Characterising Across-Stack Optimisations for Deep Convolutional Neural Networks

Jack Turner, Jose Cano Reyes, Valentin Radu, Elliot Crowley, Michael O'Boyle, Amos Storkey

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract / Description of output

Convolutional Neural Networks (CNNs) are extremely computationally demanding, presenting a large barrier to their deployment on resource-constrained devices. Since such systems are where some of their most useful applications lie (e.g. obstacle detection for mobile robots, vision-based medical assistive devices), significant bodies of work from both Machine Learning and system level perspectives have attempted to provide optimisations that will make CNNs available to edge devices. In this paper we unify the two viewpoints in a Deep Learning Inference Stack and take an across-stack approach by implementing and evaluating the most common neural network compression techniques (weight pruning, channel pruning and quantisation) and optimising their parallel execution with a range of programming approaches (OpenMP, OpenCL) and hardware architectures (CPU, GPU). We provide comprehensive Pareto curves to instruct trade-offs under constraints of accuracy, execution time, and memory space, with other results contradicting perceptions developed in isolation within own research space.
Original languageEnglish
Title of host publicationProceedings of the - Workload Characterization (IISWC), 2018 IEEE International Symposium on
Place of PublicationRaleigh, North Carolina, USA
PublisherInstitute of Electrical and Electronics Engineers (IEEE)
Number of pages10
ISBN (Electronic)978-1-5386-6780-4
ISBN (Print)978-1-5386-6781-1
Publication statusPublished - 13 Dec 2018
Event2018 IEEE International Symposium on Workload Characterization - Raleigh, United States
Duration: 30 Sept 20182 Oct 2018


Conference2018 IEEE International Symposium on Workload Characterization
Abbreviated titleIISWC 2018
Country/TerritoryUnited States
Internet address


Dive into the research topics of 'Characterising Across-Stack Optimisations for Deep Convolutional Neural Networks'. Together they form a unique fingerprint.

Cite this