L-DQN: an asynchronous limited-memory distributed quasi-Newton method

Bugra Can; Saeed Soori; Maryam Mehri Dehnavi; Mert Gürbüzbalaban

doi:10.1109/CDC45484.2021.9682985

Conference paper, 2021

L-DQN: an asynchronous limited-memory distributed quasi-Newton method

Bugra Can, Saeed Soori, Maryam Mehri Dehnavi, Mert Gürbüzbalaban

IEEE Conference on Decision and Control (CDC), 2021.

Publisher (DOI)arXiv:2108.09365

In brief

A quasi-Newton method for a primary–worker system in which workers hold their own data and update asynchronously. L-DQN keeps limited-memory curvature information so that each update is cheap and the primary node never waits for the slowest worker, and the analysis shows the method still converges linearly despite the asynchrony, with experiments on large-scale problems.

Topics

Distributed & decentralized learning

Cite

@inproceedings{can2021ldqn,
  title   = {L-DQN: an asynchronous limited-memory distributed quasi-Newton method},
  author  = {Bugra Can and Saeed Soori and Maryam Mehri Dehnavi and Mert Gürbüzbalaban},
  year    = {2021},
  booktitle = {IEEE Conference on Decision and Control (CDC)},
  doi     = {10.1109/CDC45484.2021.9682985},
}