IDEAL: inexact decentralized accelerated augmented Lagrangian method

Yossi Arjevani; Joan Bruna; Bugra Can; Mert Gürbüzbalaban; Stefanie Jegelka; Hongzhou Lin

Conference paper, 2020

IDEAL: inexact decentralized accelerated augmented Lagrangian method

Yossi Arjevani, Joan Bruna, Bugra Can, Mert Gürbüzbalaban, Stefanie Jegelka, Hongzhou Lin

Advances in Neural Information Processing Systems (NeurIPS), 33, pp. 20648–20659, 2020.

Publisher page arXiv:2006.06733

In brief

A framework for decentralized optimization that reduces the problem to an augmented Lagrangian solved inexactly with any inner solver. With an accelerated inner method, IDEAL attains optimal communication complexity in the condition number and the network’s spectral gap for smooth strongly convex problems, unifying and improving on several earlier decentralized accelerated methods.

Topics

Distributed & decentralized learning Momentum & acceleration

Cite

@inproceedings{arjevani2020ideal,
  title   = {IDEAL: inexact decentralized accelerated augmented Lagrangian method},
  author  = {Yossi Arjevani and Joan Bruna and Bugra Can and Mert Gürbüzbalaban and Stefanie Jegelka and Hongzhou Lin},
  year    = {2020},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  volume  = {33},
  pages   = {20648--20659},
  url     = {https://proceedings.neurips.cc/paper/2020/hash/ed77eab0b8ff85d0a6a8365df1846978-Abstract.html},
  note    = {arXiv:2006.06733},
}