A Review of Privacy Enhancement Methods for Federated Learning in Healthcare Systems

Probably not a substantial claim: may not pass S3 (load-bearing).

Through Scopus search, 216 records were returned.

“Through Scopus search, 216 records were returned.”

Probably not a substantial claim: may not pass S4 (based).

There is no review conducted on privacy enhancement methods used in federated learning in healthcare systems.

“There is no review conducted on privacy enhancement methods used in FL in healthcare systems.”

• AssumptionRests on an assumption: The 40 articles identified through the Scopus search are representative of the existing privacy enhancement methods for federated learning in healthcare systems.

May combine two claims.

The existing work in this review presented approaches including adding gradient [42], extra elements [43], hashed EHR address [44], and additional rules [45].

“The existing work in this review presented the approaches including adding gradient [42], extra elements [43], hashed EHR address [44], and additional rules [45].”

• cites [42] Chang, Y., Fang, C., and Sun, W. (2021). A blockchain-based federated learning method for smart healthcare. Comput. Intell. Neurosci., 2021. sourceNot checked yet

  Not checked in the free run.

• cites [43] Passerat-Palmbach, J., Farnan, T., McCoy, M., Harris, J.D., Manion, S.T., Flannery, H.L., and Gleim, B. (2020, January 3–6). Blockchain-orchestrated machine learning for privacy preserving federated learning in electronic health data. Proceedings of the 2020 IEEE International Conference on Blockchain (Blockchain), Toronto, ON, Canada. sourceNot checked yet

  Not checked in the free run.

• cites [44] Salim, 2022, Federated learning-based secure electronic health record sharing scheme in medical informatics, IEEE J. Biomed. Health Inform. sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S4 (based).

Mixed approaches have shown high performance in FL by addressing challenges associated with privacy preservation.

“These mixed approaches have shown high performance in FL by addressing the challenges associated with privacy preservation.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Various techniques (such as ML and AI models, noise injection, encryption) have been used in existing literature to improve privacy of data in health systems.

“As discussed in the Section 4.1, various techniques (such as the use of various ML and AI models, injection of noise, use of encryption techniques etc) have been used in the existing literature to improve the privacy of data in health systems.”

May combine two claims.

Federated learning provides mitigation of privacy risks compared to centralized machine learning models but is still vulnerable to attacks when extracting information from the gradients sent by participants to the aggregation server.

“FL provides mitigation of privacy risks as compared to centralized machine learning models but is still vulnerable to attacks when it comes to extracting information from the gradients sent by the participants to the aggregation server.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Scopus has the widest coverage in peer-reviewed scholarly journals and publications among bibliographic databases.

“the relevant academic publishing was collected from Scopus which has been supported as having the widest coverage in peer-reviewed scholarly journals and publications among bibliographic databases [21,22,23].”

• cites [21] Gu, 2017, Characterisation of academic journals in the digital age, Scientometrics sourceNot checked yet

  Not checked in the free run.

• cites [22] 2005, Google Scholar: The pros and the cons, Online Inf. Rev. sourceNot checked yet

  Not checked in the free run.

• cites [23] Pranckutė, R. (2021). Web of Science (WoS) and Scopus: The titans of bibliographic information in today’s academic world. Publications, 9. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Federated learning provides a distributed machine learning system that enables participants to train their local model on local data and eliminates the requirement of data sharing.

“Compared to the traditional machine learning system which collects data from all sources and trains data at a central server, FL provides a distributed machine learning system that enables participants to train their local model on local data and eliminates the requirement of data sharing [1].”

• cites [1] Rieke, 2020, The future of digital health with federated learning, NPJ Digit. Med. sourceNot checked yet

  Not checked in the free run.

Classified as probably not a substantial claim.

In healthcare, federated learning allows multiple healthcare institutions to collaborate and train machine learning models based on decentralized data without sharing sensitive patient information.

“In the context of healthcare, FL allows multiple healthcare institutions to collaborate and train machine learning models based on decentralized data without the need of sharing sensitive patient information.”

May combine two claims.

Differential privacy provides required privacy guarantees at the cost of reducing the prediction accuracy of the machine learning model.

“DP provides the required privacy guarantees at the cost of reducing the prediction accuracy of the machine learning model.”

May combine two claims.

Blockchain technology can help overcome challenges of ensuring participating devices are trustworthy and not malicious, and fairness and transparency in reward distribution, to manage the federated learning process without intermediaries.

“Blockchain technology can help overcome these challenges to manage the FL process without the need for intermediaries.”

• cites [11] Nguyen, 2021, Federated learning meets blockchain in edge computing: Opportunities and challenges, IEEE Internet Things J. sourceNot checked yet

  Not checked in the free run.

• cites [39] Shayan, 2020, Biscotti: A blockchain system for private and secure federated learning, IEEE Trans. Parallel Distrib. Syst. sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

A multi-modal machine learning model uses homomorphic encryption to securely share data required for training machine learning models.

“A multi-modal machine learning model is proposed to effectively solve the problem of data islands and uses homomorphic encryption to securely share data required for training machine learning models [36].”

• cites [36] Ji, J., Yan, D., and Mu, Z. (2022, January 15–17). Personnel status detection model suitable for vertical federated learning structure. Proceedings of the 2022 6th International Conference on Machine Learning and Soft Computing, Haikou, China. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

In federated learning for COVID-19 detection using chest X-rays and symptoms across multiple hospitals, non-IID privacy preservation is achieved by adding differential privacy stochastic gradient descent, which provides resilience to adaptive attacks with auxiliary information.

“In addition to the measures to improve model accuracy for data, non-IID privacy preservation is achieved by adding the differential privacy stochastic gradient descent (DP-SGD) which provides required resilience to adaptive attacks auxiliary information [26].”

• cites [26] Ho, T.T., Tran, K.D., and Huang, Y. (2022). FedSGDCOVID: Federated SGD COVID-19 Detection under Local Differential Privacy Using Chest X-ray Images and Symptom Information. Sensors, 22. sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

An FL-based anomaly detection system using differential privacy provides privacy guarantees to data owners and uses machine learning to secure network traffic.

“An FL-based anomaly detection system is proposed that uses DP to provide privacy guarantees to data owners and uses machine learning to secure network traffic [29].”

• cites [29] Cholakoska, A., Pfitzner, B., Gjoreski, H., Rakovic, V., Arnrich, B., and Kalendar, M. (2021, January 21–26). Differentially Private Federated Learning for Anomaly Detection in eHealth Networks. Proceedings of the 2021 ACM International Joint Conference on Pervasive and Ubiquitous Computing and Proceedings of the 2021 ACM International Symposium on Wearable Computers, New York, NY, USA. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

An end-to-end pipeline consisting of federated learning and differential privacy for health data streams using clustering reduces model training time with high accuracy, and prediction accuracy only decreases by 2% due to differential privacy.

“An end-to-end pipeline consisting of FL and DR is proposed for health data streams using a clustering mechanism to reduce model training time with high accuracy [31]. DP mechanism is used by the system to provide privacy guarantees and results showed that prediction accuracy only decreases by 2% for the trained model due to this change.”

• cites [31] Imtiaz, S., Horchidan, S.F., Abbas, Z., Arsalan, M., Chaudhry, H.N., and Vlassov, V. (2020, January 10–13). Privacy preserving time-series forecasting of user health data streams. 2020, Atlanta, GA, USA. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

An edge aggregator that introduces artificial noise to client model parameters in three steps still provided good model accuracy while achieving privacy due to the introduction of the noise.

“Therefore, an edge aggregator is proposed that introduces artificial noise to the client model parameters in three steps [27]. The introduction of this aggregator still provided good model accuracy while achieving privacy due to the introduction of the noise.”

• cites [27] Akter, 2022, Edge Intelligence: Federated Learning-Based Privacy Protection Framework for Smart Healthcare Systems, IEEE J. Biomed. Health Inform. sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

A blockchain-enabled collaborative federated learning architecture using differential privacy for safeguarding user data and blockchain for secure sharing of data provides security to processes requiring data sharing, per theoretical analysis.

“A blockchain-enabled collaborative FL architecture is proposed that uses DP for safeguarding user data and blockchain for secure sharing of data [30]. The theoretical analysis showed that the system provides security to the processes that require sharing of data.”

• cites [30] Das, P., Singh, M., and Roy, D.G. (2021, January 7–11). A secure softwarized blockchain-based federated health alliance for next generation IoT networks. Proceedings of the 2021 IEEE Globecom Workshops (GC Wkshps), Madrid, Spain. sourceNot checked yet

  Not checked in the free run.

Classified as probably not a substantial claim.

A privacy-aware and resource-saving collaborative learning protocol provides privacy for electronic health records in resource-saving scenarios by outsourcing the major learning component to cloud servers and using lightweight data perturbation and packed partially homomorphic encryption.

“A privacy-aware and resource-saving collaborative learning protocol is proposed to provide privacy for EHRs when the FL system is engaged in resource-saving scenario [38].”

• cites [38] Hao, M., Li, H., Xu, G., Liu, Z., and Chen, Z. (2020, January 7–11). Privacy-aware and resource-saving collaborative learning for healthcare in cloud computing. Proceedings of the ICC 2020 IEEE International Conference on Communications (ICC), Dublin, Ireland. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Statistical feature selection and differential privacy approaches reduce data transmission and add noise to local models before sending them to the federated learning server, and results showed good accuracy with the desired level of privacy.

“Statistical feature selection and differential privacy approaches are adopted [28] to reduce the data transmission and to add noise to the local models before sending those to the FL server for aggregation. The results showed good accuracy with the desired level of privacy applied to the model parameters.”

• cites [28] Islam, T.U., Ghasemi, R., and Mohammed, N. (2022, January 26–29). Privacy-Preserving Federated Learning Model for Healthcare Data. Proceedings of the 2022 IEEE 12th Annual Computing and Communication Workshop and Conference (CCWC), Las Vegas, NV, USA. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Chang, Fang, and Sun proposed an FL-chain system with adaptive differential privacy that adjusts noise on the gradient to balance privacy and accuracy in a MIoT environment, with a gradient verification-based consensus protocol preventing malicious attacks and single point of failure.

“Chang, Fang, and Sun [42] proposed an FL-chain system with an adaptive differential privacy algorithm. Technically, the system adjusts the amount of noise by adding on the gradient to achieve the balance between privacy and accuracy in a MIoT environment. The key strength of this FL-chain system is the gradient verification-based consensus protocol which is implemented to prevent malicious attacks”

• cites [42] Chang, Y., Fang, C., and Sun, W. (2021). A blockchain-based federated learning method for smart healthcare. Comput. Intell. Neurosci., 2021. sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Lakhan et al. implemented fraud detection using rule-based blockchain in an IoMT federated learning network, proposing an FL-BETS framework with dynamic heuristics to identify and ensure privacy preservation and fraud of data at various levels.

“Lakhan et al. [45] implemented fraud detection using the rule-based blockchain in the IoMT FL network. In this study, an FL-chain-enabled task scheduling (FL-BETS) framework with different dynamic heuristics was proposed. The goal of FL-BETS is to identify and ensure the privacy preservation and fraud of data at various levels.”

• cites [45] Lakhan, 2022, Federated-learning based privacy preservation and fraud-enabled blockchain IoMT system for healthcare, IEEE J. Biomed. Health Inform. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Salim and Park presented a secure EHR scheme in a hospital using federated learning and blockchain, where IPFS stores private data, EHR hashed addresses are recorded on a Consortium Blockchain, access is granted by patients via smart contracts, and EHR data are trained locally and globally.

“Salim and Park [44] presented a secure EHR scheme in a hospital that adopted FL and blockchain technology. In the scheme, Inter Planetary File System (IPFS) stores private data. In the IPFS, all EHR hashed addresses are recorded using a Consortium Blockchain-based network. Access to the EHR is granted by individual patients enabled by smart contracts. EHR data are trained both locally and globally”

• cites [44] Salim, 2022, Federated learning-based secure electronic health record sharing scheme in medical informatics, IEEE J. Biomed. Health Inform. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Gupta et al. proposed a hierarchical federated learning approach for IoMT allowing data aggregation at different levels enabling multiple healthcare organisations to collaborate securely, with grouping based on disease groups and FEDTIMEDIS LSTM for anomaly detection.

“Gupta et al. [46] proposed a hierarchical FL approach for the Internet of Medical Things (IoMT) application which allows data aggression at different levels enabling multiple healthcare organisations to collaborate with each other securely. A novel approach of grouping is proposed where the data is grouped based on different disease groups. The authors also developed a novel technique, called”

• cites [46] Gupta, D., Kayode, O., Bhatt, S., Gupta, M., and Tosun, A.S. (2021, January 13–15). Hierarchical Federated Learning based Anomaly Detection using Digital Twins for Smart Healthcare. Proceedings of the 2021 IEEE 7th International Conference on Collaboration and Internet Computing (CIC), Atlanta, GA, USA. sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Singh et al. introduced a Hierarchical FL model based on Dew-Cloud technology offering enhanced data confidentiality for IoMT, with HLSTM deployed at distributed Dew servers for intrusion detection, and the authors claim it can gain users' trust.

“Singh et al. [47] introduced a Hierarchical FL (HFL) model based on Dew-Cloud technology, which offers an enhanced level of data confidentiality for Internet of Medical Things (IoMT) applications. In this study, for detecting intrusion in incoming traffic a hierarchical long-term memory (HLSTM) model has also been deployed at distributed Dew servers. The authors claim that the proposed model can”

• cites [47] Singh, 2022, Dew-Cloud-Based Hierarchical Federated Learning for Intrusion Detection in IoMT, IEEE J. Biomed. Health Inform. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

In an FL-chain system, participating devices can be identified and verified using their public key, ensuring only trustworthy devices participate.

“In an FL-chain (FL with blockchain technology) system, the participating devices can be identified and verified using their public key [40]. This ensures that only trustworthy devices are allowed to participate in the learning process.”

• cites [40] Ali, 2021, Integration of blockchain and federated learning for Internet of Things: Recent advances and future challenges, Comput. Secur. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Abdellatif et al. presented a hierarchical federated learning application to enhance privacy in healthcare, with End User, edge node, and centralized server layers; End Users train a deep learning model without sharing raw data by connecting to edge nodes, which connect to a centralized server; the method achieves efficiency considering varying data distributions and wireless conditions.

“Abdellatif et al. [48] presented a hierarchical FL application that aims to enhance privacy in healthcare applications. The proposed architecture comprises different layers such as End User (EU) layer, Edge node layer, and Centralized Server layer. In this approach, EUs collaborate to train a deep learning model without sharing raw data, by connecting to the edge nodes within their communication”

• cites [48] Abdellatif, 2022, Communication-efficient hierarchical federated learning for IoT heterogeneous systems with imbalanced data, Future Gener. Comput. Syst. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Wang et al. proposes PPFLEC using a privacy protection protocol based on a shared secret and weight mask, a random mask scheme of secret sharing, lightweight features for IoMT on the edge, and uses a hash function and digital signature to ensure message integrity and protect from replay attacks.

“Wang et al. [50] proposes a Privacy Protection Scheme for FL under Edge Computing (PPFLEC) that uses a newly proposed privacy protection protocol based on a shared secret and weight mask. The proposed protocol uses a random mask scheme of secret sharing and focuses on achieving lightweight features to make it suitable for the Internet of Medical Things on the edge. This work also uses a hash”

• cites [50] Wang, 2022, Privacy-preserving federated learning for internet of medical things under edge computing, IEEE J. Biomed. Health Inform.Not checked yet

  Not checked in the free run.

May combine two claims.

Wang et al. proposed a novel protocol for parameter aggregation in federated learning for edge devices with limited capacity, aiming to improve security while reducing communication and computational overhead, using orthogonal gradient aggregation to protect previously learned knowledge when related training samples are removed and ensure privacy.

“Wang et al. [52] proposed a novel protocol for parameter aggression in FL and its application for edge devices that have limited capacity. The goal is to improve security while reducing communication and computational overhead. The authors has proposed a novel approach of orthogonal gradient aggregation (OGA) that can protect previously learned knowledge when the related training samples are”

• cites [52] Wang, 2022, Secure and efficient parameters aggregation protocol for federated incremental learning and its applications, Int. J. Intell. Syst. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Peer-to-peer sharing in federated learning facilitates exchange of model updates between participating devices and enhances efficiency by reducing computational burden on centralized servers, and can improve user privacy without a central intermediary.

“To achieve high efficiency, peer-to-peer sharing (P2PS) in FL is utilised to facilitate the exchange of model updates between participating devices and enhance the efficiency of the learning process by reducing the computational burden on centralized servers. In addition, P2PS can also improve user privacy without the need for a central intermediary to store and process sensitive data.”

May combine two claims.

Akter et al. proposed an edge aggregator that introduces noise to clients' model parameters and employs differential privacy while considering the privacy budget, and results showed high model accuracy.

“To achieve this goal, Akter et al. [27] proposed an edge aggregator that introduces noise to the clients’ model parameters and by employing DP while considering the privacy budget, the results showed high model accuracy.”

• cites [27] Akter, 2022, Edge Intelligence: Federated Learning-Based Privacy Protection Framework for Smart Healthcare Systems, IEEE J. Biomed. Health Inform. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

A border gateway and cloud aggregator scheme using a mask and digital signature, with a common mask vector added before sending the updated model to gateways, shows higher accuracy than systems using differential privacy because no noise is added.

“Border gateways collect the initial model from the cloud and then train the model with the data collected from edge intelligent devices. A mask and digital signature are used to secure model parameters after training at the gateways. The aggregator at the cloud adds a common mask vector before sending the updated model to the gateways. As no noise is added to the model parameters so the model”

• cites [50] Wang, 2022, Privacy-preserving federated learning for internet of medical things under edge computing, IEEE J. Biomed. Health Inform.Not checked yet

  Not checked in the free run.

May combine two claims.

Chen et al. proposed PFL-IU, an efficient and privacy-preserving system compatible with irrelevant updates, and results showed good performance in privacy, accuracy, convergence and efficiency.

“With the increasing EHRs collected from smartphones and wearable devices, Chen et al. [49] proposed an efficient and privacy-preserving system (PFL-IU), which is compatible with irrelevant updates. The results showed that the PFL-IU has good performance in terms of privacy, accuracy, convergence and efficiency.”

• cites [49] Chen, H., Li, H., Xu, G., Zhang, Y., and Luo, X. (2020, January 20–23). Achieving privacy-preserving federated learning with irrelevant updates over e-health applications. Proceedings of the ICC 2020–2020 IEEE International Conference on Communications (ICC), Dublin, Ireland. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Hakak et al. proposed a framework providing privacy by incorporating cloud, edge and application modules, with local and central aggregation, local controller saving data, encrypted local weights aggregated locally, sent to cloud aggregator, using homomorphic encryption.

“Hakak et al. [51] proposed a framework that provides privacy of data by incorporating cloud, edge and application modules. By providing aggregation at the local device and central cloud level, the privacy of data is maintained. Based on application requirements local controller saves the data and the updated encrypted local weights from the local devices are aggregated locally by the local aggreg”

• cites [51] Hakak, S., Ray, S., Khan, W.Z., and Scheme, E. (2020, January 10–13). A Framework for Edge-Assisted Healthcare Data Analytics using Federated Learning. Proceedings of the 2020 IEEE International Conference on Big Data (Big Data), Atlanta, GA, USA. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Gandhi et al. proposed a federated learning system to maintain integrity of sensitive medical data using deep-learning models for cancer text classification, and results showed effectiveness in addressing privacy issues.

“Gandhi et al. [35] proposed an FL system to maintain the integrity of sensitive medical data by using a few deep-learning models for cancer text classification. Their results showed that the proposed framework is effective in addressing the privacy issues of sensitive medical data for cancer text classification.”

• cites [35] Gandhi, N., Mishra, S., Bharti, S.K., and Bhagat, K. (2021, January 9–11). Leveraging towards Privacy-preserving using Federated Machine Learning for Healthcare Systems. Proceedings of the 2021 IEEE International Conference on Electronics, Computing and Communication Technologies (CONECCT), Bangalore, India. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

DP has no contribution to data governance in an FL system; data quality, data access policies and consent management are not part of the DP approach.

“DP has no contribution to data governance in an FL system, aspects like data quality, data access policies and consent management are not part of the DP approach.”

May combine two claims.

A syntactic anonymity approach for federated learning systems provides required privacy guarantees and still has good model accuracy, providing interpretable universal privacy guarantees.

“Instead of the commonly used differential privacy approach, the syntactic approach is proposed for FL systems that provide required guarantees for privacy and still have good model accuracy [53]. The proposed syntactic anonymity approach provides interpretable universal privacy guarantees.”

• cites [53] Choudhury, O., Gkoulalas-Divanis, A., Salonidis, T., Sylla, I., Park, Y., Hsu, G., and Das, A. (September, January 29). A syntactic approach for privacy-preserving federated learning. Proceedings of the 24th European Conference on Artificial Intelligence (ECAI 2020), Santiago de Compostela, Spain.Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Combining blockchain with homomorphic encryption presents a decentralised, transparent privacy-preserving solution for smart parking with minimum computation and communication overhead.

“For example, by combining blockchain with homomorphic encryption a decentralised, transparent privacy-preserving solution is presented for smart parking with minimum computation and communication overhead [2].”

• cites [2] Haritha, 2023, Asymmetric Consortium Blockchain and Homomorphically Polynomial-Based PIR for Secured Smart Parking Systems, Comput. Mater. Contin.Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

Hybrid approaches improve the overall performance of FL models by leveraging the strengths of different models.

“The purpose of the hybrid approaches is to improve the overall performance of the models used in FL by leveraging the strengths of different models.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Domain-specific knowledge can be incorporated into training to improve model robustness.

“In addition, domain-specific knowledge can be incorporated into the training to improve the model’s robustness.”

May combine two claims.

The federated learning system using a syntactic privacy model used healthcare data to predict adverse drug reactions and mortality rates and benchmarked against the centralized model.

“The FL system used healthcare data to predict adverse drug reactions and mortality rates and benchmarked against the centralized model.”

• cites [53] Choudhury, O., Gkoulalas-Divanis, A., Salonidis, T., Sylla, I., Park, Y., Hsu, G., and Das, A. (September, January 29). A syntactic approach for privacy-preserving federated learning. Proceedings of the 24th European Conference on Artificial Intelligence (ECAI 2020), Santiago de Compostela, Spain.Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

All clients may use the same encryption/decryption methods, which provides transparency of the homomorphic encryption mechanism.

“All clients may use the same encryption/decryption methods which provides transparency of the homomorphic encryption mechanism.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The decentralized and trust-enhancing properties of blockchain technology facilitate collaboration in FL systems.

“The decentralized and trust-enhancing properties of blockchain technology facilitate collaboration in FL systems.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The transparent nature of blockchain enables all participants in FL systems to have visibility into updates and changes made to the model.

“The transparent nature of the blockchain enables all participants in the FL systems to have visibility into the updates and changes made to the model.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Participants have higher trust in the shared model and the integrity of updates due to the consensus mechanism and immutability.

“Participants have higher trust in the shared model and the integrity of the updates due to the consensus mechanism and immutability.”

May combine two claims.

Collaboration is supported in HE-based systems because clients keep control of their data and allow computations on encrypted shared model information without raising privacy concerns.

“Collaboration is supported in HE-based systems as clients keep control of their data and allow computations on the encrypted shared model information without raising privacy concerns.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Smart contracts can be utilized to define and enforce data usage policies, ensuring that data is shared and utilized in accordance with predefined rules and regulations.

“Smart contracts can be utilized to define and enforce data usage policies, ensuring that data is shared and utilized in accordance with predefined rules and regulations.”

May combine two claims.

Blockchain technology in FL systems enhances privacy, security, and transparency.

“The strengths of using blockchain technology in FL systems lie in its ability to enhance privacy, security, and transparency.”

Probably not a substantial claim: may not pass S4 (based).

DP introduces noise to data, which adversely affects the prediction accuracy of ML algorithms.

“DP does that by introducing noise to the data, which adversely affects the prediction accuracy of the ML algorithms.”

Classified as probably not a substantial claim.

Blockchain's architecture, including local data storage on individual devices, cryptographic algorithms and immutability, provides enhanced data privacy and security.

“The inherent architecture of blockchain including the local data storage on individual devices, in addition to the cryptographic algorithms and immutability feature in the blockchain technology, provides enhanced data privacy and security.”

May combine two claims.

All DP approaches add computational overhead and increase design complexity because clients perform additional tasks for adding noise to model parameters before sending them to the server.

“Similarly, as the clients need to perform additional tasks for adding noise to the model parameters before sending those to the server so all DP approaches add computational overhead and increase design complexity.”

May combine two claims.

The transparent nature of blockchain can expose certain information, which may be harmful to sensitive data in FL-chain systems.

“The transparent nature of blockchain can expose certain information. This may be harmful to sensitive data in FL-chain systems.”

May combine two claims.

HE-based approaches for FL have weaknesses in resource usage; the encryption process adds design complexity, increased computational overhead and increased energy consumption for resource-constrained IoT devices.

“The encryption process that clients need to perform adds design complexity, increased computational overhead and also results in increased energy consumption for resource constraint IoT devices.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Governance is an important factor in HE because management of encryption keys is a challenge.

“Governance is an important factor in this case as the management of encryption keys is a challenge.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Integrating blockchain into an FL system requires significant development work and expertise, smart contracts, consensus mechanisms, and interoperability with existing FL protocols.

“Integrating blockchain into an FL system requires significant development work and expertise. It requires smart contracts and consensus mechanisms, and interoperability with existing FL protocols.”

May combine two claims.

Increased computational overhead in FL-chain could result in slower training speed, impact efficiency, and face scalability challenges.

“This could result in a slower training speed, impact the efficiency of the system, and face scalability challenges.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The distributed nature of the system and collaboration at the tier level adds a privacy and security layer to the system.

“The distributed nature of the system and collaboration at the tier level adds a privacy and security layer to the system.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Computational complexity can impact the efficiency and speed of FL systems, especially during the training phase.

“The computational complexity can impact the efficiency and speed of the FL systems, especially during the training phase.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The transparency offered by blockchain makes all transactions visible to all participants, enhancing trust and auditability but conflicting with privacy requirements in FL systems.

“The transparency offered by blockchain makes all transactions visible to all participants in the network, which enhances trust and auditability but also conflicts with privacy requirements in FL systems.”

May combine two claims.

Cryptographic algorithms and consensus mechanisms enhance security but come at a cost in terms of computational overhead.

“Cryptographic algorithms and consensus mechanisms enhance the security feature, however, it comes at a cost in terms of computational overhead.”

May combine two claims.

Hierarchical approaches have inherent strengths in decentralization and transparency because adding multiple edge servers and having a transparent grouping mechanism can make these approaches suitable for a large number of FL applications.

“Hierarchical approaches have inherent strengths in terms of decentralization and transparency as adding multiple edge servers and having a transparent grouping mechanism can make these approaches suitable for a large number of FL applications.”

Probably not a substantial claim: may not pass S3 (load-bearing).

From a governance perspective, integrating blockchain in FL systems may require extensive coordination and agreement among participants, potentially slowing decision-making.

“From a governance perspective, to integrate blockchain technology in FL systems, reaching consensus on system updates, protocol changes, or policy modifications may require extensive coordination and agreement among participants, potentially slowing down decision-making processes.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The complexity of the FL chain could increase development time and budget.

“The complexity of the FL chain could increase development time and budget.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Incorporating blockchain into an FL system requires detailed design and integration, including developing smart contracts, defining consensus mechanisms, and ensuring compatibility with existing FL protocols.

“Incorporating blockchain into an FL system requires detailed design and integration. Developing smart contracts, defining consensus mechanisms, and ensuring compatibility with the”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Blockchain's decentralized architecture enhances security and transparency but comes at the cost of scalability.

“Blockchain’s decentralized architecture enhances security and transparency. However, it comes at the cost of scalability.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Governance and collaboration can be considered strengths of hierarchical systems because they provide a structural framework for managing and controlling data and grouping clients.

“Governance and collaboration can be considered strengths of hierarchical systems as they provide a structural framework for managing and controlling data and grouping clients.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Consensus on system updates, policy modifications, and protocol changes may require extensive coordination and agreement among participants, which may slow decision-making.

“Consensus on system updates, policy modifications, and protocol changes may require extensive coordination and agreement among participants, which may slow down the decision-making process.”

May combine two claims.

P2PS minimizes exposure of sensitive user data to third-party entities and reduces the risk of data breaches.

“The exposure of sensitive user data to third-party entities and reduces the risk of data breaches is minimised.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The introduction of multiple edge servers introduces more vulnerable entities in the system where all participant information and data need to be secured.

“The introduction of multiple edge servers introduces more vulnerable entities in the system where all participant information and data need to be secured.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

The overhead of maintaining network connections increases and communication reliability decreases with a growing number of participants.

“In addition, the overhead of maintaining network connections increases and communication reliability decreases with the growing number of participants.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Transmission of large model parameters among participants results in high network bandwidth and increases communication latency.

“The transmission of large model parameters among participants results in high network bandwidth and increase communication latency.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

As the number of participating clients increases, computational overhead and energy consumption at the cloud and edge servers increase but will not affect the clients.

“However, as the number of participating clients increases computational overhead and energy consumption at the cloud and edge servers increase but will not affect the clients.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Some malicious participants might exploit vulnerabilities in the system to gain unauthorized access to sensitive user data or manipulate the learning process.

“In addition, some malicious participants might try to exploit vulnerabilities in the system to gain unauthorized access to sensitive user data or manipulate the learning process.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Managing large-scale FL systems with a massive number of participants can be difficult.

“Although the P2PS is able to provide offer scalability as mentioned in the advantages, managing large-scale FL systems with a massive number of participants can be difficult.”

May combine two claims.

Due to decentralized mechanisms for resource discovery and P2P connections, it is difficult to discover available and suitable participants with the desired computational resources and challenging to identify capable and trustworthy participants.

“Due to the decentralized mechanisms for resource discovery and P2P connections, it is difficult to discover available and suitable participants with the desired computational resources and it is challenging to identify capable and trustworthy participants in the learning process.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

In P2P FL systems, communication between participants and model updates could lead to overhead and high communication costs.

“In P2P FL systems, The communication between participants and model updates could lead to overhead and high communication costs.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Governance and design complexity are considerable challenges in hierarchical systems because management of a large number of clients and servers and designing appropriate aggregation mechanisms are non-trivial.

“However, governance and design complexity are considerable challenges in these systems as management of a large number of clients and servers and designing appropriate aggregation mechanisms are non-trivial in those cases.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

P2PS techniques utilize computing resources across multiple participants where each participant contributes its own processing power, which can significantly improve the overall computational capabilities of FL systems.

“P2PS techniques utilize the computing resources across multiple participants where each participant contributes its own processing power, which can significantly improve the overall computational capabilities of the FL systems.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Due to the introduction of multiple tiers, the global model may take longer to converge.

“Also, due to the introduction of multiple tiers, the global model may take longer to converge and this trade-off needs to be considered for systems requiring real-time task management.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Direct communication among participants in P2PS architecture could bring security risks, such as an increased risk of data leakage and unauthorized access.

“Direct communication among participants in P2PS architecture could bring security risks in the FL systems, such as an increased risk of data leakage and unauthorized access.”

Probably not a substantial claim: may not pass S3 (load-bearing).

It is very challenging to coordinate and synchronize model updates across numerous participants.

“It is very challenging to coordinate and synchronize model updates across numerous participants.”

May combine two claims.

P2PS in FL allows direct communication between participants, bypassing the need for a central server, alleviating network congestion and reducing overall communication latency.

“P2PS in FL allows for direct communication between participants, bypassing the need for a central server. By doing so, the P2PS is able to alleviate network congestion and reduce the overall communication latency in the system.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

P2PS reduces the impact of individual failures, making FL systems resilient.

“Additionally, P2PS reduces the impact of individual failures which makes the FL systems resilient.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Energy efficiency is achieved by local processing, but complex models can consume more energy.

“Energy efficiency is achieved by local processing, but complex models can consume more energy.”

Probably not a substantial claim: may not pass S4 (based).

They explore techniques that can provide strong privacy guarantees while maintaining acceptable levels of model accuracy and performance.

“They explore techniques that can provide strong privacy guarantees while still maintaining acceptable levels of model accuracy and performance.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Heterogeneity among devices caused by variations in hardware and software can complicate optimization and compatibility.

“Heterogeneity among devices caused by variations in hardware and software can complicate optimization and compatibility.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Network bandwidth and latency could hassle data synchronization and model updates.

“Network bandwidth and latency could hassle data synchronization and model updates.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Transferring large model parameters across participants can consume significant network bandwidth and increase latency.

“Transferring large model parameters across participants can consume significant network bandwidth and increase latency.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

These approaches strive for compatibility and interoperability with existing FL frameworks and infrastructures and encourage continuous improvement and innovation in privacy protection.

“These approaches strive for compatibility and interoperability with existing FL frameworks and infrastructures and encourage continuous improvement and innovation in privacy protection.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Due to the distributed architecture, scalability can be achieved, allowing for larger datasets and improved model performance.

“Additionally, due to the distributed architecture, scalability can be achieved, allowing for larger datasets and improved model performance.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Mixed, hybrid, and miscellaneous approaches try to leverage their strengths and compensate for their weaknesses to provide a more comprehensive and robust privacy framework.

“Mixed, hybrid, and miscellaneous approaches try to leverage their strengths and compensate for their weaknesses so as to provide a more comprehensive and robust privacy framework.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

P2P allows participants to directly share model updates, which results in trust concerns.

“However, P2P allows participants directly share model updates, which results in trust concerns.”

Probably not a substantial claim: may not pass S4 (based).

They improve the resilience of the FL system against various privacy attacks by integrating multiple techniques.

“In addition, they improve the resilience of the FL system against various privacy attacks by integrating multiple techniques.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

P2PS provides scalability by distributing computational tasks among participants, but it is difficult to coordinate and synchronize the learning process as the number of participants increases.

“P2PS provides scalability by distributing computational tasks among participants, it is difficult to coordinate and synchronize the learning process as the number of participants increases.”

Probably not a substantial claim: may not pass S3 (load-bearing).

With the lack of centralized control, P2PS could introduce security risks.

“With the lack of centralized control, the P2PS could introduce security risks.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

The specific strengths and effectiveness of mixed, hybrid, and miscellaneous approaches depend on the selection and combination of techniques, as well as their proper implementation and configurations.

“It’s worth noting that the specific strengths and effectiveness of mixed, hybrid, and miscellaneous approaches depend on the selection and combination of techniques, as well as their proper implementation and configurations.”

Probably not a substantial claim: may not pass S3 (load-bearing).

They provide flexibility and adaptability to accommodate diverse privacy requirements and system constraints.

“They also provide flexibility and adaptability to accommodate diverse privacy requirements and system constraints.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Scalability could be affected due to heterogeneity among various types of edge devices, requiring additional efforts for compatibility.

“Scalability could be affected due to heterogeneity among various types of edge devices, which requires additional efforts for compatibility.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Model complexity and training capabilities are restricted due to the limited computational resources provided by edge devices.

“The model complexity and training capabilities are restricted due to the limited computational resources provided by the edge devices.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based). May combine two claims.

Security could set the alarm due to the vulnerability of edge devices, requiring robust security measures which governance is not provided by edge devices.

“Security could set the alarm due to the vulnerability of edge devices, requiring robust security measures which governance is not provided by edge devices.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

As data distribution and quality may vary among devices, privacy preservation can affect model performance, potentially impacting model generalization.

“As data distribution and quality may vary among devices, privacy preservation can affect model performance, potentially impacting model generalization.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Model performance could be affected by data imbalance among distributed datasets.

“Model performance could be affected by the data imbalance among the distributed datasets.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Transmitting only model updates and aggregated results reduces communication costs, saves network resources and minimizes latency.

“Via this technique, transmitting only model updates and aggregated results reduces communication costs, saves network resources and minimizes latency.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Limited computational resources on edge devices could compromise model accuracy.

“The limited computational resources on edge devices could compromise model accuracy.”

Probably not a substantial claim: may not pass S3 (load-bearing).

These approaches allow for fine-grained privacy control over different aspects of the FL process.

“These approaches allow for fine-grained privacy control over different aspects of the FL process.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

The decentralized architecture of intelligence on edge devices ensures preserving data confidentiality and compliance with regulations.

“Its decentralized architecture ensures preserving data confidentiality and compliance with regulations.”

Classified as probably not a substantial claim.

The first and most important strength of adopting intelligence on edge devices in FL systems lies in privacy enhancement.

“The first and most important strength of adopting intelligence on edge devices in FL systems lies in privacy enhancement.”

Classified as probably not a substantial claim. May combine two claims.

The integration of multiple privacy techniques may introduce new attack vectors or vulnerabilities and require fine-tuning of various configuration parameters.

“The integration of multiple privacy techniques may introduce new attack vectors or vulnerabilities and require fine-tuning of various configuration parameters.”

Classified as probably not a substantial claim.

The combination of privacy techniques may introduce a trade-off between privacy protection and model utility.

“The combination of privacy techniques may introduce a trade-off between privacy protection and model utility.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

The field of privacy enhancement technology in FL is still evolving, and standardization is limited.

“The field of privacy enhancement technology in FL is still evolving, and standardization is limited.”

Probably not a substantial claim: may not pass S4 (based).

DP provides data privacy by introducing noise that adversely affects the prediction accuracy of ML algorithms used in FL systems.

“As discussed before, the differential privacy technique provides data privacy by introducing noise that adversely affects the prediction accuracy of the ML algorithms used in FL systems.”

May combine two claims.

HE is very effective for FL privacy preservation because most operations required by FL systems, such as aggregation, model updates, and gradient calculations, can be performed without accessing raw data.

“This approach is very effective for FL privacy preservation as most of the operations required by the FL systems like aggregation, model updates, and gradient calculations can be performed without accessing the raw data [34,35,36,37,38].”

• cites [34] Xie, 2020, Federated Diabetes Mellitus Analysis via Homomorphic Encryption, Proc. J. Phys. Conf. Ser. IOP Publ. sourceNot checked yet

  Not checked in the free run.

• cites [35] Gandhi, N., Mishra, S., Bharti, S.K., and Bhagat, K. (2021, January 9–11). Leveraging towards Privacy-preserving using Federated Machine Learning for Healthcare Systems. Proceedings of the 2021 IEEE International Conference on Electronics, Computing and Communication Technologies (CONECCT), Bangalore, India. sourceNot checked yet

  Not checked in the free run.

• cites [36] Ji, J., Yan, D., and Mu, Z. (2022, January 15–17). Personnel status detection model suitable for vertical federated learning structure. Proceedings of the 2022 6th International Conference on Machine Learning and Soft Computing, Haikou, China. sourceNot checked yet

  Not checked in the free run.

• cites [37] Ma, 2021, Pocket diagnosis: Secure federated learning against poisoning attack in the cloud, IEEE Trans. Serv. Comput. sourceNot checked yet

  Not checked in the free run.

• cites [38] Hao, M., Li, H., Xu, G., Liu, Z., and Chen, Z. (2020, January 7–11). Privacy-aware and resource-saving collaborative learning for healthcare in cloud computing. Proceedings of the ICC 2020 IEEE International Conference on Communications (ICC), Dublin, Ireland. sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Existing aggregation techniques for FL, such as FedAvg, use randomized approaches to achieve differential privacy, but their performance may deteriorate in the presence of adversarial attacks.

“The basis existing aggregation techniques for FL, such as the Federated Averaging Algorithm (FedAvg), use randomized approaches to achieve differential privacy, but their performance may deteriorate in the presence of adversarial attacks.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Integrating multiple privacy techniques in a mixed or hybrid approach can increase the complexity of the FL system.

“However, integrating multiple privacy techniques in a mixed or hybrid approach can increase the complexity of the FL system.”

Probably not a substantial claim: may not pass S4 (based).

Privacy techniques such as encryption, secure multiparty computation, or differential privacy can introduce computational complexity and communication requirements.

“Privacy techniques such as encryption, secure multiparty computation, or differential privacy can introduce computational complexity and communication requirements.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Combining different privacy techniques may introduce compatibility and interoperability challenges.

“In addition, combining different privacy techniques may introduce compatibility and interoperability challenges.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

DP involves evaluations of privacy guarantees and existing privacy evaluation metrics are not designed for distributed systems like FL.

“DP involves evaluations of privacy guarantees and the existing privacy evaluation metrics are not designed for distributed systems like FL.”

Classified as probably not a substantial claim. May combine two claims.

Trade-offs arise due to the combination of different techniques and the complexity of integrating them, including Complexity vs. Effectiveness, Privacy vs. Utility, Performance vs. Privacy, Compatibility and Interoperability vs. Customization, Security vs. Complexity, Standardization vs. Flexibility.

“Thus, the trade-offs arise due to the combination of different techniques and the complexity of integrating them, including Complexity vs. Effectiveness, Privacy vs. Utility, Performance vs. Privacy, Compatibility and Interoperability vs. Customization, Security vs. Complexity, Standardization vs. Flexibility.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Homomorphic encryption can be computationally expensive, and performing computations on encrypted data requires significant computational resources.

“Homomorphic encryption can be computationally expensive, and performing computations on encrypted data requires significant computational resources.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The evolving blockchain technology creates the concept of a token economy where community revenue can be allocated to users who create value.

“The evolving blockchain technology creates the concept of a token economy where the community’s revenue can be allocated to the users who create value in the system [69].”

• cites [69] Lee, 2019, A decentralized token economy: How blockchain and cryptocurrency can revolutionize business, Bus. Horizons sourceNot checked yet

  Not checked in the free run.

May combine two claims.

HE provides a relevant and suitable option for FL systems because even cloud or edge servers performing aggregation cannot access the data and only clients have access to raw data.

“This encryption method provides a very relevant and suitable option for FL systems as even the cloud or edge servers performing the aggregation processes cannot access the data and only clients have access to the raw data.”

May combine two claims.

The combined technologies of FL and blockchain effectively enhance privacy, security, transparency, collaboration, and governance.

“The combined technologies effectively enhance privacy, security, transparency, collaboration, and governance.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

These approaches all require the second privacy-enhancing technique.

“These approaches all require the second privacy-enhancing technique.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

In a decentralized environment, blockchain enables a secure and tamper-proof process for data sharing and computation using encrypted public keys.

“In the decentralized environment, blockchain enables the secure and tamper-proof process for data sharing and computation using encrypted public keys.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Therefore participants' data at the local level are better protected in terms of privacy and security.

“Therefore the participants’ data at the local level are better protected in terms of privacy and security.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Intelligence on edge devices in FL systems brings effectiveness through improved privacy protection, enhanced data security, scalability, reduced network latency, real-time decision-making, fault tolerance and resource efficiency.

“Intelligence on the edge devices in FL systems brings effectiveness through improved privacy protection, enhanced data security, scalability, reduced network latency, real-time decision-making, fault tolerance and resource efficiency.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

Errors occurring in data transmission or training outcomes can erode the trust in the FL system.

“Lastly, errors occurring in data transmission or training outcomes can erode the trust in the FL system.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

Mixed, hybrid, and miscellaneous approaches can be utilized in FL systems where complementary privacy techniques, fine-grained privacy control, balancing privacy and utility, flexibility and adaptability, robust defence against attacks, and compatibility and interoperability are needed.

“While the mixed, hybrid, and miscellaneous approaches can be utilized in FL systems where complementary privacy techniques, fine-grained privacy control, balancing privacy and utility, flexibility and adaptability, robust defence against attacks, and compatibility and interoperability are needed.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

During the aggregation and transmission phases, sensitive patient data face the risks of data breaches, unauthorized access, identity theft, or even malicious modification.

“During the aggregation and transmission phases, sensitive patient data face the risks of data breaches, unauthorized access, identity theft, or even malicious modification.”

• AssumptionRests on an assumption: Patient data is sensitive, confidential and private.

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Approaches like DP and HE provide significant privacy enhancements but add computational and governance overhead.

“Approaches like DP and HE provide significant privacy enhancements but add computational and governance overhead.”

Probably not a substantial claim: may not pass S4 (based).

P2PS can be used in FL systems where distributed computing power, scalability, privacy, fault tolerance, and geographical distribution are required.

“The P2PS can be used in FL systems where distributed computing power, scalability, privacy, fault tolerance, and geographical distribution are required.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Evaluating the effectiveness and security of mixed approaches is challenging as they may require specialized evaluation methodologies and benchmarks.

“Evaluating the effectiveness and security of these approaches is challenging as they may require specialized evaluation methodologies and benchmarks to assess their impact on privacy, utility, and overall system performance.”

Probably not a substantial claim: may not pass S2 (checkable).

The inherent characteristics of FL architecture and additional privacy enhancement techniques still leave the intelligence builder facing ethical challenges.

“Although the inherent characteristics of FL architecture and additional privacy enhancement techniques, this intelligence builder faces ethical challenges.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

More decentralized approaches like blockchain and P2PS provide robustness, scalability and fault tolerance with limited privacy guarantees.

“On the other hand, more decentralized approaches like blockchain and P2PS provide robustness, scalability and fault tolerance with limited privacy guarantees.”

Probably not a substantial claim: may not pass S4 (based).

The effectiveness of mixed, hybrid, and miscellaneous approaches to privacy enhancement in an FL system depends on several factors, including the specific techniques employed, the nature of the data, and the use case.

“The effectiveness of mixed, hybrid, and miscellaneous approaches to privacy enhancement technology in an FL system depends on several factors, including the specific techniques employed, the nature of the data, and the use case.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Biased data could affect FL training outcomes.

“Furthermore, biased data could affect FL training outcomes.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

FL opens further opportunities in novel research and business avenues by providing security and privacy for user data.

“FL opens further opportunities in the novel research and business avenues by providing security and privacy for user data.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

FL streamlines the healthcare process and improves healthcare decisions and diagnoses.

“This streamlines the healthcare process and improves healthcare decisions and diagnoses.”

May combine two claims.

Mixed approaches can contribute to improved data confidentiality by encrypting data or performing computations without exposing raw data, and can reduce the risk of data leakage or unauthorized access.

“These approaches can contribute to improved data confidentiality by encrypting data or performing computations without exposing the raw data, these approaches can reduce the risk of data leakage or unauthorized access.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

The underlying decentralized architecture for P2PS in FL is a complex and dynamic process.

“In addition, although the decentralized architecture for P2PS in FL is very important, the underlying architecture is a complex and dynamic process that needs to be optimized.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

Intelligence on the edge device can play a big role in improving privacy and efficiency in FL.

“To summarise, we emphasise that intelligence on the edge device can play a big role in improving privacy and efficiency in FL.”

May combine two claims.

Mixed approaches allow for optimizing the trade-off between privacy and utility in FL, ensuring that privacy enhancement does not compromise effectiveness and utility.

“They allow for optimizing the trade-off between privacy and utility in FL, ensuring that privacy enhancement does not compromise the effectiveness and utility of the FL system.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The implications of using P2PS technology in FL highlight the potential for improved privacy, enhanced collaboration, and distributed computing power.

“The implications of using P2PS technology in FL highlight the potential for improved privacy, enhanced collaboration, and distributed computing power.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

P2PS is able to provide high privacy guarantees, improved data availability, distributed computing power, geographic distribution, collaborative learning, better scalability, and increased fault tolerance.

“P2PS is able to provide high privacy guarantees, improved data availability, distributed computing power, geographic distribution, collaborative learning, better scalability, and increased fault tolerance.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

The innovative FL applications in healthcare systems are still at a very early stage.

“The innovative FL applications in healthcare systems are still at a very early stage.”

Probably not a substantial claim: may not pass S4 (based).

Intelligence on edge devices can be applied in FL systems where privacy, reduced communication, real-time processing, robustness, energy efficiency, and scalability are critical factors.

“For intelligence on edge devices, they can be applied in FL systems where privacy, reduced communication, real-time processing, robustness, energy efficiency, and scalability are critical factors.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

Consensus mechanisms are a suitable solution for the P2PS network to reach agreement on data or models.

“Besides, consensus mechanisms are a suitable solution for the P2PS network to reach agreement on data or models.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

While mixed approaches offer the potential for improved privacy protection, their effectiveness can vary.

“While these approaches offer the potential for improved privacy protection, their effectiveness can vary.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

The effectiveness of P2PS technology in an FL system depends on the specific implementation, network characteristics, security measures, and governance mechanisms in place.

“The effectiveness of P2PS technology in an FL system depends on the specific implementation, network characteristics, security measures, and governance mechanisms in place.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The integration of multiple privacy techniques could increase computational complexity, interoperability, and standardization challenges.

“However, the integration of multiple privacy techniques could increase computational complexity, interoperability, and standardization challenges.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Intelligence on edge devices in FL systems holds significant relevance in various contexts and domains, including scalability, real-time processing, bandwidth optimization, energy efficiency, and offline functionality.

“Besides privacy enhancement, intelligence on edge devices in FL systems holds significant relevance in various contexts and domains, including but not limited to scalability, real-time processing, bandwidth optimization, energy efficiency, and offline functionality.”