Acceleration for Compressed Gradient Descent in Distributed Optimization
Jul 12, 2020
Speakers
Dmitry Kovalev
Speaker · 0 followers
Peter Richtarik
Speaker · 0 followers
About
Due to the high communication cost in distributed and federated learning problems, methods relying on sparsification or quantization of communicated messages are becoming increasingly popular. While in other contexts the best performing gradient-type methods invariably rely on some form of acceleration to reduce the number of iterations, there are no methods which combine the benefits of both gradient compression and acceleration. In this paper, we remedy this situation and propose the first accelerated compressed gradient descent (ACGD) methods. In the single machine regime, we prove that ACGD enjoys the rate O((1+ω)√(L/μ)log1/ϵ) for μ-strongly convex problems and O((1+ω)√(L/ϵ)) for convex problems, respectively, where L is the smoothness constant and ω is the variance parameter of an unbiased compression operator. Our results improve upon the existing non-accelerated rates O((1+ω)L/μlog1/ϵ) and O((1+ω)L/ϵ), respectively, and recover the best known rates of accelerated gradient descent as a special case when no compression (ω=0) is applied. We further propose a distributed variant of ACGD and establish the rate Õ(ω+√(L/μ) +√((ω/n+√(ω/n))ω L/μ)), where n is the number of machines and Õ hides the logarithmic factor log1/ϵ . This improves upon the previous best result Õ(ω + L/μ+ω L/nμ) achieved by the DIANA method. Finally, we conduct several experiments on real-world datasets which corroborate our theoretical results and confirm the practical superiority of our methods.Due to the high communication cost in distributed and federated learning problems, methods relying on sparsification or quantization of communicated messages are becoming increasingly popular. While in other contexts the best performing gradient-type methods invariably rely on some form of acceleration to reduce the number of iterations, there are no methods which combine the benefits of both gradient compression and acceleration. In this paper, we remedy this situation and propose the first acc…
Categories
Marketing & Business Strategy
Category · 1.4k presentations
About ICML 2020
The International Conference on Machine Learning (ICML) is the premier gathering of professionals dedicated to the advancement of the branch of artificial intelligence known as machine learning. ICML is globally renowned for presenting and publishing cutting-edge research on all aspects of machine learning used in closely related areas like artificial intelligence, statistics and data science, as well as important application areas such as machine vision, computational biology, speech recognition, and robotics. ICML is one of the fastest growing artificial intelligence conferences in the world. Participants at ICML span a wide range of backgrounds, from academic and industrial researchers, to entrepreneurs and engineers, to graduate students and postdocs.
Like the format? Trust SlidesLive to capture your next event!
Professional recording and live streaming, delivered globally.
Sharing
Recommended Videos
Presentations on similar topic, category or speaker
Global Concavity and Optimization in a Class of Dynamic Discrete Choice Models
Watch later
Yiding Feng, …
Experimental approach to measuring algorithmic bias in computer vision
Watch later
Poster #14
Watch later
Intrinsic Reward Driven Imitation Learning via Generative Model
Watch later
Xingrui Yu, …
Good Subnetworks Provably Exist: Pruning via Greedy Forward Selection
Watch later
Mao Ye, …