Robust power optimization for device-to-device communication in a multi-cell network under partial CSI

Ramezani-Kebrya, Ali; Dong, Min; Liang, Ben; Boudreau, Gary; Seyedmehdi, S. Hossein

doi:10.1109/icc.2017.7996669

Cited by 12 publications

(1 citation statement)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…There are two main directions in the literature for tackling the communication cost issue in FL. The first approach consists of algorithms that aim to reduce the number of transmitted bits by applying a carefully chosen gradient compression scheme, such as quantization [2,5,31,16,38,40], sparsification [1,27,3,49,48,32], or other more sophisticated strategies [19,46,52,47,6,10]. The second approach-one that we investigate in this paperinstead focuses on increasing the total amount of local computation in between the communication rounds in the hope that this will reduce the total number of communication rounds needed to build a model of sufficient quality [43,55,39,24,37].…”

Section: Introductionmentioning

confidence: 99%

Local SGD: Unified Theory and New Efficient Methods

Gorbunov,

Hanzely,

Richtárik

2020

Preprint

View full text Add to dashboard Cite

We present a unified framework for analyzing local SGD methods in the convex and strongly convex regimes for distributed/federated training of supervised machine learning models. We recover several known methods as a special case of our general framework, including Local-SGD/FedAvg, SCAFFOLD, and several variants of SGD not originally designed for federated learning. Our framework covers both the identical and heterogeneous data settings, supports both random and deterministic number of local steps, and can work with a wide array of local stochastic gradient estimators, including shifted estimators which are able to adjust the fixed points of local iterations for faster convergence. As an application of our framework, we develop multiple novel FL optimizers which are superior to existing methods. In particular, we develop the first linearly converging local SGD method which does not require any data homogeneity or other strong assumptions.

show abstract