J.

Electrical Systems 20-5s (2024): 165-171

1
M Katyayani "Federated Learning: Advancements,
2
Dr. Kumar Applications, and Future Directions for
Keshamoni Collaborative Machine Learning in
3
A. Sree Rama Distributed Environments"
Chandra Murthy
4
Dr. K. Usha Rani
5
Sreenivasulu Reddy L
6
Dr.Yaswanth Kumar
Alapati

Abstract: Federated Learning (FL) has become widely recognized as a feasible method for training machine learning models on
decentralized devices, ensuring the preservation of data privacy. This study offers an extensive overview of the latest progress in federated
learning methods, their applications, and the challenges they entail. We begin by introducing the concept of federated learning and its
significance in distributed environments. Next, we delve into a range of methodologies aimed at improving the effectiveness, scalability,
and confidentiality of federated learning. These encompass optimization algorithms, communication protocols, and mechanisms designed
to uphold privacy. Moreover, we investigate the broad spectrum of applications where federated learning finds utility, spanning healthcare,
the Internet of Things (IoT), and edge computing. This exploration illuminates tangible scenarios and advantages in real-world settings.
Additionally, we analyze the challenges and limitations inherent in federated learning, including communication overhead, non-IID data
distribution, and model heterogeneity. We review recent research efforts aimed at addressing these challenges, such as federated averaging
variants, adaptive client selection, and robust aggregation techniques. Finally, we outline future research directions and potential avenues
for the advancement of federated learning, emphasizing the need for standardized benchmarks, federated learning frameworks, and
interdisciplinary collaborations.
Keywords: Federated Learning, Machine Learning, Privacy Preservation, Decentralized Devices, Optimization Algorithms,
Communication Protocols, Healthcare Applications.

I. INTRODUCTION
In the past few years, Federated Learning (FL) has gained recognition as an innovative method in machine
learning, providing a new approach to address the issues related to data privacy and decentralization. Unlike
conventional centralized learning techniques that aggregate data into a central repository for training models,
federated learning allows model training directly on decentralized devices while preserving the privacy of raw
data. This paradigm shift has significant implications for various applications, particularly in distributed settings
where data privacy and regulatory compliance are paramount concerns [14].
Overview of Federated Learning: Federated Learning is a machine learning paradigm that enables model
training across a network of decentralized devices while keeping data local to each device. Instead of
centralizing data in a single location, FL distributes the learning process to individual devices, allowing them to
collaboratively learn a global model without sharing raw data. The distributed learning approach offers several
advantages, including enhanced privacy protection, reduced communication overhead, and scalability to large
datasets [12] distributed across diverse locations.

1ResearchScholar, Department of English, Koneru Lakshmaiah Education Foundation, Vaddeswaram, Guntur, Andhra Pradesh, India,
[email protected]
2Assistant Professor, Department of ECE, Vaagdevi Engineering College, Warangal. Telanagan, India, [email protected]
3Sr. Assistant Professor, Department of Computer Science and Engineering, Lakireddy Bali Reddy College of Engineering, NTR Dist.,
Mylavaram. Andhra Pradesh, [email protected]
4
Associate Professor, Department of English, Koneru Lakshmaiah Education Foundation, Vaddeswaram, Guntur, Andhra Pradesh 522502,
India. [email protected]
5Associate Professor, Department of Mathematics, School of Advanced Sciences Kalasalingam Academy of Research and Education,
Krishnankoil, Tamilnadu, India, [email protected]
6 Associate Professor, Department of Information Technology, RVR&JC College of Engineering, Guntur-19. [email protected]

Copyright © JES 2024 on-line : journal.esrgroups.org

165
 J. Electrical Systems 20-5s (2024): 165-171

Importance of FL in Distributed Settings: The importance of Federated Learning becomes particularly

evident in distributed settings characterized by a vast network of devices generating data continuously. In
domains like healthcare, the Internet of Things (IoT), and edge computing settings, where sensitive data
originates locally on devices such as smart phones, sensors, and edge servers, “conventional centralized methods
encounter notable obstacles concerning data privacy, security [3], and adherence to regulations. Federated
Learning addresses these challenges by allowing machine learning models to be trained directly on edge devices
without compromising data privacy, thereby enabling collaborative learning while preserving the confidentiality
of sensitive information.
Objectives of the Review Paper: The main goal of this review is to offer a thorough summary of recent
developments in Federated Learning methodologies, applications, and the challenges they entail. Specifically,
our focus is on:

• Explore the latest techniques proposed to enhance the efficiency, scalability, and privacy of Federated
Learning.

• Examine diverse applications of Federated Learning across domains such as healthcare, IoT, and edge
computing, highlighting real-world use cases and benefits.

• Analyze the challenges and limitations inherent in Federated Learning, including communication overhead,
non-IID data distribution, and model heterogeneity.

• Review recent research efforts aimed at addressing these challenges and outline future research directions
and potential avenues for the advancement of Federated Learning.
By fulfilling these objectives, we seek to contribute to a deeper understanding of Federated Learning [16] and its
implications for machine learning in distributed environments.
II. TECHNIQUES IN FEDERATED LEARNING
Federated Learning (FL) encompasses a range of techniques designed to enable collaborative model training
across decentralized devices while addressing challenges such as communication overhead, data privacy, and
model convergence. In this section, we discuss key techniques employed in Federated Learning, including
optimization algorithms, communication protocols [5], and privacy-preserving mechanisms.
1.1 Optimization Algorithms: Optimization algorithms play a crucial role in Federated Learning by
facilitating the training of machine learning models across distributed devices [2] while mitigating challenges
such as non-IID (non-identically distributed) data and device heterogeneity. Numerous optimization strategies
have been suggested to enhance the speed of convergence and efficacy of Federated Learning (FL) models [7].
These include:

• Federated Averaging: Federated Averaging (FedAvg) is a fundamental optimization algorithm in FL

that aggregates model updates from multiple devices to compute a global model. FedAvg aims to minimize the
discrepancy between local models and the global model [6] by iteratively updating model parameters using
weighted averaging.

• Federated Stochastic Gradient Descent (FSGD): FSGD extends traditional Stochastic Gradient Descent
(SGD) to the federated setting, allowing devices to compute local gradients using their local data and then
synchronize model updates with a central server. FSGD algorithms often incorporate techniques such as
momentum and adaptive learning rates to enhance convergence.

• Federated Learning with Differential Privacy (FLDP): FLDP integrates differential privacy
mechanisms into Federated Learning to ensure robust privacy protection while training machine learning models
on sensitive data. By adding noise to the model updates before aggregation, FLDP algorithms provide strong
privacy guarantees without compromising model utility.
1.2 Communication Protocols: Efficient communication protocols are essential for facilitating
communication between decentralized devices and coordinating the model training process in Federated
Learning. These protocols aim to minimize communication overhead while ensuring timely synchronization of
model updates. Some common communication protocols used in FL include:

• In the architecture of parameter servers, a central parameter server coordinates communication between
devices by storing and distributing model parameters. Devices compute local gradients using their data and then
communicate these gradients to the parameter server for aggregation.

166
 J. Electrical Systems 20-5s (2024): 165-171

• Peer-to-Peer (P2P) Communication: P2P communication protocols enable direct communication

between devices participating in Federated Learning, bypassing the need for a central server. P2P protocols
reduce communication latency and bandwidth consumption, making them suitable for FL applications in
resource-constrained environments.
1.3 Privacy-Preserving Mechanisms: Privacy-preserving mechanisms are integral to Federated Learning
to ensure that sensitive data remains protected throughout the model training process. These mechanisms
employ cryptographic techniques, differential privacy, and federated learning frameworks to safeguard user
privacy while enabling collaborative model training. Some prominent privacy-preserving mechanisms in FL
include:

• Secure Aggregation: Secure aggregation protocols enable devices to aggregate model updates without
revealing individual contributions [16], thereby preserving the privacy of local data. Secure multi-party
computation (MPC) and homomorphic encryption are commonly used techniques for achieving secure
aggregation in FL.

• Federated Learning with Homomorphic Encryption (FLHE): FLHE leverages homomorphic encryption
to perform computations on encrypted data [16] without decrypting it, enabling devices to collaborate on model
training without sharing raw data. FLHE ensures end-to-end data privacy while allowing model updates to be
aggregated securely.
Discussion of Techniques to Improve Efficiency and Privacy in FL: Efficiency and privacy are paramount
concerns in Federated Learning, and ongoing research efforts focus on developing techniques to enhance both
aspects simultaneously. Optimization algorithms such as Federated Averaging and Federated Learning with
Differential Privacy (FLDP) aim to improve model convergence while preserving user privacy. Communication
protocols such as Parameter Server Architecture and Peer-to-Peer (P2P) Communication optimize
communication overhead and latency in FL systems. Privacy-preserving mechanisms like Secure Aggregation
and Federated Learning with Homomorphic Encryption (FLHE) provide robust privacy guarantees while
enabling collaborative model training across decentralized devices. By leveraging these techniques, Federated
Learning can achieve scalable, privacy-preserving machine learning in distributed settings.
III. APPLICATIONS OF FEDERATED LEARNING
Federated Learning (FL) has gained significant traction across various domains due to its ability to train
machine learning models collaboratively on decentralized data sources while preserving data privacy. In this
section, we explore the diverse applications of Federated Learning in domains such as healthcare, Internet of
Things (IoT) [9], and edge computing.
1.4 Healthcare: In healthcare, Federated Learning enables the development of predictive models and
diagnostic tools while ensuring patient data privacy and regulatory compliance. FL facilitates collaborative
model training across multiple healthcare institutions without centralizing sensitive patient information.
Applications of Federated Learning in healthcare [10] include:

• Disease Prediction: FL allows healthcare providers to develop predictive models for diseases such as
cancer, diabetes, and cardiovascular disorders by aggregating data from diverse sources without sharing patient
records.

• Medical Imaging Analysis: FL techniques enable the development of image-based diagnostic tools for
medical imaging modalities such as MRI, CT scans, and X-rays. Federated Learning preserves patient privacy
while improving the accuracy of diagnostic models.

• Drug Discovery: FL enables pharmaceutical companies and research institutions to leverage distributed
data sources for drug discovery and development. Collaborative model training facilitates the identification of
novel drug candidates while protecting proprietary research data.
1.5 Internet of Things (IoT): The Internet of Things (IoT) encompasses a wide range of interconnected
devices, sensors, and systems that generate vast amounts of data. Federated Learning enables edge devices in
IoT networks to collaboratively train machine learning models without transmitting raw data to centralized
servers [13]. Applications of Federated Learning in IoT include:

• Predictive Maintenance: FL allows IoT devices to analyze sensor data in real-time and predict
equipment failures or maintenance needs without compromising data privacy. Collaborative model training
enhances predictive accuracy while reducing communication overhead.

167
 J. Electrical Systems 20-5s (2024): 165-171

• Environmental Monitoring: Federated Learning enables distributed sensors deployed in environmental

monitoring networks to collectively analyze data and detect patterns related to air quality, water pollution, and
climate change. FL preserves user privacy while improving the effectiveness of environmental monitoring
systems.

• Smart Agriculture: FL techniques empower agricultural IoT devices to optimize crop management
practices, monitor soil conditions, and predict crop yields based on localized environmental data. Collaborative
model training enhances the resilience and efficiency of smart agriculture systems.
1.6 Edge Computing: Edge computing involves processing data locally on edge devices or edge servers
[13], closer to the data source, to reduce latency and bandwidth usage. Federated Learning extends edge
computing capabilities by enabling collaborative model training on edge devices while maintaining data
privacy. Applications of Federated Learning in edge computing include:

• Real-time Anomaly Detection: FL enables edge devices such as surveillance cameras, industrial
sensors, and IoT devices to detect anomalies and security threats in real-time by collaboratively analyzing
streaming data. Federated Learning enhances anomaly detection accuracy while minimizing communication
latency.

• Personalized Recommendations: FL techniques allow edge devices such as smart phones, smart
watches, and IoT devices to generate personalized recommendations for users based on their preferences and
behavior patterns. Collaborative model training on edge devices ensures user privacy and data sovereignty.

• Autonomous Vehicles: Federated Learning enables autonomous vehicles to learn and adapt to diverse
driving conditions, traffic patterns, and road environments by collaboratively training machine learning models
on edge devices installed in vehicles. FL enhances the safety and performance of autonomous driving systems
while preserving passenger privacy.
In summary, Federated Learning offers a wide range of applications and use cases across domains such as
healthcare, Internet of Things (IoT), and edge computing, enabling collaborative model training on
decentralized data sources while preserving data privacy and security [11].
IV. CHALLENGES AND LIMITATIONS
Federated Learning (FL) presents several challenges and limitations that need to be addressed to ensure its
effectiveness and scalability in real-world applications. In this section, we discuss key challenges, including
communication overhead, non-IID (Non-Independently and Identically Distributed) data distribution [17], and
model heterogeneity, and analyze their impact on FL performance.
1.7 Communication Overhead: One of the primary challenges in Federated Learning is the
communication overhead associated with transmitting model updates between the central server and
participating edge devices or nodes. As FL involves iterative model training rounds, frequent communication of
model parameters incurs significant bandwidth and latency overhead. This communication overhead can lead to
slower convergence rates, increased training time, and higher energy consumption, particularly in resource-
constrained environments.
1.8 Non-IID Data Distribution: FL assumes that data samples across participating nodes are IID,
meaning they are independently and identically distributed. However, in practical scenarios, data distribution
among edge devices or nodes may exhibit significant heterogeneity, resulting in non-IID data distributions.
Non-IID data distribution can arise due to variations in data collection environments, device types, user
demographics, and other factors. This poses challenges for aggregating and reconciling model updates from
heterogeneous data sources, leading to suboptimal global model performance and convergence issues.
1.9 Model Heterogeneity: Model heterogeneity refers to the diversity of machine learning models,
architectures, and hyperparameters across participating edge devices or nodes in a Federated Learning setting. In
FL, edge devices may have different computational capabilities, storage capacities, and model architectures,
resulting in model heterogeneity. Model heterogeneity complicates the aggregation of local model updates at the
central server and may require techniques such as model compression, knowledge distillation, or adaptive
aggregation to reconcile disparate model parameters effectively.
Analysis of challenges in FL and their impact on performance

• Communication overhead significantly affects FL performance by increasing training time and

resource consumption. Mitigating communication overhead requires optimizing communication protocols,

168
 J. Electrical Systems 20-5s (2024): 165-171

implementing efficient compression techniques for model updates, and exploring federated learning techniques
that reduce communication frequency.

• Non-IID data distribution introduces bias and variance in the global model, leading to suboptimal
performance and convergence issues. Addressing non-IID data distribution involves strategies such as data
augmentation, adaptive federated learning algorithms, and personalized model updates to account for local data
characteristics.

• Model heterogeneity complicates the aggregation of model updates and may result in a loss of
information during the federated averaging process. To mitigate model heterogeneity, techniques such as model
distillation, meta-learning, and adaptive aggregation methods can be employed to harmonize model parameters
across heterogeneous edge devices while preserving model diversity.
In conclusion, addressing challenges such as communication overhead, non-IID data distribution, and model
heterogeneity is essential to enhance the performance and scalability of Federated Learning in distributed
settings. By developing efficient communication protocols, adapting algorithms to handle non-IID data, and
harmonizing heterogeneous models, FL can realize its full potential in various applications while preserving
data privacy and security.
V. MITIGATING CHALLENGES: RECENT ADVANCES
Addressing the challenges associated with Federated Learning (FL) requires innovative techniques and
methodologies to improve efficiency, robustness, and scalability. In this section, we discuss recent advances in
FL that aim to mitigate challenges such as communication overhead, non-IID data distribution, and model
heterogeneity.
1.10 Federated Averaging Variants: Federated Averaging (FedAvg) is a fundamental algorithm used in
FL for aggregating model updates from distributed clients. Recent advances have focused on enhancing FedAvg
and developing variants that improve convergence speed, communication efficiency, and robustness to non-IID
data. Variants of FedAvg include Federated Averaging with Local Adapting Steps (FedAvg-LAS), which allows
clients to adapt their learning rates locally based on data characteristics, and Federated Averaging with
Momentum (FedAvg-M), which incorporates momentum to stabilize training and accelerate convergence.
1.11 Adaptive Client Selection: Adaptive client selection mechanisms aim to address the challenge of non-
IID data distribution by dynamically selecting clients for participation in each FL round based on their data
relevance or model performance. Recent approaches leverage techniques such as reinforcement learning, meta-
learning, and Bayesian optimization to adaptively select clients with representative data distributions, thereby
improving the global model's performance and convergence rate.
1.12 Robust Aggregation Techniques: Robust aggregation techniques focus on mitigating the impact of
outliers, malicious clients, or noisy updates during the aggregation process in FL. Recent advances include
techniques such as trimmed mean aggregation, which discards extreme updates before aggregation, and
Byzantine-robust aggregation, which employs cryptographic methods to detect and exclude malicious clients'
contributions. Additionally, differential privacy mechanisms can be integrated into the aggregation process to
preserve privacy while aggregating model updates securely.
Review of recent techniques to overcome challenges in FL: Recent advances in Federated Learning have led
to the development of innovative techniques and methodologies aimed at mitigating challenges and improving
the efficiency, robustness, and scalability of FL algorithms. By enhancing Federated Averaging variants,
employing adaptive client selection mechanisms, and implementing robust aggregation techniques, researchers
have made significant strides in overcoming communication overhead, non-IID data distribution, and model
heterogeneity challenges in FL. These advancements pave the way for the widespread adoption of FL across
various domains, offering scalable and privacy-preserving solutions for distributed machine learning tasks [16]”.
VI. FUTURE DIRECTIONS AND PROSPECTS
As Federated Learning (FL) continues to evolve, several future directions and prospects emerge, offering
opportunities for advancement and innovation. In this section, we outline potential areas of development that
could shape the future of FL.
Standardized Benchmarks and Evaluation Metrics: Establishing standardized benchmarks and evaluation
metrics is crucial for comparing the performance of FL algorithms across different datasets and settings. Future
research should focus on developing comprehensive benchmark datasets and evaluation protocols that capture
the diversity of FL applications while considering factors such as data heterogeneity, privacy constraints, and

169
 J. Electrical Systems 20-5s (2024): 165-171

communication overhead. Standardized benchmarks will facilitate fair comparisons between FL methods and
enable researchers to identify promising techniques more effectively.
Federated Learning Frameworks: The development of open-source FL frameworks is essential for fostering
collaboration and accelerating research in the field. Future efforts should prioritize the creation of user-friendly
FL platforms that provide modular implementations of FL algorithms, customizable privacy-preserving
mechanisms, and robust communication protocols. These frameworks should support a wide range of machine
learning tasks and enable seamless integration with existing deep learning libraries, facilitating experimentation
and deployment in real-world scenarios.
Interdisciplinary Collaborations: Collaborations between researchers from diverse disciplines, including
computer science, statistics, privacy, and domain-specific domains, are essential for advancing FL research and
addressing complex challenges. Future directions in FL should encourage interdisciplinary collaborations to
leverage domain knowledge, explore novel methodologies, and develop tailored solutions for specific
application domains. By fostering cross-disciplinary dialogue and collaboration, researchers can unlock new
insights and accelerate the development of innovative FL techniques.
Ethical Considerations and Societal Impacts: As FL technologies become more prevalent, it is imperative to
address ethical considerations and societal impacts associated with their deployment. Future research should
prioritize the development of ethical guidelines and best practices for FL, ensuring fairness, transparency, and
accountability in algorithmic decision-making. Moreover, researchers should actively engage with stakeholders,
including policymakers, industry partners, and affected communities, to assess the potential social, economic,
and ethical implications of FL applications. By integrating ethical considerations into the design and
implementation of FL systems, researchers can promote responsible innovation and mitigate unintended
consequences.
In summary, future directions in Federated Learning should focus on establishing standardized benchmarks,
developing user-friendly frameworks, fostering interdisciplinary collaborations, and addressing ethical
considerations and societal impacts. By pursuing these avenues of research, the FL community can advance the
state-of-the-art, promote responsible deployment, and realize the full potential of FL in diverse application
domains.
VII. CONCLUSION
In this review paper, we provided a comprehensive overview of Federated Learning (FL), an emerging paradigm
for collaborative machine learning in distributed settings. We discussed the importance of FL in scenarios where
data cannot be centralized due to privacy concerns, communication constraints, or regulatory requirements. Our
exploration covered various aspects of FL, including optimization algorithms, communication protocols,
privacy-preserving mechanisms, applications across different domains, challenges, recent advances, and future
directions.
Key findings from our review include:

• Techniques in Federated Learning: We examined optimization algorithms, communication

protocols, and privacy-preserving mechanisms employed in FL systems to improve efficiency and protect
sensitive data.

• Applications of Federated Learning: We explored FL applications in healthcare, Internet of Things

(IoT), and edge computing, highlighting its potential to enable collaborative learning across distributed devices
and environments.

• Challenges and Limitations: We identified communication overhead, non-IID data distribution, and
model heterogeneity as significant challenges in FL, which can impact performance and convergence.

• Mitigating Challenges: Recent Advances: We discussed recent techniques such as federated

averaging variants, adaptive client selection, and robust aggregation techniques designed to address challenges
and improve the effectiveness of FL algorithms.

• Future Directions and Prospects: We outlined future directions in FL, including the development of
standardized benchmarks, user-friendly frameworks, interdisciplinary collaborations, and ethical considerations,
to foster responsible innovation and realize the full potential of FL.
In conclusion, Federated Learning represents a promising approach to collaborative machine learning in
distributed environments, offering opportunities for privacy-preserving data analysis and knowledge sharing.

170
 J. Electrical Systems 20-5s (2024): 165-171

While FL presents challenges such as communication overhead and data distribution heterogeneity, recent
advances and future directions hold promise for overcoming these obstacles and advancing the field. As FL
continues to evolve, interdisciplinary collaborations, ethical considerations, and standardized evaluation
frameworks will be critical for ensuring its responsible deployment and societal impact.
Looking ahead, the future of Federated Learning is bright, with potential applications across various domains
and opportunities for innovation and collaboration. By addressing challenges, embracing interdisciplinary
research, and prioritizing ethical considerations, we can unlock the full potential of FL and drive positive change
in the machine learning landscape.
REFERENCES
[1] Kairouz, Peter, et al. "Advances and open problems in federated learning." arXiv preprint arXiv: 1912.04977 (2019).

[2] McMahan, H. Brendan, et al. "Communication-efficient learning of deep networks from decentralized data." Artificial
Intelligence and Statistics. PMLR, 2017.

[3] Yang, Qiang, et al. "Federated learning." Proceedings of the 2019 IEEE International Conference on Artificial
Intelligence and Virtual Reality. IEEE, 2019.

[4] Li, Tianjian, et al. "Federated learning: Challenges, methods, and future directions." IEEE Signal Processing Magazine
37.3 (2020): 50-60.

[5] Bonawitz, Keith, et al. "Towards federated learning at scale: System design." arXiv preprint arXiv: 1902.01046 (2019).

[6] Smith, Virginia, et al. "Federated learning: Strategies for improving communication efficiency." arXiv preprint arXiv:
1610.05492 (2016).

[7] Hard, Andrew, Kanishka Rao, and Rajiv Mathews. "Federated learning for mobile keyboard prediction." Proceedings
of the 2018 Workshop on Systems for ML and Open Source Software at NeurIPS. 2018.

[8] McMahan, H. Brendan, et al. "Federated learning: Collaborative machine learning without centralized training data."
Google Research Blog, April 2017.

[9] Zhao, Ruoxi, et al. "Federated learning with non-iid data." arXiv preprint arXiv: 1806.00582 (2018).

[10] Bagdasaryan, Eugene, et al. "How to backdoor federated learning." International Conference on Machine Learning.
PMLR, 2020.

[11] Maroua Drissi. "A state-of-the-art on federated learning for vehicular communications", Vehicular Communications,
2023.

[12] Omar Abdel Wahab, Azzam Mourad, Hadi Otrok, Tarik Taleb. "Federated Machine Learning: Survey, Multi-Level
Classification, Desirable Criteria and Future Directions in Communication and Networking Systems", IEEE
Communications Surveys & Tutorials, 2021.

[13] Steven M. Williamson, Victor Prybutok. "Balancing Privacy and Progress: A Review of Privacy Challenges, Systemic
Oversight, and Patient Perceptions in AI-Driven Healthcare", Applied Sciences, 2024.

[14] "Federated Learning", Springer Science and Business Media LLC, 2020.

[15] "Recent Trends in Image Processing and Pattern Recognition", Springer Science and Business Media LLC, 2024.

[16] Kumar Keshamoni. "Chapter 40 ChatGPT: An Advanceds Natural Language Processing System for Conversational AI
Applications—A Comprehensive Review and Comparative Analysis with Other Chatbots and NLP Models", Springer
Science and Business Media LLC, 2023.

[17] Ashish Rauniyar, Desta Haileselassie Hagos, Debesh Jha, Jan Erik Håkegård, Ulas Bagci, Danda B. Rawat, Vladimir
Vlassov. "Federated Learning for Medical Applications: A Taxonomy, Current Trends, Challenges, and Future
Research Directions", IEEE Internet of Things Journal, 2023.

171