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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

A pruning-based AutoML framework for run-time reconfigurability, namely RT3,is proposed in this work. This enables Transformer-based large Natural LanguageProcessing (NLP) models to be efficiently executed on resource-constrainedmobile devices and reconfigured (i.e., switching models for dynamic hardwareconditions) at run-time. Such reconfigurability is the key to save energy forbattery-powered mobile devices, which widely use dynamic voltage and frequencyscaling (DVFS) technique for hardware reconfiguration to prolong battery life.In this work, we creatively explore a hybrid block-structured pruning (BP) andpattern pruning (PP) for Transformer-based models and first attempt to combinehardware and software reconfiguration to maximally save energy forbattery-powered mobile devices. Specifically, RT3 integrates two-leveloptimizations: First, it utilizes an efficient BP as the first-step compressionfor resource-constrained mobile devices; then, RT3 heuristically generates ashrunken search space based on the first level optimization and searchesmultiple pattern sets with diverse sparsity for PP via reinforcement learningto support lightweight software reconfiguration, which corresponds to availablefrequency levels of DVFS (i.e., hardware reconfiguration). At run-time, RT3 canswitch the lightweight pattern sets within 45ms to guarantee the requiredreal-time constraint at different frequency levels. Results further show thatRT3 can prolong battery life over 4x improvement with less than 1% accuracyloss for Transformer and 1.5% score decrease for DistilBERT.

Dancing along Battery: Enabling Transformer with Run-time Reconfigurability on Mobile Devices