Machine learning

Deep Learning Optimiser

This use case develops a deep learning optimiser that uses OmpSs to reduce the energy consumption of deep learning inference significantly. The optimiser makes use of OmpSs to increase the computational architectures supported by the optimiser to CPU, GPU and FPGA. The focus and aim is to facilitate the deployment of deep learning based AI on embedded systems for users in an energy- and cost-efficient manner.

LEGaTO enabled Machine Intelligence Sweden, one of its partners, to develop a Deep Learning Optimiser that has been spun out as its own company, EmbeDL. Furthermore, OmpSs@FPGA lowers the entry barrier for FPGA-development and has thus greatly improved the efficiency of the FPGA-development cycle.

Machine Learning on the LEGaTO system

 

Components

OmpSs@FPGA has been used to develop an FPGA implementation for deployment of deep neural networks.

LEGaTO components are available here: https://legato-project.eu/software/components

 

Scientific publications

 

Events

 

News

Medium Date Article
LEGaTO website April 2020 LV-EmbeDL: Transferring LEGaTO-developed Techniques to Industry
HiPEAC info January 2020 Bright sparks - Tackling the energy challenge in computing systems
LEGaTO website December 2019 LEGaTO partner Hans Salomonsson wins HiPEAC Technology Transfer Award
LEGaTO website March 2019 Reducing energy while maintaining reliability: FPGAs and neural networks
LEGaTO website February 2018 New project to plug the software-stack support gap for energy-efficient computing