Pipeline Latency Optimization

Pipeline Latency Optimization is a methodological framework within distributed deep learning that minimizes the time delay in propagating activations and gradients across pipeline stages by strategically overlapping computation and communication while balancing stage workloads to reduce idle periods.

77

Publications

5.4K

Citations

320

Authors

101

Institutions

Publications per year

2017–2026

39

Authors

320

  1. KC
    Kevin Chan

    DEVCOM Army Research Laboratory, United States

    2 publications · 44 citations

    2

    h-index

  2. RP
    Ryan Prenger

    University of California, Berkeley, United States

    2 publications · 2.4K citations

    2

    h-index

  3. NS
    Nam Sung Kim

    University of Illinois Urbana-Champaign, United States

    2 publications · 68 citations

    2

    h-index

  4. GD
    Greg Diamos

    Google (United States), United States

    2 publications · 2.4K citations

    2

    h-index

  5. ZC
    Zeshan Chishti

    Intel (United States), United States

    2 publications · 32 citations

    2

    h-index

1–5 of 320

Venues

  1. arXiv (Cornell University)

    Repository · 6 publications · 2.7K citations

    6

    h-index

  2. IEEE Transactions on Mobile Computing

    TMCCI · 3 publications · 84 citations

    3

    h-index

  3. Classical and Quantum Gravity

    2 publications · 208 citations

    2

    h-index

  4. ACM SIGARCH Computer Architecture News

    2 publications · 39 citations

    2

    h-index

  5. Real-Time Systems

    1 publications · 28 citations

    1

    h-index

1–5