Integrating Federated Transfer Learning for Secure Multi-Tenant Data Management in Decentralized Cloud Infrastructures

Abstract

Ensuring secure multi-tenant data management while enabling cross-domain knowledge sharing is challenging in decentralized cloud infrastructures. Traditional centralized learning methods pose risks like data leakage and non-compliance with privacy regulations. To address these concerns, this research integrates Federated Transfer Learning (FTL) for secure and efficient multi-tenant data management. The proposed approach employs Federated Learning (FL) to enable collaborative model training while keeping raw data localized, preserving privacy. Additionally, BERT-based transfer learning improves knowledge sharing by adapting pre-trained models to tenant-specific tasks, enhancing efficiency and reducing computational overhead. A tenant-aware security mechanism dynamically assesses trust levels to ensure secure workload allocation and mitigate risks. Furthermore, a decentralized aggregation strategy enhances data privacy and prevents single-point failures, improving system robustness. The framework was evaluated using real-world datasets, assessing privacy, adaptability, communication overhead, and computational efficiency. Experimental results demonstrate that FTL-driven decentralized architectures achieve low latency (6ms), high throughput (850000 requests/sec at 10000 tenants), better utilization of resources (83%), and good security compliance (9.5/10). Consistency models also come with a 50% overhead for strong consistency, which shows the trade-offs involved in ensuring data consistency. The results confirm the proposed model as a scalable, efficient, and privacy-preserving solution for multi-tenant cloud environments.