Reducing communication in proximal Newton methods for sparse least squares problems

Saeed Soori; Aditya Devarakonda; Zachary Blanco; James Demmel; Mert Gürbüzbalaban; Maryam Mehri Dehnavi

doi:10.1145/3225058.3225131

Conference paper, 2018

Reducing communication in proximal Newton methods for sparse least squares problems

Saeed Soori, Aditya Devarakonda, Zachary Blanco, James Demmel, Mert Gürbüzbalaban, Maryam Mehri Dehnavi

International Conference on Parallel Processing (ICPP), 2018.

Publisher (DOI)

In brief

Proximal Newton methods for sparse least-squares problems spend most of their time communicating when run on a distributed-memory machine. Reformulating the inner iterations so that several steps’ worth of information is exchanged at once reduces the number of communication rounds by a tunable factor, and the resulting method is faster on large problems without changing what it computes.

Topics

Distributed & decentralized learning

Cite

@inproceedings{soori2018communication,
  title   = {Reducing communication in proximal Newton methods for sparse least squares problems},
  author  = {Saeed Soori and Aditya Devarakonda and Zachary Blanco and James Demmel and Mert Gürbüzbalaban and Maryam Mehri Dehnavi},
  year    = {2018},
  booktitle = {International Conference on Parallel Processing (ICPP)},
  doi     = {10.1145/3225058.3225131},
}