English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

In scalable machine learning systems, model training is often parallelizedover multiple nodes that run without tight synchronization. Most analysisresults for the related asynchronous algorithms use an upper bound on theinformation delays in the system to determine learning rates. Not only are suchbounds hard to obtain in advance, but they also result in unnecessarily slowconvergence. In this paper, we show that it is possible to use learning ratesthat depend on the actual time-varying delays in the system. We develop generalconvergence results for delay-adaptive asynchronous iterations and specializethese to proximal incremental gradient descent and block-coordinate descentalgorithms. For each of these methods, we demonstrate how delays can bemeasured on-line, present delay-adaptive step-size policies, and illustratetheir theoretical and practical advantages over the state-of-the-art.

Delay-adaptive step-sizes for asynchronous learning