Collaborative LLM Numerical Reasoning with Local Data Protection

2025-04-01

AI Summary

To address the challenge of performing complex numerical reasoning on resource-constrained edge devices while ensuring strict data locality and privacy, this paper proposes a privacy-enhanced collaborative inference framework. Methodologically, it introduces (1) a context-aware query-domain migration synthesis strategy that transforms sensitive numerical queries into semantically anonymized yet logically equivalent counterparts executable remotely, and (2) a toolchain-based answer reconstruction mechanism at the edge, integrating prompt engineering, code generation, logical constraint modeling, and differential-privacy-inspired data abstraction to recover high-fidelity results locally. Evaluated across multiple benchmarks, the framework achieves 16.2–43.6% improvements in inference accuracy and reduces data leakage risk by 2.3–44.6%, significantly outperforming state-of-the-art collaborative inference and privacy-preserving approaches.

Technology Category

• Machine Learning: Privacy
• Natural Language Processing: Safety and Robustness
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

Application Category

• Security and Privacy: Data transparency and provenance
• Responsible Web: Data and user privacy-enhancing technologies for the Web
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Numerical reasoning over documents, which demands both contextual understanding and logical inference, is challenging for low-capacity local models deployed on computation-constrained devices. Although such complex reasoning queries could be routed to powerful remote models like GPT-4, exposing local data raises significant data leakage concerns. Existing mitigation methods generate problem descriptions or examples for remote assistance. However, the inherent complexity of numerical reasoning hinders the local model from generating logically equivalent queries and accurately inferring answers with remote guidance. In this paper, we present a model collaboration framework with two key innovations: (1) a context-aware synthesis strategy that shifts the query domains while preserving logical consistency; and (2) a tool-based answer reconstruction approach that reuses the remote-generated problem-solving pattern with code snippets. Experimental results demonstrate that our method achieves better reasoning accuracy than solely using local models while providing stronger data protection than fully relying on remote models. Furthermore, our method improves accuracy by 16.2% - 43.6% while reducing data leakage by 2.3% - 44.6% compared to existing data protection approaches.

Problem

• Enhancing numerical reasoning for local models with data protection
• Reducing data leakage while using remote model assistance
• Improving accuracy and privacy in collaborative LLM frameworks

Innovation

• Context-aware synthesis for logical consistency
• Tool-based answer reconstruction with code
• Balanced local and remote model collaboration