Skip to main navigation Skip to search Skip to main content

MobiRNN: Efficient recurrent neural network execution on mobile GPU

  • Stony Brook University

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

58 Scopus citations

Abstract

In this paper, we explore optimizations to run Recurrent Neural Network (RNN) models locally on mobile devices. RNN models are widely used for Natural Language Processing, Machine translation, and other tasks. However, existing mobile applications that use RNN models do so on the cloud. To address privacy and efficiency concerns, we show how RNN models can be run locally on mobile devices. Existing work on porting deep learning models to mobile devices focus on Convolution Neural Networks (CNNs) and cannot be applied directly to RNN models. In response, we present MobiRNN, a mobile-specific optimization framework that implements GPU offloading specifically for mobile GPUs. Evaluations using an RNN model for activity recognition shows that MobiRNN does significantly decrease the latency of running RNN models on phones.

Original languageEnglish
Title of host publicationEMDL 2017 - Proceedings of the 1st International Workshop on Deep Learning for Mobile Systems and Applications, co-located with MobiSys 2017
PublisherAssociation for Computing Machinery
Pages1-6
Number of pages6
ISBN (Electronic)9781450349628
DOIs
StatePublished - Jun 23 2017
Event1st International Workshop on Deep Learning for Mobile Systems and Applications, EMDL 2017, Co-located with MobiSys 2017 - Niagara Falls, United States
Duration: Jun 23 2017Jun 23 2017

Publication series

NameEMDL 2017 - Proceedings of the 1st International Workshop on Deep Learning for Mobile Systems and Applications, co-located with MobiSys 2017

Conference

Conference1st International Workshop on Deep Learning for Mobile Systems and Applications, EMDL 2017, Co-located with MobiSys 2017
Country/TerritoryUnited States
CityNiagara Falls
Period06/23/1706/23/17

Fingerprint

Dive into the research topics of 'MobiRNN: Efficient recurrent neural network execution on mobile GPU'. Together they form a unique fingerprint.

Cite this