IEEE - Aerospace and Electronic Systems - July 2021 - 80

Feature Article:
DOI. No. 10.1109/MAES.2020.3049030
Machine Learning/Artificial Intelligence for Sensor
Data Fusion-Opportunities and Challenges
Erik Blasch , Air Force Office of Scientific Research, Arlington, VA 22203
USA
Tien Pham
Chee-Yee Chong , Independent Consultant, San Jose, CA 94024 USA
Wolfgang Koch
, Army Research Laboratory, Adelphi, MD 20783 USA
, Fraunhofer FKIE, D-53343 Wachtberg, Germany;
University of Bonn, 53113 Bonn, Germany
Henry Leung , University of Calgary, Calgary, AB T2N 1N4, Canada
Dave Braines, IBM, Hants SO21 2JN, U.K., Cardiff University, Cardiff
CF10 3AT, U.K.
Tarek Abdelzaher
61801 USA
, University of Illinois at Urbana-Champaign, Urbana, IL
INTRODUCTION
Machine learning (ML) methods are fundamental to the philosophy
of artificial intelligence (AI) approaches. With the
advent of the computer in the 1960s, both ML and sensor
datafusion (SDF) took prominence in the engineering community
with such developments of the Kalman filter, Bayesian
decision trees, expert systems, and probabilistic reasoning
Author's current addresses: Erik Blasch, Air Force
Office of Scientific Research, Arlington, VA 22203 USA
(e-mail: erik.blasch@gmail.com). Tien Pham, Army
Research Laboratory, Adelphi, MD 20783 USA (e-mail:
tien.pham1.civ@mail.mil). Chee-Yee Chong, Independent
Consultant, San Jose, CA 94024 USA (e-mail:
chee.y.chong@gmail.com). Wolfgang Koch, Fraunhofer
FKIE, D-53343 Wachtberg, Germany, and also University
of Bonn, 53113 Bonn, Germany (e-mail: wolfgang.
koch@fkie.fraunhofer.de). Henry Leung, University of
Calgary, Calgary, AB T2N 1N4, Canada (e-mail: leungh@ucalgary.ca).
Dave Braines, IBM United Kingdom
Limited, Hants SO21 2JN, U.K., and also Cardiff University,
Cardiff CF10 3AT, U.K. (e-mail: dave_braines@uk.ibm.com).
Tarek Abdelzaher, University of
Illinois at Urbana-Champaign, Urbana, IL 61801 USA
(e-mail: zaher@illinois.edu).
Manuscript received October 31, 2019, revised October
1, 2020; accepted October 4, 2020, and ready for
publication December 31, 2020.
Review handled by Peter Willett.
0885-8985/21/$26.00 ß U.S. Government work not
protected by U.S. copyright
80
[1]. While SDF expanded to multimodal analysis and informationfusioninthe
1990s[2],trendsinthe 21st century
include big data, uncertainty reduction, user coordination [3],
context reasoning [4], and multi-modal analysis, but it is only
in the last decade that a radical change in the AI/ML methods
surfaced with most notably deep learning (DL) [5], graphical
processing units, and reusable software.
AI/ML and data analytics pose both challenges and
opportunities for SDF. The challenges arise because they
appear to address the same problems as information fusion
(see Figure 1); however, SDF seeks methods for situation
and impact assessment, while AI/ML focuses on big data
analytics supporting object assessment classification.
These challenges can also be opportunities as AI/ML
spawned new research in deploying physics-based and
human-derived information fusion (PHIF), learning about
context, tracking with graph fusion, coordinating Internet
of Things (IoT) security, as well as facilitating dynamic
network analysis for multi-domain operations (MDO).
The 2019 panel members (i.e., the authors of this paper)
focused on AI/ML issues relating to SDF, illustrating potential
approaches and addressing challenges as summarized in
Table 1. Each panelist focused on the following questions.
What is the SDF problem?
Which part of the SDF problem needs machine
learning and why?
Where are opportunities to bring together AI/ML
and SDF?
When do challenges arise using machine learning
for SDF?
IEEE A&E SYSTEMS MAGAZINE
JULY 2021
IEEE - Aerospace and Electronic Systems - July 2021

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - July 2021
