# FLAG: Programmable In-Network Aggregation for Communication-Aware Federated Learning in 5G RANs

- **Authors**: Emilio Paolini, Andrea Pinto, Luca Valcarenghi, and Flavio Esposito
- **Venue**: IEEE Transactions on Network and Service Management (TNSM), 2026
- **IEEE Xplore**: [View on IEEE Xplore](https://ieeexplore.ieee.org/document/11536094)

---

## Abstract

Federated Learning (FL) enables collaborative model training without sharing raw data, making it attractive for privacy-preserving applications at the wireless edge. However, when executed over real 5G networks, FL performance degrades due to uplink congestion, heterogeneous client capabilities, and intermittent connectivity. Most existing approaches attempt to mitigate these issues indirectly by optimizing clients (through adaptive participation, local training, or selection strategies) or by optimizing models (via pruning, quantization, or compression), but they ignore potential network bottlenecks.

To address this challenge, we introduce **FLAG 🏁**, a framework that embeds line-rate model aggregation directly into the 5G gNodeB (gNB) data plane using programmable switches and P4. By offloading parameter aggregation to the radio access network infrastructure, FLAG cuts uplink traffic in half, reduces latency, and speeds up time-to-accuracy. Furthermore, we design loss-tolerant aggregation and deadline-aware scheduling mechanisms that ensure model convergence even in the presence of packet loss or straggler clients.

---

## Highlights

- **Network-Active Learning**: Integrates communication and machine learning training loops rather than treating them in isolation.
- **5G gNB Data Plane Acceleration**: Embeds parameter aggregation directly at line-rate into 5G access point network elements.
- **Resiliency & Scheduling**: Deadline-aware scheduling and packet loss recovery mechanisms designed specifically for wireless edge deployments.