IEEE - Aerospace and Electronic Systems - July 2021 - 63

Credit: Image licensed by Ingram Publishing
to provide more concise and consistent data to the AI both
at the training and inference phases. This article focuses
on AI at the edge and illustrates the benefits of integrating
data fusion with AI at the edge.
Our main contributions in this article are as follows.
(1) We propose a framework for AI and data fusion at
the edge.
(2) We provide a comparativ'e discussion of different
data fusion and AI architectures.
(3) We discuss multiple levels of fusion and different
types of AI, and how different types of AI relate to
and align with different levels of fusion.
(4) We highlight and elaborate the advantages of AI
and data fusion at the edge including latency,
energy efficiency, precision, security, privacy, cost,
scalability, and sustainability.
(5) We present experimental results for latency, energy
efficiency, and accuracy for AI and data fusion and
AI without data fusion to demonstrate the advantages
ofAI and data fusion at the edge.
The remainder of this article is organized as follows.
Section " Framework for Data Fusion and AI at the Edge "
presents a framework for data fusion and AI at the edge.
Section " Models and Architectures for Data Fusion and AI "
provides a comparative discussion of contemporary data
fusion and AI architectures. Section " Alignment of Multilevel
Fusion and AI " delineates multiple levels of fusion
and different types ofAI, and also provides a mapping ofdifferent
AI and ML types to different fusion levels. Advantages
of data fusion and AI at the edge are discussed in
section " Advantages of Data Fusion and AI at the Edge. "
Experimental results are presented in section " Experimental
Results. " Finally, section " Conclusions " concludes this
article.
FRAMEWORK FOR DATA FUSION AND AI AT THE EDGE
In this section, a high-level framework for data fusion and
AI/ML processing at the edge is developed. To explain the
JULY 2021
proposed framework, we provide a brief overview of AI
andML and elaborate on how data fusion enhances AI. Contemporary
AI is dominated by ML. Multitude ofML methods,
which can be categorized under supervised learning,
semisupervised learning, unsupervised learning, and reinforcement
learning (RL), help provide edge intelligence.
Deploying an ML model requires first training the model
and then using the trained model to perform inference. Since
training is a more resource-intensive and time-consuming
process; in the edge AI model, training is typically done on
the cloud and inference is performed at the edge. Although
AI training can be performed at edge [4], given the limited
resources ofmany edge devices, AI training will likely continue
on powerful cloud-based computers, whereas inference
will be performed at edge devices. Among the existing
ML methods, deep learning provides magnificent performance
on various tasks. For example, convolutional neural
networks (CNN) have been utilized for image classification,
object detection, and recognition. Recurrent neural networks
(RNNs) find applications in natural language processing and
multitarget tracking. Deep reinforcement learning (DRL)
can be leveraged for determining optimal policy/strategy for
accomplishing various tasks, such as trajectory optimization
of autonomous vehicles. Regardless of any specific AI/ML
methods, edge computing provides a means for performance-
and energy-efficient execution ofAI/ML algorithms.
Figure 1 depicts our proposed framework for data
fusion and AI at the edge. In the proposed framework,
data fusion, AI/ML processing, analysis, and decisionmaking
are done at the following three levels:
(1) edge-of-network sensor/IoT nodes;
(2) edge servers or fog nodes;
(3) cloud servers.
AI at both the edge-of-network sensor/IoT devices and
edge servers (fog nodes) is referred to as edge AI. Edge AI
enables computations near the edge of the network and helps
in reducing the communication burden on the core network.
We distinguish the edge AI done at senor/IoT node level and
edge server (fog node) level to highlight the difference in
compute capability of the two. In the proposed framework,
IEEE A&E SYSTEMS MAGAZINE
63
IEEE - Aerospace and Electronic Systems - July 2021

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