Existing differentially private federated learning methods often cause significant degradation

The gap

Evidence profile

Sourced from the stated research gap and limitations and future work of the source papers, classified as general, spanning 3 journals. Those papers have been cited 1 times in total.

Supporting evidence — 3 representative gaps

• AIGU-DPFL: Adaptive Differentially Private Federated Learning with Importance-Based Gradient Updates (2026) · Computers · cited 1× · doi

  Existing differentially private federated learning methods often cause significant degradation in model performance due to the noise required to satisfy differential privacy guarantees. Fixed noise-injection strategies following gradient clipping overlook the varying importance of different gradient dimensions. There is a need for an adaptive approach that can dynamically control noise intensity to satisfy differential privacy guarantees.

  generalstated research gap
  Keywords: existing differentially private federated learning methods often cause
• Exploiting explanations for model extraction via knowledge distillation and mitigation with private counterfactuals (2026) · Frontiers in Artificial Intelligence · doi

  We discuss the contribution, limitations of our current approach and potential future research directions: • Attack Scope: This work is limited to MEA that exploit CFs. Other types of privacy attacks, such as MIA and inversion, are not examined in this paper and therefore fall outside the scope of our current analysis. A future direction is to evaluate whether the proposed mitigation strategy remains effective against this broader class of privacy attacks. • Privacy-Performance Trade-off: Integrating DP into the CF generator mitigates MEA, but it also introduces limitations including proximity, by degrading explanation quality, prediction gain, and plausibility in some cases. As future work, a more systematic study of the privacy-utility trade-off is needed, including evaluating a wider range of privacy budgets and optimization settings for the DP-based CF generator to better balance protection and explanation quality. • Focus on Deep Learning Applications: This work focuses on MEA for DNNs, where knowledge distillation is particularly effective due to the expressive capacity of DNNs and their ability to learn rich data representations. This focus limits the applicability of our findings to other model families, such as traditional baselines including tree-based or ensemble Frontiers in Artificial Intelligence 15 frontiersin.org Ezzeddine et al. 10.3389/frai.2026.1746910 methods. Since KD proved effective in the DNN setting, a natural future direction is to explore how KD-based extraction strategies can be adapted to non-DNN algorithms and whether similar performance gains can be achieved.

  generallimitationsevidence 5/5
  Keywords: privacy future ective including based limitations current scope attacks direction whether performance trade generator explanation
• SecureFedShield: An Adaptive Privacy-Preserving Federated Defense Framework Against Adversarial Attacks in Financial Fraud Detection (2026) · International Journal of Creative and Open Research in Engineering and Management · doi

  References This paper presented SecureFedShield, an adaptive privacy-preserving federated learning framework for secure financial fraud detection under adversarial environments. The proposed framework integrates adaptive privacy protection, continuous trust-based client evaluation, adversarial update detection, and a unified aggregation trust-aware collaborative learning architecture. Unlike conventional federated independently address privacy preservation or adversarial robustness, complementary SecureFedShield mechanisms to enhance both model security and predictive performance. learning approaches combines robust these into that [1] H. B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, "Communication- Efficient Learning of Deep Networks from Decentralized Data," in Proc. 20th Int. Conf. Artificial Intelligence and Statistics (AISTATS), Fort Lauderdale, FL, USA, 2017, pp. 1273–1282. [2] C. Dwork, "Differential Privacy," in Proc. 33rd Int. Colloquium on Automata, Languages and Programming (ICALP), Venice, Italy, 2006, pp. 1–12. © 2026 The Author(s). Published by IJCOPE Journal. Website: https://ijcope.org/ 16 International Journal of Creative and Open Research in Engineering and Management ISSN: 3108-1754 (Online) Volume 02 Issue 08 Augustl-2026 | Impact Factor: 3.5 [3] C. Dwork and A. Roth, The Algorithmic Foundations of Differential Privacy. Boston, MA, USA: Now Publishers Inc., 2014. [4] P. Kairouz et al., "Advances and Open Problems in Federated Learning," Foundations and Trends® in Machine Learning, vol. 14, no. 1–2, pp. 1–210, 2021. [5] K. Bonawitz et al., "Practical Secure Aggregation for Privacy-Preserving Machine Learning," in Proc. ACM SIGSAC Conf. Computer and Communications Security (CCS), Dallas, TX, USA, 2017, pp. 1175– 1191. [6] B. Hitaj, G. Ateniese, and F. Perez-Cruz, "Deep Models Under the GAN: Information Leakage from Collaborative Deep Learning," in Proc. ACM SIGSAC Conf. Computer and Communications Security (CCS), Dallas, TX, USA, 2017, pp. 603–618. [7] M. Nasr, R. Shokri, and A. Houmansadr, "Comprehensive Privacy Analysis of Deep Learning: Passive and Active White-Box Inference Attacks against Centralized and Federated Learning," in Proc. IEEE Symp. Security and Privacy (SP), San Francisco, CA, USA, 2019, pp. 739–753. [8] L. Zhu, Z. Liu, and S. Han, "Deep Leakage from Gradients," in Advances in Neural Information Processing Systems (NeurIPS 2019), Vancouver, BC, Canada, 2019. [9] J. Geiping, H. Bauermeister, H. Dröge, and M. Moeller, "Inverting Gradients – How Easy Is It to Break Privacy in Federated Learning?" in Advances in Neural Information Processing Systems (NeurIPS 2020), Vancouver, BC, Canada, 2020. [10] L. Melis, C. Song, E. De Cristof

  generalfuture workevidence 5/5
  Keywords: learning privacy federated deep proc security adversarial conf advances information securefedshield adaptive preserving framework secure

Questions about this gap