IEEE Consumer Electronics Magazine - November/December 2021 - 107

the data at different end devices. In such a case,
FL is a viable solution that can be built different
ML models shared among end devices without
lacking privacy and other security functionality.
Thus, a secured and privacy-preserving ML
framework is one of the important objectives
that can be achieved through FL. Other than
this, it can be used for other purposes also, such
as to maximize the computing power at end devices
at cloud and edge computing. Based on
architecture, FL can be classified as vertical and
horizontal. Chapter 4 describes different horizontal
FL architectures (HFL), algorithm, and
their challenges. The Google GBoard is based on
HFL. Vertical Federated Learning is also known
as heterogeneous federated learning. VFL architecture
focuses on local data and ensures that
the model is built according to business requirements.
Federated AI Technology Enabler (FATE)
is platform developed by " WeBank, " which is
also a part of the Linux Foundation is the world's
first open-source FL based on VFL. In chapter 5,
different VHL architectures, algorithms, and
challenges are discussed.
In many applications, there are not enough
shared features to determine whether to build a
HFL or VFL model. In such scenarios, transfer
learning techniques can be used to build a model
known as federated transfer learning (FTL). FTL is
the most suitable approach for cross-domain
knowledge transfer. FTL is classified as instance
based, feature based, and model based. These
frameworks are discussed in chapter 6. Framework
to overcome few existing challenges in FL is discussed
in Chapter 7. The mechanism such as
the reversed-auction based approach, FL incentivizer
(FLI) payoff-sharing scheme, and evaluating
owner's data are discussed in this chapter. Few FL
model for different applications is discussed in
chapter 8-10. It is learnt that a powerful object
detectionmodel can be built by the helpof FL. Federated
learning for natural language processing
(FL-NLP) is another emerging application domain
to protect the privacy of users. In many applications
such as blockchain, healthcare, edge computing,
etc., where data cannot be combined for
MLmodel due to different reasons. Insuch applications,
FLmodel can be built without compromising
on privacy and security functionality.
In conclusion, we can say that the FL model
require ML approaches in many applications to
achieve privacy and other important parameters.
We expect FL can be the backbone of next-generation
machine learning that can satisfy many technological
needs for AI-based applications.
&REFERENCES
1. Q. Yang, Y. Liu, Y. Cheng, Y. Kang, T. Chen, and H.
Yu. " Federated learning, " Synth. Lectures Artif. Intell.
Mach. Learn., vol. 13, no. 3, pp. 1-207, 2019.
2. D. Puthal and S. P. Mohanty, " Cybersecurity issues in
AI, " IEEEConsum. Electron. Mag., vol. 10, no. 4,
pp. 33-35, Jul. 2021.
Niranjan K. Ray is an Associate Professor with the
School of Computer Engineering, KIIT Deemed to be
University, Bhubaneswar, India. Contact him at
rayniranjan@gmail.com.
Deepak Puthal is an Assistant Professor (in U.K.,
Lecturer) with the School of Computing, Newcastle
University, Newcastle upon Tyne, U.K. Contact him
at deepak.puthal@ieee.org.
Dhruva Ghai is a Professor with the Department of
Electronics and Communication Engineering, Oriental
University, Indore, India. Contact him at
dhruvaghai@orientaluniversity.in.
November/December 2021
107
IEEE Consumer Electronics Magazine - November/December 2021

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - November/December 2021
