Yet Another Challenge for the Automotive Software: Deep Learning

In automotive software, the deep learning-based systems, with their peculiar features, are playing an increasingly pervasive role. As a result, inside the automotive software engineering community, an awareness of the need of integrating the deep learning development approach with the "traditio...

Full description

Saved in:
Bibliographic Details
Published in:IEEE software p. 1
Main Authors: Falcini, Fabio, Lami, Giuseppe, Mitidieri, Alessandra
Format: Journal Article
Language:English
Published: IEEE 16.06.2017
Subjects:
Automobiles
Automotive engineering
Biological neural networks
D Software/Software Engineering
D.2 Software Engineering
D.2.0 General
D.2.0.d Standards
D.2.18 Software Engineering Process
D.2.18.a Life cycle
I Computing Methodologies
I.2 Artificial Intelligence
I.2.10 Vision and Scene Understanding
Machine learning
Software
Software engineering
Training
ISSN:0740-7459
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In automotive software, the deep learning-based systems, with their peculiar features, are playing an increasingly pervasive role. As a result, inside the automotive software engineering community, an awareness of the need of integrating the deep learning development approach with the "traditional" ones is growing, at technical, methodological and cultural levels. In particular, the innovative and data-intensive phase of deep neural network training, by means of ad-hoc training data, is pivotal in the software development of vehicle functions that rely on such a technology paradigm and thus necessitates a special focus. A development lifecycle fitting deep learning needs is introduced and an improvement initiative, based on Automotive SPICE, for an effective adoption of DNN in the automotive software applications is presented.
ISSN:0740-7459
DOI:10.1109/MS.2017.265101102