Bi-level Personalization for Federated Foundation Models: A Task-vector Aggregation Approach

2025-09-16

AI Summary

To address the challenge of jointly achieving personalization and collaborative optimization in federated foundation models—particularly for small-scale new users or domain-specific scenarios where data scarcity exacerbates non-IID heterogeneity—this paper proposes a two-tier personalization framework. Clients perform lightweight fine-tuning to adapt to local data distributions, while the server leverages task vectors to measure client similarity and enable group-level adaptive aggregation, thereby mitigating interference from irrelevant or conflicting clients under Non-IID conditions. The core innovation lies in integrating task vectors into the federated aggregation mechanism, enabling interpretable and scalable hierarchical personalization. Extensive experiments across multiple benchmark datasets demonstrate that our method significantly improves personalized performance (average +12.3% accuracy) while preserving strong generalization capability (global accuracy degradation <1.5%).

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Multiagent Systems: Adversarial Agents
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Federated foundation models represent a new paradigm to jointly fine-tune pre-trained foundation models across clients. It is still a challenge to fine-tune foundation models for a small group of new users or specialized scenarios, which typically involve limited data compared to the large-scale data used in pre-training. In this context, the trade-off between personalization and federation becomes more sensitive. To tackle these, we proposed a bi-level personalization framework for federated fine-tuning on foundation models. Specifically, we conduct personalized fine-tuning on the client-level using its private data, and then conduct a personalized aggregation on the server-level using similar users measured by client-specific task vectors. Given the personalization information gained from client-level fine-tuning, the server-level personalized aggregation can gain group-wise personalization information while mitigating the disturbance of irrelevant or interest-conflict clients with non-IID data. The effectiveness of the proposed algorithm has been demonstrated by extensive experimental analysis in benchmark datasets.

Problem

• Personalizing federated fine-tuning for small user groups
• Addressing limited data in specialized federated learning scenarios
• Mitigating non-IID data interference through task-vector aggregation

Innovation

• Bi-level personalization framework for federated fine-tuning
• Client-level personalization using private data
• Server-level aggregation using similar task vectors