IEEE - Aerospace and Electronic Systems - July 2021 - 73

Munir et al.
processing [30]. Many edge devices and systems (e.g., surveillance
systems, autonomous vehicles, and robots) have
stringent deadline requirements (in the order of microseconds
to milliseconds) and missing those deadlines can
result in catastrophes. According to Steve Roddy, the VP
of Special Project in Arm's Machine Learning Group [2]:
" Applications that people will engage within real-world
products such as controlling home devices or providing
driver assistance in a car, all of those applications are running
on the edge and many will require real-time
responses. Any delay from bouncing information to the
cloud and back could be a problem. " Sri Chandrasekaran,
Senior Director of IEEE Standards Association, has also
emphasized the importance of speed/latency for AI inference
2: " We can't overlook the importance of latency. AI
at the edge will allow for faster data transfer, which will,
in turn, benefit the many industries AI touches, especially
industrial IoT and automotive. These industries benefit
from AI at the edge because the machines and automobiles
must be able to understand many different aspects at once.
Sending data to the cloud and back is not only inefficient,
but it is also less secure and much slower, ultimately leading
to a decrease in productivity and reliability. " Research
results also verify that data fusion and AI at the edge provides
much faster response as compared to sending data to
the cloud for fusion and inference [31].
Energy Efficiency. Data fusion and AI at the edge is
much more energy efficient than the data fusion and inference
at the cloud because it takes a large amount of power
to send data over the air, whereas it takes orders of magnitude
lesser power to do computations on the device when
the data is available on-device. Since many of the edge
devices are battery powered with no energy harvesting
system (e.g., solar and thermal), sending data to the cloud
for fusion and inference and receiving the results back
will expeditiously deplete these devices of the battery
power.
Precision. AI precision or accuracy refers to the ratio
of the number of input samples that get correct prediction
to the total number of input samples [30]. Many edge
applications, such as autonomous driving and face authentication,
require ultrahigh AI accuracy. Although increase
in the number and type of sensors assist in covering large
areas, the growing number of sensors have often resulted
in an increase in false alarm rates and have compounded
the target acquisition process in case of surveillance applications
due to the fact that sensors can provide inaccurate,
incomplete, or inconsistent data. Data fusion at the edge
can help filter the outliers and malicious readings, which
results in an improved AI/ML model with better precision
than a model that is trained on outliers and malicious data.
Furthermore, in many edge applications, AI inference
accuracy is also affected by the speed at which an application
needs to process the input data. For a video analytics
application under a fast feeding rate, some input samples
JULY 2021
may be skipped due to limited resources of edge devices.
In such conditions, data fusion can help improve the accuracy
by fusing a few adjacent frames of video and presenting
the fused frames to the AI model for inference. The
number of frames to be fused will depend on the application
and available resources of an edge device. The fused
frame will be able to capture the salient information in the
video frames and will provide better prediction accuracy
instead of skipping the video frames if an edge device
could not handle the feed rate.
Security. Security is another advantage that is provided
by data fusion and AI at the edge. In case of fusion and AI
at the cloud, data need to travel from edge devices to the
cloud and from the cloud to edge devices hundreds ofmiles
over multiple channels including wireless channels and
Internet thus exposing the data en route to attackers. The
data sent to and received from the cloud can be compromised
over the wireless channels, wired channels, intermediate
routers, or even the cloud computers itself. Data
fusion and AI at the edge devices minimizes the data transfer
and thus alleviates the security issues associated with
data transfer over multiple channels.
Privacy. Privacy preservation is another advantage of
data fusion and AI at the edge. Often the data requiring
inference is private and contains sensitive information
(e.g., medical records, personal photos, financial reports,
and target information in defense applications). Sending
this private data to the cloud for AI provides no privacy
guarantees to the user. Microsoft Research has proposed a
homomorphic encryption based solution referred to as
CryptoNets [32] that permits a data owner to send their
data in encrypted form to the cloud service for inference.
Since the cloud does not have access to the encryption
key, the user data remain confidential. In CryptoNets, the
cloud service is able to provide inference on the encrypted
data and return the results to the user in an encrypted
form. However, the overhead of homomorphic encryption
limits the applicability of this solution on resource-constrained
edge devices. Data fusion and AI at the edge enables
inference at the edge and thus alleviates the privacy
issues associated with sending data to the cloud.
Cost. Data fusion and AI at the edge also provide cost
advantages. Due to advancements in semiconductor technology,
the cost of system-on-chips (SoCs) is decreasing
with increasing capability to perform fusion and AI on
these SoCs. These edge SoCs cost much lesser than building
the apparatus and infrastructure required to perform
fusion and inference in the cloud.
Scalability. The number of edge devices is continuously
on the rise approaching 100 billion in near future
and producing hundreds of zettabytes of data. If each edge
device has to send all the data back to the cloud data center
for fusion and AI inference, it will put an enormous
pressure on the network bandwidth likely causing the network
to collapse (denial of service) as well as will require
IEEE A&E SYSTEMS MAGAZINE
73
IEEE - Aerospace and Electronic Systems - July 2021

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