Title: TITAN-FedAnil+: A Trust-Centric Adaptive Framework for Blockchain-Integrated Federated Learning in Resource-Limited Smart Enterprises

Abstract:

Federated Learning (FL) has gained traction as a robust mechanism for fostering collaborative intelligence while safeguarding data privacy. Despite its promise, the field faces persistent hurdles, notably data heterogeneity stemming from non-independent and identically distributed (non-IID) datasets and vulnerabilities associated with decentralized security. These issues are particularly acute within enterprise settings characterized by limited computational resources.

To address these challenges, this study introduces TITAN-FedAnil+, a novel Trust-Based Adaptive Network designed to enhance blockchain-enabled federated learning for intelligent enterprises. The core innovation of this framework lies in its use of affinity propagation-based adaptive clustered aggregation. This technique allows the system to detect and filter out malicious model updates dynamically, without the need for prior assumptions regarding the quantity of attackers.

Furthermore, the framework leverages GPU-accelerated vectorization to significantly boost computational performance. It also incorporates a signed state jump mechanism, which facilitates lightweight resynchronization within the blockchain infrastructure.

Experimental evaluations highlight the framework's efficiency, revealing substantial decreases in memory consumption. Specifically, TITAN-FedAnil+ achieved memory savings of up to 81% over 50 communication rounds when tested on constrained edge devices with 8 GB of memory, outperforming the baseline framework. These findings suggest that TITAN-FedAnil+ offers a viable solution for enhancing robustness, scalability, and resource efficiency in secure federated learning applications tailored for intelligent enterprise environments.

Source: arXiv Generated at: 2026-06-04 00:00:00 UTC