3PC: Three Point Compressors for Communication-Efficient Distributed Training and a Better Theory for Lazy Aggregation

Peter Richtarik, Igor Sokolov, Ilyas Fatkhullin, Elnur Gasanov, Zhize Li, Eduard Gorbunov

Research output: Chapter in Book/Report/Conference proceedingConference contribution


We propose and study a new class of gradient compressors for communication-efficient training—three point compressors (3PC)—as well as efficient distributed nonconvex optimization algorithms that can take advantage of them. Unlike most established approaches, which rely on a static compressor choice (e.g., TopK), our class allows the compressors to evolve throughout the training process, with the aim of improving the theoretical communication complexity and practical efficiency of the underlying methods. We show that our general approach can recover the recently proposed state-of-the-art error feedback mechanism EF21 (Richtárik et al, 2021) and its theoretical properties as a special case, but also leads to a number of new efficient methods. Notably, our approach allows us to improve upon the state-of-the-art in the algorithmic and theoretical foundations of the lazy aggregation literature (Liu et al, 2017; Lan et al, 2017). As a by-product that may be of independent interest, we provide a new and fundamental link between the lazy aggregation and error feedback literature. A special feature of our work is that we do not require the compressors to be unbiased.
Original languageEnglish (US)
Title of host publication39th International Conference on Machine Learning (ICML)
StatePublished - 2022

Bibliographical note

KAUST Repository Item: Exported on 2023-05-23
Acknowledgements: The work of P. Richtarik, I. Sokolov, E. Gasanov and Z. Li was supported by the KAUST baseline research funding scheme. The work of E. Gorbunov was supported by Russian Science Foundation (project No. 21-71-30005). The work of I. Fatkhullin was supported by ETH AI Center doctoral fellowship.


Dive into the research topics of '3PC: Three Point Compressors for Communication-Efficient Distributed Training and a Better Theory for Lazy Aggregation'. Together they form a unique fingerprint.

Cite this