LSRP: A Leader-Subordinate Retrieval Framework for Privacy-Preserving Cloud-Device Collaboration

TL;DR

This work tackles privacy-preserving cloud–device collaboration by addressing underutilization of cloud LLM problem-solving and limited on-device data integration. It introduces LSRP, a leader–subordinate framework where the cloud LLM acts as a dynamic leader guided by U-U-RAG, and the on-device SLM serves as the privacy-preserving subordinate augmented by SMFB-DPO feedback. Key innovations include the U-U-RAG mechanism for task-specific leader strategy selection and the SMFB-DPO loop that refines cloud guidance using on-device feedback, with NSGA-II used to balance multiple quality metrics. Empirical results on CoGenis and Movie Explain datasets show substantial improvements in QA relevance and personalization over baselines, demonstrating the approach’s practical potential for private, personalized AI services. The work advances the state of cloud–device collaboration by combining dynamic guidance with on-device data advantages while maintaining privacy, and provides a reproducible framework with detailed evaluation metrics.

Abstract

Cloud-device collaboration leverages on-cloud Large Language Models (LLMs) for handling public user queries and on-device Small Language Models (SLMs) for processing private user data, collectively forming a powerful and privacy-preserving solution. However, existing approaches often fail to fully leverage the scalable problem-solving capabilities of on-cloud LLMs while underutilizing the advantage of on-device SLMs in accessing and processing personalized data. This leads to two interconnected issues: 1) Limited utilization of the problem-solving capabilities of on-cloud LLMs, which fail to align with personalized user-task needs, and 2) Inadequate integration of user data into on-device SLM responses, resulting in mismatches in contextual user information. In this paper, we propose a Leader-Subordinate Retrieval framework for Privacy-preserving cloud-device collaboration (LSRP), a novel solution that bridges these gaps by: 1) enhancing on-cloud LLM guidance to on-device SLM through a dynamic selection of task-specific leader strategies named as user-to-user retrieval-augmented generation (U-U-RAG), and 2) integrating the data advantages of on-device SLMs through small model feedback Direct Preference Optimization (SMFB-DPO) for aligning the on-cloud LLM with the on-device SLM. Experiments on two datasets demonstrate that LSRP consistently outperforms state-of-the-art baselines, significantly improving question-answer relevance and personalization, while preserving user privacy through efficient on-device retrieval. Our code is available at: https://github.com/Applied-Machine-Learning-Lab/LSRP.

Figures (5)

• Figure 1: Illustration of the limitations of the device-cloud framework in privacy-preserving tasks.
• Figure 2: Main framework of LSRP.
• Figure 3: Experiment on different $k$ settings.
• Figure 4: Experiment on retrieval on-device and on-cloud settings.
• Figure 5: Experiment on four leader strategies and LSRP before and after SMFB-DPO.