Concepedia

Mixed Precision Training

Mixed Precision Training is a computational optimization method in deep learning that, as a variant of gradient-based training, utilizes a combination of lower-precision (e.g., FP16) and higher-precision (e.g., FP32) numerical formats during forward and backward passes to reduce memory usage and accelerate training while maintaining numerical stability and model accuracy.

80

Publications

7.7K

Citations

432

Authors

112

Institutions

Publications per year

2017–2026

79

Parents

Publications

80

fairseq: A Fast, Extensible Toolkit for Sequence Modeling

Myle Ott, Sergey Edunov, Alexei Baevski et al. · 2019 · 2.5K citations · Full text

, , +17

Mixed Precision Training

Paulius Micikevicius, Sharan Narang, Jonah Alben et al. · arXiv (Cornell University) · 2017 · 881 citations · Full text

, , +15

HAWQ: Hessian AWare Quantization of Neural Networks With Mixed-Precision

Zhen Dong, Zhewei Yao, Amir Gholami et al. · 2019 · 458 citations

, , +14

ZeroQ: A Novel Zero Shot Quantization Framework

Yaohui Cai, Zhewei Yao, Zhen Dong et al. · 2020 · 378 citations

, , +12

Highly Scalable Deep Learning Training System with Mixed-Precision: Training ImageNet in Four Minutes

Xianyan Jia, Shutao Song, Wei He et al. · arXiv (Cornell University) · 2018 · 313 citations · Full text

, , +17

1–5 of 80

Authors

432

  1. BG
    Boris Ginsburg

    Nvidia (United States), United States

    5 publications · 1.3K citations

    5

    h-index

  2. NM
    Naveen Mellempudi

    Intel (United States), United States

    5 publications · 278 citations

    5

    h-index

  3. OK
    Oleksii Kuchaiev

    University of California, Irvine, United States

    5 publications · 1.3K citations

    5

    h-index

  4. BK
    Bharat Kaul

    Intel (United States), United States

    5 publications · 278 citations

    5

    h-index

  5. ML
    Manuel Le Gallo

    IBM Research - Zurich, Switzerland

    4 publications · 232 citations

    4

    h-index

1–5 of 432

Institutions

112

  1. Intel (United States)

    Santa Clara, United States

    5 h-index

    23 publications

    960 citations

  2. Peking University

    Beijing, China

    3 h-index

    8 publications

    649 citations

  3. Google (United States)

    Mountain View, United States

    3 h-index

    3 publications

    400 citations

  4. University of California, Berkeley

    Berkeley, United States

    3 h-index

    15 publications

    4.3K citations

  5. Harvard University Press

    Cambridge, United States

    3 h-index

    3 publications

    135 citations

  6. University of Illinois Urbana-Champaign

    Urbana, United States

    3 h-index

    10 publications

    406 citations

  7. University of Illinois Chicago

    Chicago, United States

    2 h-index

    2 publications

    122 citations

  8. IBM Research - Zurich

    Rüschlikon, Switzerland

    2 h-index

    14 publications

    1.5K citations

1–8 of 112

Venues

  1. arXiv (Cornell University)

    Repository · 24 publications · 2.5K citations

    18

    h-index

  2. Proceedings of the AAAI Conference on Artificial Intelligence

    Conference · 3 publications · 127 citations

    3

    h-index

  3. Sensors

    2 publications · 39 citations

    2

    h-index

  4. 2

    h-index

  5. bioRxiv (Cold Spring Harbor Laboratory)

    Repository · 1 publications · 39 citations

    1

    h-index

1–5