2016

Normalization Propagation: A Parametric Technique for Removing Internal Covariate Shift in Deep Networks

Arpit, Devansh, Zhou, Yingbo, Kota, Bhargava U. et al.

Understand

While the authors of Batch Normalization (BN) identify and address an important problem involved in training deep networks-- Internal Covariate Shift-- the current solution has certain drawbacks.

  • Specifically, BN depends on batch statistics for layerwise input normalization during training which makes the estimates of mean and standard deviation of input (distribution) to hidden layers inaccurate for validation due to shifting parameter values (especially during initial training epochs).
  • Also, BN cannot be used with batch-size 1 during training.
  • We address these drawbacks by proposing a non-adaptive normalization technique for removing internal covariate shift, that we call Normalization Propagation.

Built on

  • Independent component analysis: algorithms and applications

    Hyvärinen, Aapo and Oja, Erkki · 2000

    Earlier work this paper cites.

  • Improving predictive inference under covariate shift by weighting the log-likelihood function

    Shimodaira, Hidetoshi · 2000

    Earlier work this paper cites.

  • Learning Multiple Layers of Features from Tiny Images

    Krizhevsky, Alex · 2009

    Earlier work this paper cites.

  • Understanding the difficulty of training deep feedforward neural networks

    Glorot, Xavier and Bengio, Yoshua · 2010

    Earlier work this paper cites.

  • Sparse representation for computer vision and pattern recognition

    Wright, John, Ma, Yi, Mairal, Julien, Sapiro, Guillermo, Huang, Thomas S, and Yan, Shuicheng · 2010

    Earlier work this paper cites.

  • Reading digits in natural images with unsupervised feature learning

    Netzer, Yuval, Wang, Tao, Coates, Adam, Bissacco, Alessandro, Wu, Bo, and Ng, Andrew Y · 2011

    Earlier work this paper cites.

Similar

  • Efficient backprop

    LeCun, Yann A, Bottou, Léon, Orr, Genevieve B, and Müller, Klaus-Robert · 2012

    Cited alongside, same era.

  • Lecture 6.5-rmsprop: Divide the gradient by a running average of its recent magnitude

    Tieleman, Tijmen and Hinton, Geoffrey · 2012

    Cited alongside, same era.

  • Maxout networks

    Goodfellow, Ian J., Warde-farley, David, Mirza, Mehdi, Courville, Aaron, and Bengio, Yoshua · 2013

    Cited alongside, same era.

  • k-sparse autoencoders

    Original

    Makhzani, Alireza and Frey, Brendan · 2013

    Cited alongside, same era.

  • Exact solutions to the nonlinear dynamics of learning in deep linear neural networks

    Original

    Saxe, Andrew M, McClelland, James L, and Ganguli, Surya · 2013

    Cited alongside, same era.

  • Network in network

    Lin, Min, Chen, Qiang, and Yan, Shuicheng · 2014

    Cited alongside, same era.

Then

  • Learning activation functions to improve deep neural networks

    Agostinelli, Forest, Hoffman, Matthew, Sadowski, Peter, and Baldi, Pierre · 2015

    Later among the works it cites.

  • Natural neural networks

    Desjardins, Guillaume, Simonyan, Karen, Pascanu, Razvan, et al · 2015

    Later among the works it cites.

  • Delving deep into rectifiers: Surpassing human-level performance on imagenet classification

    He, Kaiming, Zhang, Xiangyu, Ren, Shaoqing, and Sun, Jian · 2015

    Later among the works it cites.

  • Batch normalization: Accelerating deep network training by reducing internal covariate shift

    Ioffe, Sergey and Szegedy, Christian · 2015

    Later among the works it cites.

  • Deeply-supervised nets

    Lee, Chen-Yu, Xie, Saining, Gallagher, Patrick, Zhang, Zhengyou, and Tu, Zhuowen · 2015

    Later among the works it cites.

Beyond the bibliography

alphaXiv searches the wider corpus for related work and actual follow-ups.

Open on alphaXiv

alphaXiv is searching for related work…