IEEE - Aerospace and Electronic Systems - July 2021 - 82

Machine Learning/Artificial Intelligence for Sensor Data Fusion-Opportunities and Challenges
Table 1.
Summary of the Panel Ideas and Issues of AI/ML and Information Fusion
SDF Problem Use of ML
Abdelzaher Physics-based
and humanderived
IF
Blasch
Baines
Chong
Koch
Leung
Pham
User
augmentation
Contextual
support
Data
Association
Perceiving and
action
Big data
analysis of
edge-sensing
Highdimensional
learning
Interpretable
Analysis
Training
from
Data
Data processing
for object
assessment
Image fusion Change
Detection
Multi-domain
coordination
Rapidly learn,
adapt, and
reason to act
management and not on " mission management " policies of
AI/ML. Finally, the green colors represent a good overlap
of AI and SDF; while blue is somewhat, and red did not
have an intersection in the panel discussions. The area on
the upper right of Figure 2 shows that future research is
needed in AI/ML algorithms to support situation and
impact assessment as well as process refinement. The rest
ofthe article presents key discussions from the panelists.
HISTORY OF INFORMATION FUSION AND AI/ML
Chee Chong highlighted the basis of control theory since
the 1940s that spawned interest in SDF techniques. In the
1980s, few companies were doing information fusion as
there was a lack of computer processing methods. The
2000s saw industrial companies exploring big data computing
as the power to process massive amounts of data
existed (see Table 2). The challenge with many current AI
Challenges of ML
Unlabeled data
Heterogeneous
analysis
Determining the
various users
Relevant Models
Combining data and
models for situation
assessment
Real-time labeling
Inference in sparse and
congested areas
New Research
Coordination of DL through
multiple AI/ML networks
Model-based methods to
address the unknown
Explainable results
Context-based AI
Need common terms for
ethical, social, and usable
deployment
Joint multimodal image
data fusion
Learning at the edge
approaches is that companies are trying to package products
without doing the research, testing and evaluation of
the results, or determining the ethical and social implications
of the developments [6].
EXAMPLE: TARGET TRACKING
The SDF community has existed for 40 years with its roots
in the 1980s. The dawn of information fusion began with
data fusion with such groups as Garvey at SRII doing AI,
Reid at Lockheed working multiple hypothesis tracking,
Bar-shalom at Systems Control developing thejoint probabilistic
data association filter, and Chong at Advanced
Table 2.
Changes in Computation Power
Performance
1980
Machine
Speed
Operations
Memory
Figure 2.
Intersection ofAI/ML issues with that of sensor-data fusion.
82
Vax 11/750
Minicomputer
3 MHz
120K FLOPS
2 MB RAM
Communications 1200 baud dialup
Modem
IEEE A&E SYSTEMS MAGAZINE
2019
IPhone
XS
2.49 GHz
409
GFLOPS
4 GB
RAM
1GB
network
JULY 2021
IEEE - Aerospace and Electronic Systems - July 2021

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