GQ-FSL: Green Quantized Federated Split Learning

2026-07-31

AI Summary

This work addresses the challenges of deploying deep neural networks (DNNs) at the wireless edge, where device energy budgets and resource constraints severely limit performance. While existing federated split learning approaches mitigate some limitations, they still incur substantial communication and computational overhead. To overcome this, we propose GQ-FSL, a green quantized federated split learning framework that introduces asymmetric stochastic quantization into federated split learning for the first time, enabling heterogeneous precision configurations for client and server submodels. By jointly modeling energy consumption and convergence bounds, GQ-FSL decouples device-level energy usage from global model performance and optimizes both the split point and quantization precision. Experimental results demonstrate that, under target accuracy requirements, GQ-FSL significantly reduces total system energy consumption compared to quantized federated learning and full-precision federated split learning, enabling efficient deployment of large-scale DNNs on resource-constrained edge devices.

Abstract

Deploying state-of-the-art deep neural networks (DNNs) at the wireless edge is severely bottlenecked by the strict energy and resource constraints of mobile devices. While federated split learning (FSL) mitigates on-device computation by offloading workloads to an edge server, this may introduce systemic overheads, while the continuous exchange of cut-layer data, and submodels still incurs significant energy consumption (EC). To address this, we propose a green quantized FSL (GQ-FSL) framework that incorporates stochastic quantization for both local collaborative training and wireless transmissions. Notably, GQ-FSL supports asymmetric precision levels for the client- and server-side submodels, effectively decoupling device energy constraints from global convergence degradation. To quantify these tradeoffs, we develop parameterized energy models for the split architecture and derive a theoretical convergence bound under statistically heterogeneous data. Building on that, we formulate a joint optimization problem to configure the DNN split point and precision levels, minimizing the total system EC while satisfying a strict target accuracy constraint. Ultimately, we demonstrate that GQ-FSL enables large-scale DNN deployment on resource-constrained devices, achieving superior energy efficiency compared to quantized federated learning and full-precision FSL.

Problem

• energy consumption
• federated split learning
• resource-constrained devices
• wireless edge
• deep neural networks

Innovation

• quantized federated learning
• split learning
• asymmetric precision
• energy efficiency
• stochastic quantization