SECURED DATA SHARING IN FEDERATED LEARNING FOR CRYPTOGRAPHICS TECHNIQUES

This review systematically analyzed the three cornerstonetechniquesforprivacyinfederatedlearning (FL): Homomorphic Encryption (HE), Secure Multi- Party Computation (SMPC), and Differential Privacy (DP). Our analysis highlights a fundamental trade-off, referred to as the “Practitioner’s Dilemma”, between security and usability. Cryptographic methods like HE and SMPC offer high model accuracy and strong developer guarantees but introduce significant computational and communication overhead, negativelyimpactingend-user experience.Conversely, statisticalapproachessuchasLocalDifferentialPrivacy (LDP)providelightweight, low-overheadsolutionsbut degrade model accuracy, creating a noisy training process. The review demonstrates that there is no one- size-fits-all solution; instead, selecting a privacy- preserving FL strategy requires careful, application- specifictrade-offsbetweenperformance,accuracy,and security.   Moreover,thestudyunderscoresthatsecuringFLis a socio-technical challenge that demands interdisciplinary collaboration. Cryptographers, machinelearningengineers, andprivacytheorists must align their objectives to balance theoretical guarantees with practical efficiency. In high-stakes applications like medical diagnosis, accuracy may outweigh computational cost, while in low-stakes scenarios, efficiency may take priority. Despite these challenges, rapid progress in HE, SMPC, and DP shows a promising path forward. By synthesizing existing techniques and clearly articulating the trade-offs, this review provides a framework for practitioners and researcherstobuildtrustworthy,privacy-preservingFL systems, enabling responsible deployment of collaborativeAIwhileguidingfutureinnovationsinthe field.   Future research in privacy-preserving federated learningshouldfocus onaddressingthe inherent trade- offsbetweenmodelaccuracy,computationalefficiency, anddata confidentiality.Onepromisingdirectionisthe development of hybrid cryptographic frameworks that combineHomomorphicEncryption,SecureMulti-Party Computation, and Differential Privacy to achieve stronger privacy guarantees while reducing computational and communication overhead. Another important area is dynamic and adaptive privacy mechanisms, which can tailor privacy levels based on the sensitivity of data, application requirements, or resource constraints of client devices. Scalability and real-timeperformanceinlarge-scale,heterogeneousFL networks also remain significant challenges, particularly in IoT and edge computing scenarios. Additionally, research should explore cross- disciplinary approaches, fostering collaboration between cryptographers, machine learning engineers, and domain experts to create protocols that are both theoreticallyrobustandpracticallydeployable.Finally, developing standardized benchmarks and evaluation frameworks for secure federated learning will enable objectivecomparisonof privacy-preservingtechniques andguidethedesignofmoreefficient,trustworthy,and application-specific FL systems.

1.      S.Zhou,L.Wang,L.Chen,Y.WangandK. Yuan, “Group verifiable secure aggregate federated learning based on secret sharing,” Scientific Reports, vol. 15, 2025. 2.      W. Yao, T. Zhou, Y. Han and X. Wang, “Verifiable secure aggregation scheme for privacy protection in federated learning networks,” Discover Computing, vol. 28, 2025. 3.      Y.Zhang, W. Zhang, Y. Ke and C. Shen, “VeCoSA-FI: A verifiable and collusion- resistant secure aggregation protocol in federated learning with fair incentive mechanism,” Journal of King Saud University – Computer and Information Sciences, 2025. 4.      X.Ma, “A secure aggregation method for federated learning based on homomorphic encryption,”AdvancesinEngineeringIn[5] X.Zhang,Y.LuoandT.Li,“Areviewof researchonsecureaggregationfor federated learning,” Future Internet, vol. 17, no. 7, 2025innovation,vol.16,no.2,pp.80-85, 2025. 5.      X. Zhang, Y. Luo and T. Li, “A review of researchonsecureaggregationfor federated learning,” Future Internet, vol. 17, no. 7, 2025. 6.      R. Taiello, C. Gritti, M. Önen and M. Lorenzi, “Buffalo: A practical secure aggregation protocol for buffered asynchronous federated learning,” ACM CODASPY, 2025. 7.      J.Li, H. Chen and Q. Wang, “Efficient verifiable secure aggregation protocols for federated learning,” Journal of Information Security and Applications, 2025. 8.      Y.Liu,Z.ChenandJ.Wu,“Aneffectiveand verifiable secure aggregation scheme with privacy preserving for federated learning,” Journal of Systems Architecture, vol. 161, 2025. 9.      D.Commey and G. V. Crosby, “PQS-BFL: A post-quantum secure blockchain-based federated learning framework,” arXiv preprint, 2025. 10.   I. Emmaka and T. V. X. Phuong, “One-shot secure aggregation: A hybrid cryptographic protocol for private federated learning in IoT,” arXiv preprint, 2025.