Deploying Large AI Models on Resource-Limited Devices with Split Federated Learning

2025-04-12

AI Summary

To address the challenges of resource constraints, privacy sensitivity, and high communication overhead on edge devices, this paper proposes SFLAM—a quantized split federated fine-tuning framework. SFLAM is the first to deeply integrate split learning, model quantization, adaptive power control, and dynamic bandwidth allocation into a unified edge–cloud collaborative training paradigm. We theoretically model and characterize the fundamental latency–energy trade-off boundary in edge training. Experimental results demonstrate that, compared to baseline methods, SFLAM reduces memory footprint by 62%, cuts communication latency by 47%, improves energy efficiency by 3.1×, and scales to over one thousand edge devices. This work provides a systematic solution for lightweight, privacy-preserving, and energy-efficient large-model fine-tuning at the edge.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Search and Optimization: Learning to Search
• Natural Language Processing: (Large) Language Models

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Systems and Infrastructure for Web, Mobile and WoT: Federated Web and WoT systems, including distributed, federated and edge-based data processing
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Large Artificial Intelligence Models (LAMs) powered by massive datasets, extensive parameter scales, and extensive computational resources, leading to significant transformations across various industries. Yet, their practical deployment on resource-limited mobile edge devices is hindered by critical challenges such as data privacy, constrained resources, and high overhead costs. Addressing this gap, this paper proposes a novel framework, named Quantized Split Federated Fine-Tuning Large AI Model (SFLAM). By partitioning the training load between edge devices and servers using a split learning paradigm, SFLAM can facilitate the operation of large models on devices and significantly lowers the memory requirements on edge devices. Additionally, SFLAM incorporates quantization management, power control, and bandwidth allocation strategies to enhance training efficiency while concurrently reducing energy consumption and communication latency. A theoretical analysis exploring the latency-energy trade-off is presented, and the framework's efficacy is validated via comprehensive simulations. The findings indicate that SFLAM achieves superior performance in terms of learning efficiency and scalability compared to conventional methods, thereby providing a valuable approach for enabling advanced AI services in resource-constrained scenarios.

Problem

• Enabling large AI models on resource-limited edge devices
• Reducing memory and computational overhead for edge deployment
• Optimizing energy and latency in federated learning systems

Innovation

• Split learning paradigm for edge-server load partitioning
• Quantization management to reduce memory requirements
• Power and bandwidth optimization for efficiency