What question did this study set out to answer?

This research aims to develop an efficient model for dynamic resource allocation in 6G networks using federated learning techniques.

March 5, 2026Open Access

Federated Learning for Dynamic Resource Allocation in 6G Network Slicing

RKR. Krishnamoorthy CBC. N. Gireesh BabuInstitute of Technology Management PGPratik GiteInstitute of Chemical Technology

Key Points

This research aims to develop an efficient model for dynamic resource allocation in 6G networks using federated learning techniques.
Introduced Federated Learning-based Dynamic Resource Allocation Model (FL-DRAM)
Utilized reinforcement learning (RL) and hierarchical federated learning (HFL)
Implemented accuracy evaluation using a Wireless Network Slicing Dataset from Kaggle
Developed using mruby for efficient performance
Evaluated with security mechanisms including Blockchain and Differential Privacy
PerFedRL achieved an accuracy of 97.23%, outperforming conventional methods
AFUM reduced training overhead by 35%
Dynamic resource allocation resulted in 95.32% bandwidth utilization efficiency
Energy savings rate reached 47.84%
Adversarial robustness improved by 99.12% through privacy mechanisms

Abstract

The advancement of 6G networks necessitates efficient resource allocation to maintain optimal Quality of Service (QoS) across dynamically managed network slices. Traditional centralized approaches to network resource allocation pose significant scalability, privacy, and computational bottlenecks. This study introduces a Federated Learning-based Dynamic Resource Allocation Model (FL-DRAM) that integrates reinforcement learning (RL) and hierarchical federated learning (HFL) to optimize bandwidth, latency, and reliability across diverse network slices. The proposed model leverages a Wireless Network Slicing Dataset from Kaggle, featuring key QoS parameters such as latency, jitter, throughput, and packet loss. The execution of the proposed model is implemented using mruby, a lightweight Ruby implementation designed for efficient performance in constrained environments. Experimental results demonstrate that the Personalized Federated Learning Model (PerFedRL) achieves an accuracy of 97.23%, outperforming conventional approaches such as FedAvg (90.45%) and FedProx (92.38%). The Adaptive Federated Update Mechanism (AFUM) enhances convergence time, reducing training overhead by 35%, while reinforcement learning-based dynamic resource allocation achieves bandwidth utilization efficiency of 95.32% and an energy savings rate of 47.84%. Additionally, security mechanisms such as Blockchain-Enabled Federated Learning (BFL) and Differential Privacy (DP) improve adversarial robustness by 99.12%, ensuring enhanced data protection. This research validates the efficacy of federated learning in 6G resource optimization, demonstrating significant improvements in latency, resource utilization, and computational efficiency. The study highlights the potential of hierarchical federated learning models to drive intelligent, scalable, and privacy-preserving resource management in next-generation networks. In order to provide adaptive, scalable, and privacy-preserving resource allocation for dynamic 6G network slicing, this work presents FL-DRAM, a hierarchical federated learning system closely associated with RL. In contrast to previous HFL + RL systems, FL-DRAM provides system-wide QoS and security benefits by integrating PerFedRL personalization with AFUM asynchronous updates, as shown in a new 6G Digital Twin configuration.

Ask AI

Helpful

Bookmark

View Full Paper