IEEE - Aerospace and Electronic Systems - July 2021 - 64

Artificial Intelligence and Data Fusion at the Edge
Figure 1.
Framework for data fusion and AI at edge.
AI is first performed at the lowest tier of the network edge
comprising of senor nodes and IoT devices. These edge-ofnetwork
sensor/IoT devices can be sensor nodes sensing particular
features (e.g., temperature, humidity, and pressure),
smart phones, smart vehicles, video and imaging devices
(e.g., cameras including night vision imaging cameras and
thermal imaging cameras), and even airborne vehicles, such
as unmanned aerial vehicles (UAVs), equipped with different
sensors. These edge-of-network IoT devices typically
possess limited computation and communication capabilities.
Given the increasing proliferation ofAI-driven applications,
modern IoT devices can be outfitted with various ML
accelerators to speed up the execution of ML algorithms in
an energy-efficient manner as depicted in Figure 1. For
example, IoT nodes equipped with cameras can execute
hardware-accelerated lightweight CNN algorithms, such as
MobileNet [5], to perform object classification.
Although modern and futuristic IoT devices can be
equipped with AI accelerators [6] as depicted in Figure 1,
many of the contemporary edge-of-network sensors/IoT
devices have limited computational capabilities that are not
adequate to carry out complex AI tasks. In order to boost
the computational and AI capabilities at the network edge,
edge servers are installed at the base stations in the edge
computing paradigm. Edge servers possess much more
resources and computational capability than the edge-of64
network
sensors/IoT devices. Since edge servers receive
data from many edge-of-network devices with diverse
application requirements, performance requirements from
edge server are also diverse. When AI/ML processing with
stringent latency requirements (e.g., for hard real-time systems)
is required from edge servers, edge servers are
designed with high-performance stationary computing servers
with stable power source and high bandwidth network
connections. For mobile edge processing, edge servers can
be power/energy constrained. For example, UAVs, which
are typically powered by batteries, can be used as mobile
edge servers or devices. These mobile edge devices often
employ AI accelerators for fast and energy-efficient AI/ML
processing [7], [8]. Moreover, edge servers or fog nodes are
more amenable for integration ofML accelerators, such as
CNN, RNN, andMLP accelerators as depicted in Figure 1.
Each edge server manages a cluster ofedge-of-network
sensors/IoT devices in its vicinity. Edge servers provide
applications, content, context, services, and storage to edgeof-network
IoT devices. For example, edge servers can
assist with complex image processing and storage tasks for
imaging data acquired from camera sensors at IoT devices.
Since edge servers are in close proximity to edge-of-network
IoT devices, the edge servers can perform the offloaded
tasks from IoT devices with much lower latency and much
lower communication overhead as compared to the cloud.
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
