LUTNet An energy-efficient AI network of elementary lookup tables

Goal

The project partners develop an innovative approach for energy-efficient artificial intelligence (AI) based on artificial neural networks in Field Programmable Gate Arrays (FPGAs). The aim of the project is to develop a fundamentally new AI structure based on classical CNN architectures for an FPGA using the example of 1-dimensional stream processing (artifact recognition in ECG). With the help of machine learning methods, the signal is evaluated and the result is classified. A given detection rate must be achieved and the energy requirement minimized as far as possible.

Persons
Publications
An Embedded CNN Implementation for On-Device ECG Analysis

Alwyn Burger, Chao Qian, Gregor Schiele, Domenik Helms; 2020 IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops); 0March / 2020

Modelling neural networks as SDFG representations for energy efficient hardware

Mark Kettner and Behnam Razi Perjikolaei and Wolfgang Nebel; 009 / 2020

FPGA based low latency, low power stream processing AI

Domenik Helms, Mark Kettner, Behnam Razi Perjikolaei, Lukas Einhaus, Christopher Ringhofer, Gregor Schiele; European Workshop on On-Board Data Processing; 0June / 2021

Partners
Universität Duisburg-Essen - Fachgebiet Eingebettete Systeme der Informatik
www.uni-due.de/es/

Duration

Start: 01.10.2019
End: 30.09.2020

Source of funding