Skip to main content
SpringerLink
Log in

Channel Selection Using Gumbel Softmax

  • Conference paper
  • First Online:
Computer Vision – ECCV 2020 (ECCV 2020)

Part of the book series: Lecture Notes in Computer Science ((LNIP,volume 12372))

Included in the following conference series:

Abstract

Important applications such as mobile computing require reducing the computational costs of neural network inference. Ideally, applications would specify their preferred tradeoff between accuracy and speed, and the network would optimize this end-to-end, using classification error to remove parts of the network. Increasing speed can be done either during training – e.g., pruning filters – or during inference – e.g., conditionally executing a subset of the layers. We propose a single end-to-end framework that can improve inference efficiency in both settings. We use a combination of batch activation loss and classification loss, and Gumbel reparameterization to learn network structure. We train end-to-end, and the same technique supports pruning as well as conditional computation. We obtain promising experimental results for ImageNet classification with ResNet (45–52% less computation).

This is a preview of subscription content, log in via an institution to check access.

Access this chapter

Log in via an institution
Chapter
USD 29.95
Price excludes VAT (USA)
  • Available as PDF
  • Read on any device
  • Instant download
  • Own it forever
eBook
USD 84.99
Price excludes VAT (USA)
  • Available as EPUB and PDF
  • Read on any device
  • Instant download
  • Own it forever
Softcover Book
USD 109.99
Price excludes VAT (USA)
  • Compact, lightweight edition
  • Dispatched in 3 to 5 business days
  • Free shipping worldwide - see info

Tax calculation will be finalised at checkout

Purchases are for personal use only

Institutional subscriptions

Notes

  1. 1.

    See https://github.com/andreasveit/convnet-aig.

  2. 2.

    Note that the number we use for their FLOPs is different from what they report. They report lower FLOPs for the baseline ResNet-50 architecture (3.8 GFLOPs versus our 4.028). To normalize the comparison, we added 0.2 GFLOPs to their results.

References

  1. Bengio, Y.: Deep learning of representations: looking forward. CoRR abs/1305.0445 (2013). http://arxiv.org/abs/1305.0445

  2. Bengio, Y., Léonard, N., Courville, A.C.: Estimating or propagating gradients through stochastic neurons for conditional computation. CoRR abs/1308.3432 (2013). http://arxiv.org/abs/1308.3432

  3. Bolukbasi, T., Wang, J., Dekel, O., Saligrama, V.: Adaptive neural networks for efficient inference. In: ICML, pp. 527–536 (2017)

    Google Scholar 

  4. Cai, H., Zhu, L., Han, S.: ProxylessNAS: direct neural architecture search on target task and hardware. arXiv:1812.00332 (2018)

  5. Courbariaux, M., Bengio, Y., David, J.P.: BinaryConnect: training deep neural networks with binary weights during propagations. In: NIPS, pp. 3123–3131 (2015)

    Google Scholar 

  6. Deng, J., Dong, W., Socher, R., Li, L.J., Li, K., Fei-Fei, L.: ImageNet: a large-scale hierarchical image database. In: CVPR, pp. 248–255. IEEE (2009)

    Google Scholar 

  7. Figurnov, M., et al.: Spatially adaptive computation time for residual networks. In: CVPR (2017)

    Google Scholar 

  8. Gal, Y., Hron, J., Kendall, A.: Concrete dropout. In: Advances in Neural Information Processing Systems, pp. 3581–3590 (2017)

    Google Scholar 

  9. Graves, A.: Adaptive computation time for recurrent neural networks. CoRR abs/1603.08983 (2016)

    Google Scholar 

  10. Gumbel, E.J.: Statistical theory of extreme values and some practical applications. NBS Appl. Math. Ser. 33 (1954)

    Google Scholar 

  11. Han, S., Mao, H., Dally, W.J.: Deep compression: compressing deep neural networks with pruning, trained quantization and Huffman coding. arXiv preprint arXiv:1510.00149 (2015)

  12. Han, S., Pool, J., Tran, J., Dally, W.: Learning both weights and connections for efficient neural network. In: Advances in Neural Information Processing Systems, pp. 1135–1143 (2015)

    Google Scholar 

  13. Hassibi, B., Stork, D.G.: Second order derivatives for network pruning: optimal brain surgeon. In: Advances in Neural Information Processing Systems, pp. 164–171 (1993)

    Google Scholar 

  14. He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: CVPR, pp. 770–778 (2016)

    Google Scholar 

  15. He, Y., Liu, P., Wang, Z., Hu, Z., Yang, Y.: Filter pruning via geometric median for deep convolutional neural networks acceleration. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 4340–4349 (2019)

    Google Scholar 

  16. He, Y., Lin, J., Liu, Z., Wang, H., Li, L.J., Han, S.: AMC: AutoML for model compression and acceleration on mobile devices. In: ECCV, pp. 784–800 (2018)

    Google Scholar 

  17. He, Y., Zhang, X., Sun, J.: Channel pruning for accelerating very deep neural networks. In: ICCV (2017)

    Google Scholar 

  18. Hu, H., Peng, R., Tai, Y.W., Tang, C.K.: Network trimming: a data-driven neuron pruning approach towards efficient deep architectures. arXiv preprint arXiv:1607.03250 (2016)

  19. Huang, G., Chen, D., Li, T., Wu, F., van der Maaten, L., Weinberger, K.Q.: Multi-scale dense convolutional networks for efficient prediction. CoRR, abs/1703.09844 2 (2017)

    Google Scholar 

  20. Huang, G., Sun, Y., Liu, Z., Sedra, D., Weinberger, K.Q.: Deep networks with stochastic depth. In: Leibe, B., Matas, J., Sebe, N., Welling, M. (eds.) ECCV 2016. LNCS, vol. 9908, pp. 646–661. Springer, Cham (2016). https://doi.org/10.1007/978-3-319-46493-0_39

    Chapter  Google Scholar 

  21. Huang, Z., Wang, N.: Data-driven sparse structure selection for deep neural networks. In: ECCV (2018)

    Google Scholar 

  22. Jang, E., Gu, S., Poole, B.: Categorical reparameterization with gumbel-softmax. In: ICLR (2017). arXiv preprint arXiv:1611.01144

  23. Kingma, D.P., Salimans, T., Welling, M.: Variational dropout and the local reparameterization trick. In: Cortes, C., Lawrence, N.D., Lee, D.D., Sugiyama, M., Garnett, R. (eds.) NIPS, pp. 2575–2583. Curran Associates, Inc. (2015). http://papers.nips.cc/paper/5666-variational-dropout-and-the-local-reparameterization-trick.pdf

  24. Krizhevsky, A., Sutskever, I., Hinton, G.E.: ImageNet classification with deep convolutional neural networks. In: Pereira, F., Burges, C.J.C., Bottou, L., Weinberger, K.Q. (eds.) Advances in Neural Information Processing Systems 25, pp. 1097–1105. Curran Associates, Inc. (2012). http://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf

  25. LeCun, Y., Bengio, Y., Hinton, G.: Deep learning. Nature 521, 436–444 (2015). https://doi.org/10.1038/nature14539

  26. LeCun, Y., Denker, J.S., Solla, S.A.: Optimal brain damage. In: Advances in Neural Information Processing Systems, pp. 598–605 (1990)

    Google Scholar 

  27. Li, H., Kadav, A., Durdanovic, I., Samet, H., Graf, H.P.: Pruning filters for efficient ConvNets. In: ICLR (2017)

    Google Scholar 

  28. Li, H., Lin, Z., Shen, X., Brandt, J., Hua, G.: A convolutional neural network cascade for face detection. In: CVPR, pp. 5325–5334 (2015)

    Google Scholar 

  29. Liu, Z., Sun, M., Zhou, T., Huang, G., Darrell, T.: Rethinking the value of network pruning. In: ICLR (2019). https://openreview.net/forum?id=rJlnB3C5Ym

  30. Louizos, C., Welling, M., Kingma, D.P.: Learning sparse neural networks through \( l\_0 \) regularization. In: ICLR (2017). arXiv preprint arXiv:1712.01312

  31. Luo, J.H., Wu, J.: AutoPruner: an end-to-end trainable filter pruning method for efficient deep model inference. arXiv preprint arXiv:1805.08941 (2018)

  32. Luo, J.H., Wu, J., Lin, W.: ThiNet: a filter level pruning method for deep neural network compression. In: ICCV (2017). arXiv preprint arXiv:1707.06342

  33. Maddison, C.J., Mnih, A., Teh, Y.W.: The concrete distribution: a continuous relaxation of discrete random variables. arXiv preprint arXiv:1611.00712 (2016)

  34. McGill, M., Perona, P.: Deciding how to decide: dynamic routing in artificial neural networks. In: ICML (2017). arXiv preprint arXiv:1703.06217

  35. Molchanov, D., Ashukha, A., Vetrov, D.: Variational dropout sparsifies deep neural networks. In: Precup, D., Teh, Y.W. (eds.) ICML, vol. 70, pp. 2498–2507. PMLR, 06–11 August 2017. http://proceedings.mlr.press/v70/molchanov17a.html

  36. Molchanov, P., Tyree, S., Karras, T., Aila, T., Kautz, J.: Pruning convolutional neural networks for resource efficient inference. In: ICLR (2016). arXiv preprint arXiv:1611.06440

  37. Mozer, M.C., Smolensky, P.: Skeletonization: A technique for trimming the fat from a network via relevance assessment. In: Advances in Neural Information Processing Systems, pp. 107–115 (1989)

    Google Scholar 

  38. Paszke, A., et al.: Automatic differentiation in PyTorch. In: NIPS-W (2017)

    Google Scholar 

  39. Sandler, M., Howard, A., Zhu, M., Zhmoginov, A., Chen, L.C.: MobileNetV2: inverted residuals and linear bottlenecks. In: CVPR, pp. 4510–4520 (2018)

    Google Scholar 

  40. Shazeer, N., et al.: Outrageously large neural networks: the sparsely-gated mixture-of-experts layer. arXiv preprint arXiv:1701.06538 (2017)

  41. Shirakawa, S., Iwata, Y., Akimoto, Y.: Dynamic optimization of neural network structures using probabilistic modeling. In: Thirty-Second AAAI Conference on Artificial Intelligence (2018)

    Google Scholar 

  42. Simonyan, K., Zisserman, A.: Very deep convolutional networks for large-scale image recognition. CoRR abs/1409.1556 (2014)

    Google Scholar 

  43. Srinivas, S., Babu, R.V.: Generalized dropout. arXiv preprint arXiv:1611.06791 (2016)

  44. Srinivas, S., Babu, V.: Learning neural network architectures using backpropagation. In: BMVC, pp. 104.1–104.11, September 2016. https://dx.doi.org/10.5244/C.30.104

  45. Srinivas, S., Subramanya, A., Babu, R.V.: Training sparse neural networks. In: CVPR Workshops, July 2017

    Google Scholar 

  46. Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., Salakhutdinov, R.: Dropout: a simple way to prevent neural networks from overfitting. J. Mach. Learn. Res. 15(1), 1929–1958 (2014)

    MathSciNet  MATH  Google Scholar 

  47. Suau, X., Zappella, L., Palakkode, V., Apostoloff, N.: Principal filter analysis for guided network compression. arXiv preprint arXiv:1807.10585 (2018)

  48. Teerapittayanon, S., McDanel, B., Kung, H.: BranchyNet: fast inference via early exiting from deep neural networks. In: ICPR, pp. 2464–2469. IEEE (2016)

    Google Scholar 

  49. Veit, A., Belongie, S.: Convolutional networks with adaptive inference graphs. In: ECCV (2017). https://github.com/andreasveit/convnet-aig

  50. Viola, P., Jones, M.J.: Robust real-time face detection. Int. J. Comput. Vis. 57(2), 137–154 (2004)

    Article  Google Scholar 

  51. Yang, F., Choi, W., Lin, Y.: Exploit all the layers: fast and accurate CNN object detector with scale dependent pooling and cascaded rejection classifiers. In: CVPR, pp. 2129–2137 (2016)

    Google Scholar 

  52. You, Z., Yan, K., Ye, J., Ma, M., Wang, P.: Gate decorator: global filter pruning method for accelerating deep convolutional neural networks. In: Advances in Neural Information Processing Systems, pp. 2130–2141 (2019)

    Google Scholar 

  53. Zhu, M., Gupta, S.: To prune, or not to prune: exploring the efficacy of pruning for model compression. arXiv preprint arXiv:1710.01878 (2017)

  54. Zoph, B., Vasudevan, V., Shlens, J., Le, Q.V.: Learning transferable architectures for scalable image recognition. In: CVPR, pp. 8697–8710 (2018)

    Google Scholar 

Download references

Acknowledgments

We thank Andreas Veit and Serge Belongie for their invaluable insights on AIG and several reviewers for helpful comments. This work was generously supported by Google Cloud, without whose help it could not have been completed. It was funded by NSF grant IIS-1447473, by a gift from Sensetime, and by a Google Faculty Research Award.

Author information

Authors and Affiliations

  1. Cornell Tech, New York, USA

    Charles Herrmann, Richard Strong Bowen & Ramin Zabih

  2. Google Research, New York, USA

    Ramin Zabih

Authors
  1. Charles Herrmann
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Richard Strong Bowen
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Ramin Zabih
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Charles Herrmann .

Editor information

Editors and Affiliations

  1. University of Oxford, Oxford, UK

    Andrea Vedaldi

  2. Graz University of Technology, Graz, Austria

    Horst Bischof

  3. University of Freiburg, Freiburg im Breisgau, Germany

    Thomas Brox

  4. University of North Carolina at Chapel Hill, Chapel Hill, NC, USA

    Jan-Michael Frahm

1 Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary material 1 (pdf 1430 KB)

Rights and permissions

Reprints and permissions

Copyright information

© 2020 Springer Nature Switzerland AG

About this paper

Check for updates. Verify currency and authenticity via CrossMark

Cite this paper

Herrmann, C., Bowen, R.S., Zabih, R. (2020). Channel Selection Using Gumbel Softmax. In: Vedaldi, A., Bischof, H., Brox, T., Frahm, JM. (eds) Computer Vision – ECCV 2020. ECCV 2020. Lecture Notes in Computer Science(), vol 12372. Springer, Cham. https://doi.org/10.1007/978-3-030-58583-9_15

Download citation

  • DOI: https://doi.org/10.1007/978-3-030-58583-9_15

  • Published:

  • Publisher Name: Springer, Cham

  • Print ISBN: 978-3-030-58582-2

  • Online ISBN: 978-3-030-58583-9

  • eBook Packages: Computer ScienceComputer Science (R0)

Publish with us

Policies and ethics

Search

Navigation