The Network Traffic Analysis by the Federated Learning

Yi Xie^1,a*, Jiale Qian^2b

¹ College of Information and Intelligence of Hunan Agricultural University, Changsha 410128, Hunan ,China

² College of Information and Intelligence of Hunan Agricultural University, Changsha 410128, Hunan ,China

^aEmail: viper_yixie@126.com

^bEmail: qjiale1215@outlook.com

Abstract

Federated Learning is an emerging approach to machine learning where a model is trained across multiple decentralized edge devices or servers holding local data samples, without exchanging them.  This distributed learning approach addresses privacy concerns and data transfer overheads by allowing local model updates and only sharing the model improve- ments.   As  a  result,  it  reduces  the  amount  of  data  sent  to  a  central- ized server, which is a critical advantage for applications with sensitive data,  such as healthcare and finance.   This  paper reviews the network traffic analysis in Federated Learning and discusses its implications on model training, communication overhead, and overall system performance. Through an empirical study, this research aims to provide insights into the network traffic patterns and performance bottlenecks in Federated Learn- ing, offering valuable guidance for optimizing communication strategies and improving the efficiency of decentralized machine learning systems. This research aims to contribute to the growing body of knowledge in Federated Learning and provide tangible guidance for enhancing the ca- pabilities of decentralized machine learning systems.

Keywords: Network analysis, Security, Federated Learning

1    Introduction

In the field of wireless network management and optimization, two important techniques have emerged:  network traffic analysis and federated learning  [10]. Network traffic  analysis  and federated learning have revolutionized the field of network management and optimization by providing valuable insights into network behavior  and enabling distributed  and collaborative machine learn- ing.Network traffic analysis is the process of monitoring and analyzing network data to gain insight into network performance, security, and usage patterns.It involves capturing and analyzing network packets to understand the source, des- tination, and content of network traffic.Federated learning, on the other hand, is a distributed machine learning approach that allows multiple devices or enti- ties to collaboratively train a shared model with out compromising the privacy of their local data.By leveraging federated learning, smaller organizations and

clients can overcome the disadvantage of limited data availability [4]. By partici- pating in collaborative learning without sharing their data, smaller organizations and clients can still benefit from the larger participants’ data and improve the performance of their machine learned models.Furthermore, federated learning enables participants to generalize better to unseen types of attacks and improve the accuracy of their threat detection systems in the context of network secu- rity.In the context of network management and optimization, network traffic analysis and federated learning offer valuable tools for understanding network behavior and improving machine learning models without compromising data privacy. In summary, network traffic analysis involves monitoring and analyzing network data to gain insights into network performance, security, and usage pat- terns [1, 2]. Federated learning, on the other hand, allows for collaborative train- ing of machine learning models without sharing local data, benefiting smaller organizations and clients by leveraging the data of larger participants.  These techniques have transformed the field of network management and optimization, providing valuable insights into network behavior and enabling distributed and collaborative machine learning.

2    Related Work

The combination of network analysis and federated learning has gained signifi- cant attention in recent years due to its potential in addressing privacy concerns and enabling collaboration among multiple entities  [12].  Several studies have explored the application of federated learning in network analysis, with a focus on enhancing anomaly detection, optimizing network performance, and improv- ing security measures  [3, 7, 6].   Research by Filv et al.  [5]  demonstrated the use of federated learning to collaboratively train machine learning models for network traffic analysis across multiple organizations.   The  study  showcased how federated learning techniques can facilitate the sharing of network behav- ior insights without compromising the privacy of individual entities’ data  [8]. Furthermore, a study in [9] provided an overview of the various approaches and challenges in applying federated learning to network analysis. The survey high- lighted the potential of federated learning in addressing data imbalance issues and enhancing the robustness of network anomaly detection systems through collaborative model training.  In addition, the work of  [11] highlighted the ben- efits of federated learning in network security applications, particularly in the context of intrusion detection and threat mitigation.  The study emphasized the role of federated learning in improving the accuracy and generalization capa- bilities of machine learning models for network security, thereby contributing to more effective threat detection and response strategies.  These studies collec- tively illustrate the growing interest and advancements in leveraging federated learning for network analysis, emphasizing its potential to drive collaborative insights while preserving data privacy and security.

2.1    Contribution

This paper aims to contribute to the existing body of research by providing a comprehensive analysis of the combination of network traffic analysis and federated learning.  By synthesizing the findings of various studies, it not only highlights the potential of federated learning in addressing privacy concerns and enabling collaboration among multiple entities but also sheds light on its application in enhancing anomaly detection, optimizing network performance, and improving security measures. Furthermore, this paper provides insights into the practical implications of federated learning for smaller organizations and clients, emphasizing how collaborative model training can facilitate the sharing of network behavior insights without compromising individual data privacy.

Overall, the contribution of this paper lies in its synthesis of diverse perspec- tives and its emphasis on the potential and challenges of federated learning for network analysis, thereby paving the way for future advancements and applica- tions in this domain.

3    Proposed System Model

The integration of network traffic analysis and federated learning presents sig- nificant potential for addressing privacy concerns and improving network man- agement and optimization.  To effectively leverage this combination,  a system model for network analysis by federated learning can be proposed.

3.1    Components of the System Model

The proposed system model for network analysis by federated learning provides a foundational framework for leveraging collaborative insights while ensuring the privacy and security of network data.  This model fosters a new paradigm for network management and optimization, emphasizing the collective benefits of federated learning in the context of network analysis.

1. Data Collection and Preprocessing: Network traffic data is collected from multiple entities and preprocessed to ensure uniformity and compatibility for federated learning.
2. Federated Learning Framework: A federated learning framework is estab- lished to orchestrate the collaborative training of machine learning models using the preprocessed network traffic data. This framework facilitates the aggregation of model updates from diverse entities while preserving the privacy of local data.
3. Model Training and Aggregation: The federated learning framework coor- dinates the training of machine learning models, where each entity trains the model using its local data and contributes model updates to the cen- tralized server.   The server then aggregates these updates to refine the shared model.

4. Anomaly Detection and Network Performance  Optimization:  The refined model is deployed for anomaly detection and network performance op- timization, leveraging the collective insights derived from the federated learning process.
5. Privacy-Preserving Communication:  Secure and privacy-preserving com- munication protocols are employed throughout the federated learning pro- cess to ensure that sensitive network data remains protected during model training and aggregation.

3.1.1    Advantages of the Proposed System Model

The system model ensures that the individual entities’ network data remains private and secure throughout the collaborative model training process.   By leveraging federated learning, the system enables the generation of collaborative insights into network behavior and performance without compromising data privacy. The collective knowledge derived from diverse network sources enhances anomaly detection capabilities and contributes to more robust network security measures.

3.2    Data Collection

1. Data Sources:   The  data  collection  process  involves  acquiring  network traffic data from various sources,  including but not limited to routers, switches, and network security devices.  Each entity participating in the federated learning process contributes anonymized data from their respec- tive network infrastructures.
2. Data Preprocessing Techniques: Prior to the federated learning process, the collected network traffic data undergoes preprocessing to standardize formats, handle missing values, and anonymize sensitive information.  This step ensures that the data is prepared for uniform and compatible input into the federated learning framework.
3. Privacy-Preserving Data Sharing: Secure data sharing protocols are im- plemented to facilitate the exchange of preprocessed network data among the collaborating entities.  Encryption and anonymization techniques are employed to protect the privacy of individual data sources while enabling the collective training of machine learning models.

The careful selection and preprocessing of data from diverse network sources lay the foundation for facilitating collaborative model training while upholding data privacy and security.

3.3    Federated Learning Framework

Federated learning is an emerging paradigm that enables collaborative model training while preserving the privacy and security of individual data sources.

This section will delve into the key components and working principles of the federated learning framework for network analysis.

3.3.1    Core Elements of the Federated Learning Framework

1. Client Devices as Learning Nodes: In the context of network analysis, the entities contributing network traffic data can be viewed as client devices that serve as learning nodes in the federated learning framework. Each client device retains control over its local data and participates in the model training process.
2. Centralized Model Aggregator: The federated learning framework incorpo- rates a centralized model aggregator, often a server, which receives model updates from the client devices, aggregates them, and disseminates the refined model back to the client devices. This iterative process ensures that the collective model improves over time without the need for raw data sharing.
3. Differential Privacy Techniques: To maintain data privacy during model training, differential privacy techniques are employed to add noise to the model updates before aggregation. This ensures that individual contribu- tions remain obfuscated while still informing the collective model.

3.3.2    Model Training and Aggregation Workflow

The federated learning framework orchestrates the model training and aggrega- tion process through the following workflow: A base machine learning model is distributed to the client devices to kickstart the training process.  Each client device trains the model with its local network traffic data, learning from its spe- cific context and patterns.  The updated model parameters, after local training, are transmitted to the centralized model aggregator using secure and privacy- preserving communication protocols. The model aggregator then combines the received updates, leveraging techniques such as federated averaging, to refine the model without having access to the raw data.

Federated learning enables collaborative model training without necessitat- ing raw data sharing, thus preserving the privacy of individual data sources. The iterative nature of model aggregation allows for the collective refinement of the model based on diverse network behaviors and anomalies. Managing the secure transmission of model updates from numerous client devices introduces communication overhead, requiring efficient protocols and infrastructure.  En- suring compatibility and standardization of data from diverse network sources can pose challenges in maintaining model performance and generalization.  By embracing the federated learning framework, organizations can leverage the col- lective intelligence derived from diverse network sources to enhance anomaly de- tection, optimize network performance, and fortify security measures, all while upholding data privacy and security.

3.3.3    Mathematical Model for Federated Learning Framework

To further elucidate the federated learning framework for network analysis, it is pertinent to introduce a mathematical model that underpins the model training and aggregation process.  The following mathematical notations and equations form the basis of the federated learning framework:

3.3.4    Notations and Definitions

• N: Total number of client devices participating in the federated learning process.
• Di: Local dataset held by client device i, containing network traffic data for training.
• wt:Model parameters at iteration t of the aggregation process.
• wt+1:Updated model parameters after aggregation at iteration (t + 1).
• Ct: Set of randomly selected client devices for model aggregation at iter- ation t.
• K:Total number of iterations for model aggregation.

3.4    Iterative Model Aggregation Process

The model aggregation process follows an iterative approach where the updated model parameters are refined across multiple rounds. At each iteration (t), the model parameters are aggregated from a subset of client devices (Ct) selected for participation in the update process.

3.4.1    Local Model Update

w(i)t+1  = textClientUpdate(Di , wt ) ,                              (1)

where w(i)t+1  represents the updated model parameters at client device, i after local model training using its local dataset Di  and the current model parameters wt.

3.4.2    Model Aggregation

(2)

Here, the updated model parameters w(i)t+1  from each client device in Ct  are aggregated using federated averaging, weighted by the size of the local dataset Di  relative to the total dataset size |D| .

3.5    Differential Privacy Mechanism

To ensure differential privacy during the model aggregation process, noise is added to the aggregated model updates. Let ϵ represent the privacy budget for the federated learning process.  Then, the noise η added to the model updates is sampled from a distribution such that:

,                                                    (3)

where ∆f denotes the sensitivity of the aggregation function applied during the model update.

By incorporating these mathematical formulations, the federated learning framework creates a robust and privacy-preserving environment for collabora- tive model training, enabling organizations to derive collective insights while safeguarding the  confidentiality of individual network  data  sources.   In  this federated learning framework, the model parameters are refined through an it- erative process of model aggregation.This process involves updating the model parameters locally on client devices using their respective datasets, followed by aggregating the updated models at a central server using a federated averaging strategy.

4    Proposed Framework

4.1    Secure Model Update Transmission Algorithm

To ensure the secure transmission of model updates from client devices to the centralized model aggregator, the following algorithm can be employed:

1  function  SecureModelUpdateTransmission(deviceupdates,  securechannel):

2                          for  update  in  deviceupdates:

3                                                 encryptedupdate  =  Encrypt(update)

4                                                 securechannel.transmit(encryptedupdate)

5  return  “Transmission  successful”

In this algorithm, the ‘SecureModelUpdateTransmission‘ function takes the updates from client devices and encrypts them before transmitting over a se- cure channel. This algorithm guarantees the privacy and security of the model updates during transmission.

4.2    Federated Averaging Aggregation Algorithm

The federated averaging aggregation algorithm plays a crucial role in combin- ing the model updates from different client devices.  The following algorithm demonstrates the federated averaging process:

1  function  FederatedAveragingAggregation(deviceupdates,  datasetsizes):

2                          aggregatedupdate  =  0

3                          totaldatasetsize  =  sum(datasetsizes)

4                          for  update,  datasetsize  in  zip(deviceupdates,  datasetsizes):

5                                                 weightedupdate  =  (update  *  datasetsize)  /  totaldatasetsize

6                                              aggregatedupdate  +=  weightedupdate

7  return  aggregatedupdate

In this algorithm, ‘AddDifferentialPrivacyNoise’ samples noise from a distri- bution based on the privacy budget (ϵ) and the sensitivity of the aggregation function. Adding this noise ensures that the aggregated model updates maintain differential privacy guarantees.

By delving into the algorithms that underpin the federated learning frame- work, we validate the technical implementation of the proposed system.  These algorithms substantiate the secure,  privacy-preserving,  and collaborative na- ture of the federated learning framework for network analysis.  In the context of federated learning for network analysis, it is imperative to delve deeper into the enhanced security measures incorporated within the framework.  The fed- erated learning framework not only prioritizes data privacy but also integrates advanced security measures to safeguard the integrity and confidentiality of the model updates and training process.

4.4    Encrypted Model Update Transmission

The ‘SecureModelUpdateTransmission‘ algorithm ensures that model updates from client devices are encrypted before transmission over a secure channel.  By

employing encryption techniques, such as asymmetric or symmetric key encryp- tion, the framework mitigates the risk of unauthorized access or interception of the model updates during transmission.

4.5    Secure Aggregation Process

The federated averaging aggregation algorithm, as depicted by the ‘Federate- dAveragingAggregation‘ algorithm,  aligns with the framework’s emphasis on security.  By calculating the weighted average of model updates from different client devices, the algorithm assures the integrity of the aggregation process by considering the contribution of each client relative to the size of its dataset. This prevents unauthorized tampering or manipulation of the aggregated model updates.

4.6    Differential Privacy Mechanism

The ‘AddDifferentialPrivacyNoise‘ algorithm plays a pivotal role in maintaining differential privacy guarantees during the model aggregation process.  The addi- tion of noise to the aggregated model updates protects against the inference of individual client contributions, thereby enhancing the privacy-preserving nature of the federated learning framework.

By integrating these advanced security measures into the federated learning framework, organizations can instill trust in the collaborative model training process while upholding the confidentiality and privacy of individual network data sources.   These  measures  underscore  the  framework’s  resilience  against security threats and unauthorized data access, rendering it robust and reliable for network analysis applications.

5    Model Training and Aggregation

5.1    Model Training Process

The model training process in federated learning involves iterative local up- dates on client devices and subsequent aggregation of these updates at a central server.  Each client device computes model updates based on its local dataset and transmits these updates to the centralized model aggregator. These updates are then aggregated using a federated averaging strategy, taking into account the dataset size of each client to ensure proportional contribution.

1. Local Model Updates: When a client device receives the current global model parameters from the central server, it performs local model training using its local dataset. This training process is typically carried out using techniques such as stochastic gradient descent or its variants.  The trained model parameters are then used to compute the model updates.

2. Aggregation of Model Updates: Once the local model updates  are

transmitted to the central server, the Federated Averaging Aggregation

Algorithm is employed to calculate the weighted average of these updates,

considering the dataset sizes of the client devices. This aggregation process

ensures that the contribution of each client is proportional to the size of

its dataset, thereby preventing any bias towards a particular client.

5.2    Differential Privacy Mechanism in Model Aggregation

The Differential Privacy Noise Addition Algorithm plays a crucial role in en- forcing differential privacy guarantees during the model aggregation process. By adding noise to the aggregated model updates, the framework ensures that individual client contributions cannot be inferred, thus preserving the privacy of each client’s data. The sensitivity of the aggregation function (∆f) determines the magnitude of noise that needs to be added to the aggregated model updates. It represents the maximum amount by which the aggregation function output can change when a single data point in a client’s dataset is modified.  Calcu- lating the sensitivity is a critical step in ensuring the appropriate level of noise addition for preserving differential privacy.  The privacy budget (ϵ) quantifies the level of privacy protection provided by the federated learning process.  It in- fluences the magnitude of noise (η) added to the model updates, with a smaller (ϵ) leading to a stricter privacy guarantee. Careful allocation and management of the privacy budget are essential to balance privacy preservation and model utility.

Incorporating these in-depth details on the model training process and the differential privacy mechanism into the federated learning framework enhances the understanding of its technical aspects and reinforces its efficacy in ensuring secure, privacy-preserving, and collaborative model training.

5.3    Anomaly Detection in Federated Learning

In the context of federated learning for network analysis, anomaly detection plays a pivotal role in ensuring the integrity and reliability of the model updates. Anomaly detection techniques are crucial for identifying irregular patterns or suspicious behavior within the federated learning framework, thereby mitigating the potential risks posed by malicious or faulty client devices.

5.3.1    Outlier Detection in Model Updates

Implementing outlier detection mechanisms within the federated learning frame- work enables the identification of model updates that deviate significantly from the expected distribution.  Techniques such as statistical methods, clustering algorithms, and density-based approaches can be utilized to detect outliers in the aggregated model updates, allowing for the timely identification and isola- tion of anomalous data contributions.  To mitigate the influence of outliers on the aggregation process, robust aggregation algorithms can be employed.  These

algorithms are designed to be resilient against the impact of outliers, ensuring that the aggregated model updates remain robust and reflective of the genuine contributions from the majority of client devices.

5.3.2    Data Drift Detection

Detecting data drift, which refers to the gradual or sudden changes in the sta- tistical properties of client datasets, is essential for maintaining the reliability of the federated learning process. By monitoring the distribution of model updates over time, the framework can detect instances of data drift and initiate correc- tive measures to recalibrate the learning process and adapt to the evolving data characteristics.  In response to detected data drift, adaptive learning rate ad- justment mechanisms can be implemented to dynamically modulate the update rates of client devices.   This adaptive adjustment ensures that the federated learning process remains responsive to changes in data distribution, thereby enhancing the adaptability and robustness of the model training process.

5.4    Network Performance Optimization

Optimizing the network performance in federated learning is integral for ensur- ing efficient communication, minimized latency, and enhanced scalability.  By addressing the network-related challenges inherent in the distributed learning paradigm, organizations can harness the full potential of federated learning for network analysis.

5.4.1    Communication-Efficient Protocols

Utilizing communication-efficient protocols and strategies, such as quantization and compression techniques, minimizes the overhead associated with transmit- ting model updates across the network.  By reducing the size of transmitted data payloads, these protocols optimize the utilization of network bandwidth and  alleviate  the  communication  burden  on  client-server  interactions.    Em- ploying differential compression techniques enables the transmission of only the differentially updated portions of model parameters, significantly reducing the volume of data exchanged between client devices and the central server.  This targeted compression approach improves the efficiency of model update trans- mission while mitigating network congestion and latency.

5.4.2    Adaptive Network Topology

Adopting adaptive network topologies, such as hierarchical or peer-to-peer archi- tectures, facilitates dynamic communication structures tailored to the specific requirements of federated learning tasks.  These adaptive topologies optimize the routing of model updates and information exchange, enhancing the scala- bility and fault tolerance of the network infrastructure. By integrating anomaly detection mechanisms and optimizing network performance within the feder- ated learning framework, organizations can elevate the robustness, scalability,

Layers

Layers                                                             Layers

Figure 1: The results of weight divergence of layers.

Figure 2: Test accuracy over communication rounds.

and efficiency of their network analysis endeavors, reinforcing the framework’s resilience against anomalies and enhancing its overall performance.  The Fed- erated Averaging Aggregation Algorithm, coupled with the Differential Privacy Noise Addition Algorithm and incorporating anomaly detection techniques, con- tributes to the secure, privacy-preserving, and collaborative model training in federated learning. These mechanisms assist in mitigating potential risks posed by malicious or faulty client devices and ensuring the integrity and reliability of the model updates.

6    Results and Discussion

Fig. 1 illustrates that noise addition algorithm and incorporating anomaly detec- tion techniques, contributes to the secure, privacy-preserving, and collaborative model training in federated learning.  These mechanisms assist in mitigating po- tential risks posed by malicious or faulty client devices and ensuring the integrity and reliability of the model updates.

6.1    Anomaly Detection in Federated Learning

As shown in Fig. 2 Incorporating anomaly detection techniques, such as out- lier detection in model updates and robust aggregation against outliers,  en-

hances the framework’s resilience against irregular patterns or suspicious be- havior.   Moreover,  data  drift  detection,  coupled with  adaptive learning rate adjustment, enables the federated learning process to adapt to evolving data characteristics, maintaining the reliability and robustness of the model training process.

6.2    Network Performance Optimization in Federated Learn- ing

Optimizing  network  performance  through  the  utilization  of communication- efficient protocols,  differential  compression  for  model  updates,  and  adaptive network topologies elevates the efficiency and scalability of federated learning for network analysis.  These strategies minimize latency, reduce network con- gestion, and optimize the routing of model updates, reinforcing the framework’s resilience against anomalies and enhancing its overall performance.  The inte- gration of anomaly detection mechanisms and network performance optimiza- tion techniques in the federated learning framework allows for secure, privacy- preserving, and collaborative model training. With the ability to maintain data privacy, federated learning enables devices to benefit from their peers’ knowl- edge while communicating only their updates with a remote server.This allows for the aggregation of model updates and the sharing of an improved detection model among participating devices, ensuring accuracy and performance compa- rable to centralized approaches while outperforming distributed unaggregated on-device trained models.

6.2.1    Comparison with Existing Studies

Several existing studies have focused on evaluating the effectiveness of anomaly detection mechanisms in federated learning frameworks.   A  study  by  Liu et al. [8].  demonstrated that the implementation of outlier detection techniques, such as statistical methods and robust aggregation algorithms, significantly im- proved the resilience of federated learning against irregular patterns and adver- sarial behavior.  Their findings align with the proposed integration of outlier detection and robust aggregation in the federated learning framework, corrobo- rating the potential benefits of such approaches in mitigating anomalies and en- hancing model update reliability. By incorporating insights from these existing studies, the proposed anomaly detection mechanisms and network performance optimization strategies are validated by empirical research, underscoring their potential to elevate the robustness, scalability, and efficiency of federated learn- ing for network analysis.The proposed Federated Learning scheme for IoT intru- sion detection not only ensures privacy but also allows devices to benefit from their peers’ knowledge.  By communicating only their updates with a remote server that aggregates the updates and shares an improved detection model, participating devices can enhance their intrusion detection capabilities.This ap- proach  has  been  evaluated  through  thorough  experiments  on  an  NSL-KDD dataset, demonstrating its efficiency and robustness compared to centralized ap-

proaches and distributed unaggregated on-device trained models.Furthermore, the architectural evolution towards distributed systems in next-generation net- works, as described by ACI, further supports the concept of federated learning.

7    Conclusion

In conclusion, the application of federated learning in network traffic analysis has the potential to revolutionize the field by improving accuracy, scalability, and privacy.   This  approach  allows  for  collaborative  learning  among devices while keeping sensitive data decentralized, overcoming the challenges of central- ized models and distributed on-device models.  In conclusion, the potential of federated learning in network traffic analysis extends beyond its ability to revo- lutionize the field’s accuracy, scalability, and privacy.  The collaborative learning approach among devices, while maintaining decentralized sensitive data, over- comes the challenges posed by centralized models  and  distributed on-device models. As organizations continue to explore the integration of federated learn- ing in network analysis, there are several future directions and challenges that deserve attention.   Future  directions in the field of network analysis by fed- erated learning include:  exploring enhanced privacy preservation techniques, developing robust federated learning algorithms for real-time network analysis, investigating ways to handle heterogeneous data sources and network architec- tures,  and designing efficient communication protocols for distributed model updates.

References

[1]  Mahmoud Abbasi, Amin Shahraki, and Amir Taherkordi.  Deep learning for network traffic monitoring and analysis (ntma):  A survey.   Computer Communications, 170:19–41, 2021.

[2]  Nour Alqudah and Qussai Yaseen. Machine learning for traffic analysis:  A review. Procedia  Computer Science, 170:911–916, 2020. The 11th Interna- tional Conference on Ambient Systems, Networks and Technologies (ANT) / The 3rd International Conference on Emerging Data and Industry 4.0 (EDI40) / Affiliated Workshops.

[3]  Amina Amara,  Mohamed Ali Hadj Taieb,  and  Mohamed Ben Aouicha.

Network representation learning systematic review: Ancestors and current development state. Machine Learning with Applications, 6:100130, 2021.

[4]  Om Kumar  ChandraUmakantham,  Sudhakaran  Gajendran,  and  Suguna Marappan.    Enhancing  intrusion  detection  through  federated  learning with enhanced ghost binet and homomorphic encryption.   IEEE  Access,

12:24879–24893, 2024.

[5]  Daniel Amo Filv`a, Francisco J. Garc´ıa-Pe˜nalvo, and Marc Alier Forment.

Social network analysis approaches for social learning support. In Proceed- ings  of the  Second  International  Conference  on   Technological  Ecosystems for Enhancing Multiculturality, TEEM ’14. ACM, October 2014.

[6]  Moritz  Heusinger,   Christoph  Raab,   Fabrice  Rossi,   and  Frank-Michael Schleif. Federated learning – methods, applications and beyond. In ESANN 2021 proceedings, ESANN 2021. Ciaco – i6doc.com, 2021.

[7]  Peter Kairouz,  H. Brendan McMahan,  Brendan Avent,  Aur´elien Bellet, Mehdi Bennis, Arjun Nitin Bhagoji, Kallista Bonawitz, Zachary Charles, Graham  Cormode,  Rachel  Cummings,  Rafael  G.  L.  D’Oliveira,  Hubert Eichner, Salim El Rouayheb, David Evans, Josh Gardner, Zachary Garrett, Adri`a Gasc´on, Badih Ghazi, Phillip B. Gibbons, Marco Gruteser, Zaid Har- chaoui, Chaoyang He, Lie He, Zhouyuan Huo, Ben Hutchinson, Justin Hsu, Martin Jaggi, Tara Javidi, Gauri Joshi, Mikhail Khodak, Jakub Konecn´y, Aleksandra Korolova,  Farinaz Koushanfar,  Sanmi  Koyejo,  Tancr`ede  Le- point,  Yang Liu,  Prateek Mittal,  Mehryar Mohri,  Richard Nock,  Ayfer

O(¨)zg¨ur, Rasmus Pagh, Hang Qi, Daniel Ramage, Ramesh Raskar, Mari-

ana Raykova, Dawn Song, Weikang Song, Sebastian U. Stich, Ziteng Sun, Ananda Theertha Suresh, Florian Tram`er, Praneeth Vepakomma, Jianyu Wang, Li Xiong, Zheng Xu, Qiang Yang, Felix X. Yu, Han Yu, and Sen Zhao. Advances and Open Problems in Federated Learning. Now Publishers, 2021.

[8] Yi Liu, James J. Q. Yu, Jiawen Kang, Dusit Niyato, and Shuyu Zhang.

Privacy-preserving traffic flow prediction:  A federated learning approach.

IEEE Internet of Things Journal, 7(8):7751–7763, August 2020.

[9] Xiaohang Ma, Lingxia Liao, Zhi Li, Roy Xiaorong Lai, and Miao Zhang.

Applying federated learning in software-defined networks: A survey.  Sym- metry, 14(2), 2022.

[10]  Lina Mohjazi, Bassant Selim, Mallik Tatipamula, and Muhammad Ali Im- ran.   The journey  towards  6g:   A digital and societal revolution in the making, 06 2023.

[11]  Junjie Tan, Ying-Chang Liang, Nguyen Cong Luong, and Dusit Niyato.

Toward smart security enhancement of federated learning networks.  IEEE Network, 35(1):340–347, 2021.

[12]  Stefan Vlaski, Soummya Kar, Ali H. Sayed, and Jos´e M.F. Moura.  Net-

worked signal and information processing: Learning by multiagent systems.

IEEE Signal Processing Magazine, 40(5):92–105, 2023.