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

Today, artificial neural networks are the state of the art for solving avariety of complex tasks, especially in image classification. Sucharchitectures consist of a sequence of stacked layers with the aim ofextracting useful information and having it processed by a classifier to makeaccurate predictions. However, intermediate information within such models isoften left unused. In other cases, such as in edge computing contexts, thesearchitectures are divided into multiple partitions that are made functional byincluding early exits, i.e. intermediate classifiers, with the goal of reducingthe computational and temporal load without extremely compromising the accuracyof the classifications. In this paper, we present Anticipate, Ensemble andPrune (AEP), a new training technique based on weighted ensembles of earlyexits, which aims at exploiting the information in the structure of networks tomaximise their performance. Through a comprehensive set of experiments, we showhow the use of this approach can yield average accuracy improvements of up to15% over traditional training. In its hybrid-weighted configuration, AEP'sinternal pruning operation also allows reducing the number of parameters by upto 41%, lowering the number of multiplications and additions by 18% and thelatency time to make inference by 16%. By using AEP, it is also possible tolearn weights that allow early exits to achieve better accuracy values thanthose obtained from single-output reference models.

Anticipate, Ensemble and Prune: Improving Convolutional Neural Networks via Aggregated Early Exits