C+zero+parameter+alternative+to+spatial+convolutions%2E+arXiv+preprint+arXiv%3A1711%2E0814)
24. B. Wu, A. Wan, X. Yue, and K. Keutzer. Squeezeseg: Convolutional neural nets with recurrent crf for real-time road-object segmentation from 3d lidar point cloud. In 2018 IEEE International Conference on Robotics and Automation (ICRA), pages 1887–1893. IEEE, 2018. [find](https://scholar.google.com/scholar?q=B%2E+Wu%2C+A%2E+Wan%2C+X%2E+Yue%2C+and+K%2E+Keutzer%2E+Squeezeseg%3A+Convolutional+neural+nets+with+recurrent+crf+for+real%2Dtime+road%2Dobject+segmentation+from+3d+lidar+point+cloud%2E+In+2018+IEEE+International+Conference+)
25. B. Wu, Y. Wang, P. Zhang, Y. Tian, P. Vajda, and K. Keutzer. Mixed precision quantization of convnets via differentiable neural architecture search. arXiv preprint arXiv:1812.00090, 2018. — [arXiv:1812.00090](https://arxiv.org/abs/1812.00090) [find](https://scholar.google.com/scholar?q=B%2E+Wu%2C+Y%2E+Wang%2C+P%2E+Zhang%2C+Y%2E+Tian%2C+P%2E+Vajda%2C+and+K%2E+Keutzer%2E+Mixed+precision+quantization+of+convnets+via+differentiable+neural+architecture+search%2E+arXiv+preprint+arXiv%3A1812%2E00090%2C+2018%2E+%E2%80%94+%5BarXiv%3A181)
26. B. Wu, X. Zhou, S. Zhao, X. Yue, and K. Keutzer. Squeezesegv2: Improved model structure and unsupervised domain adaptation for road-object segmentation from a lidar point cloud. arXiv preprint arXiv:1809.08495, 2018. — [arXiv:1809.08495](https://arxiv.org/abs/1809.08495) [find](https://scholar.google.com/scholar?q=B%2E+Wu%2C+X%2E+Zhou%2C+S%2E+Zhao%2C+X%2E+Yue%2C+and+K%2E+Keutzer%2E+Squeezesegv2%3A+Improved+model+structure+and+unsupervised+domain+adaptation+for+road%2Dobject+segmentation+from+a+lidar+point+cloud%2E+arXiv+preprint+arXiv%3A1)
27. T.-J. Yang, A. Howard, B. Chen, X. Zhang, A. Go, M. Sandler, V. Sze, and H. Adam. Netadapt: Platform-aware neural network adaptation for mobile applications. Energy, 41:46, 2018. [find](https://scholar.google.com/scholar?q=T%2E%2DJ%2E+Yang%2C+A%2E+Howard%2C+B%2E+Chen%2C+X%2E+Zhang%2C+A%2E+Go%2C+M%2E+Sandler%2C+V%2E+Sze%2C+and+H%2E+Adam%2E+Netadapt%3A+Platform%2Daware+neural+network+adaptation+for+mobile+applications%2E+Energy%2C+41%3A46%2C+2018%2E)
28. Y. Yang, Q. Huang, B. Wu, T. Zhang, L. Ma, G. Gambardella, M. Blott, L. Lavagno, K. Vissers, J. Wawrzynek, et al. Synetgy: Algorithm-hardware co-design for convnet accelerators on embedded fpgas. arXiv preprint arXiv:1811.08634, 2018. — [arXiv:1811.08634](https://arxiv.org/abs/1811.08634) [find](https://scholar.google.com/scholar?q=Y%2E+Yang%2C+Q%2E+Huang%2C+B%2E+Wu%2C+T%2E+Zhang%2C+L%2E+Ma%2C+G%2E+Gambardella%2C+M%2E+Blott%2C+L%2E+Lavagno%2C+K%2E+Vissers%2C+J%2E+Wawrzynek%2C+et+al%2E+Synetgy%3A+Algorithm%2Dhardware+co%2Ddesign+for+convnet+accelerators+on+embedded+fpgas%2E+arXi)
29. X. Zhang, X. Zhou, M. Lin, and J. Sun. Shufflenet: An extremely efficient convolutional neural network for mobile devices. arxiv 2017. arXiv preprint arXiv:1707.01083. — [arXiv:1707.01083](https://arxiv.org/abs/1707.01083) [find](https://scholar.google.com/scholar?q=X%2E+Zhang%2C+X%2E+Zhou%2C+M%2E+Lin%2C+and+J%2E+Sun%2E+Shufflenet%3A+An+extremely+efficient+convolutional+neural+network+for+mobile+devices%2E+arxiv+2017%2E+arXiv+preprint+arXiv%3A1707%2E01083%2E+%E2%80%94+%5BarXiv%3A1707%2E01083%5D%28https%3A%2F%2Farx)
30. B. Zoph and Q. V. Le. [Neural architecture search with reinforcement learning](https://md.research.elamur.ai/free/en/papers/zoph-2016-neural-architecture-search-reinforcement-learning/). arXiv preprint arXiv:1611.01578, 2016.
31. B. Zoph, V. Vasudevan, J. Shlens, and Q. V. Le. [Learning transferable architectures for scalable image recognition](https://md.research.elamur.ai/free/en/papers/zoph-2017-learning-transferable-architectures-scalable-image/). arXiv preprint arXiv:1707.07012, 2(6), 2017.

*Updated 2 November 2026*