English

DEED: A General Quantization Scheme for Communication Efficiency in Bits

Optimization and Control 2020-06-23 v1 Machine Learning

Abstract

In distributed optimization, a popular technique to reduce communication is quantization. In this paper, we provide a general analysis framework for inexact gradient descent that is applicable to quantization schemes. We also propose a quantization scheme Double Encoding and Error Diminishing (DEED). DEED can achieve small communication complexity in three settings: frequent-communication large-memory, frequent-communication small-memory, and infrequent-communication (e.g. federated learning). More specifically, in the frequent-communication large-memory setting, DEED can be easily combined with Nesterov's method, so that the total number of bits required is O~(κlog1/ϵ)\tilde{O}( \sqrt{\kappa} \log 1/\epsilon ), where O~\tilde{O} hides numerical constant and logκ\log \kappa factors. In the frequent-communication small-memory setting, DEED combined with SGD only requires O~(κlog1/ϵ)\tilde{O}( \kappa \log 1/\epsilon) number of bits in the interpolation regime. In the infrequent communication setting, DEED combined with Federated averaging requires a smaller total number of bits than Federated Averaging. All these algorithms converge at the same rate as their non-quantized versions, while using a smaller number of bits.

Keywords

Cite

@article{arxiv.2006.11401,
  title  = {DEED: A General Quantization Scheme for Communication Efficiency in Bits},
  author = {Tian Ye and Peijun Xiao and Ruoyu Sun},
  journal= {arXiv preprint arXiv:2006.11401},
  year   = {2020}
}
R2 v1 2026-06-23T16:28:41.585Z