eduzhai > Applied Sciences > Engineering >

Coded Distributed Computing with Partial Recovery

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 13 pages

Abstract: Coded computation techniques provide robustness against straggling workers indistributed computing. However, most of the existing schemes require exactprovisioning of the straggling behaviour and ignore the computations carriedout by straggling workers. Moreover, these schemes are typically designed torecover the desired computation results accurately, while in many machinelearning and iterative optimization algorithms, faster approximate solutionsare known to result in an improvement in the overall convergence time. In thispaper, we first introduce a novel coded matrix-vector multiplication scheme,called coded computation with partial recovery (CCPR), which benefits from theadvantages of both coded and uncoded computation schemes, and reduces both thecomputation time and the decoding complexity by allowing a trade-off betweenthe accuracy and the speed of computation. We then extend this approach todistributed implementation of more general computation tasks by proposing acoded communication scheme with partial recovery, where the results of subtaskscomputed by the workers are coded before being communicated. Numericalsimulations on a large linear regression task confirm the benefits of theproposed distributed computation scheme with partial recovery in terms of thetrade-off between the computation accuracy and latency.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×