2017

Improving Deep Learning by Inverse Square Root Linear Units (ISRLUs)

Carlile, Brad, Delamarter, Guy, Kinney, Paul et al.

Understand

We introduce the "inverse square root linear unit" (ISRLU) to speed up learning in deep neural networks.

  • ISRLU has better performance than ELU but has many of the same benefits.
  • ISRLU and ELU have similar curves and characteristics.
  • Both have negative values, allowing them to push mean unit activation closer to zero, and bring the normal gradient closer to the unit natural gradient, ensuring a noise-robust deactivation state, lessening the over fitting risk.

Built on

  • Arithmetic Complexity of Computations , volume 33 of CBMS-NSF Regional Conference Series in Applied Mathematics

    Shmuel Winograd · 1980

    Earlier work this paper cites.

  • Programming the FPS T Series

    John L. Gustafson · 1986

    Earlier work this paper cites.

  • Algorithms and design: the CRAY APP shared-memory system

    Brad Carlile · 1993

    Earlier work this paper cites.

  • Long short-term memory

    Sepp Hochreiter and Jürgen Schmidhuber · 1997

    Earlier work this paper cites.

  • Natural gradient works efficiently in learning

    Shun-Ichi Amari · 1998

    Earlier work this paper cites.

  • Fast inverse square root

    Chris Lomont · 2003

    Earlier work this paper cites.

Similar

  • Deep sparse rectifier neural networks

    Xavier Glorot, Antoine Bordes, and Yoshua Bengio · 2011

    Cited alongside, same era.

  • Empirical evaluation of gated recurrent neural networks on sequence modeling

    Original

    Junyoung Chung, Çaglar Gülçehre, KyungHyun Cho, and Yoshua Bengio · 2014

    Cited alongside, same era.

  • Fast and accurate deep network learning by exponential linear units (ELUs)

    Original

    Djork-Arné Clevert, Thomas Unterthiner, and Sepp Hochreiter · 2015

    Cited alongside, same era.

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

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

    Cited alongside, same era.

  • Tensorflow: A system for large-scale machine learning

    Martín Abadi, Paul Barham, Jianmin Chen, Zhifeng Chen, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Geoffrey Irving, Michael Isard, Manjunath Kudlur, Josh Levenberg, Rajat Monga, Sherry Moore, Derek G. Murray, Benoit Steiner, Paul Tucker, Vijay Vasudevan, Pete Warden, Martin Wicke, Yuan Yu, and Xiaoqiang Zheng · 2016

    Cited alongside, same era.

  • Parallelism, compute intensity, and data vectorization

    Brad Carlile

    Cited in the paper.

Then

  • Fast algorithms for convolutional neural networks

    Andrew Lavin and Scott Gray · 2016

    Later among the works it cites.

  • Deep residual networks with exponential linear unit

    Anish Shah, Eashan Kadam, Hena Shah, Sameer Shinde, and Sandip Shingade · 2016

    Later among the works it cites.

  • Rethinking the inception architecture for computer vision

    Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna · 2016

    Later among the works it cites.

  • Intel, Sep 2017

    Vector Mathematics (VM) Performance and Accuracy Data · 2017

    Closest in time.

  • Self-normalizing neural networks

    Original

    Günter Klambauer, Thomas Unterthiner, Andreas Mayr, and Sepp Hochreiter · 2017

    Closest in time.

Beyond the bibliography

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

Open on alphaXiv

alphaXiv is searching for related work…