Taming Noise-Induced Prototype Degradation for Privacy-Preserving Personalized Federated Fine-Tuning

TL;DR

This paper introduces VPDR, a client-side privacy mechanism for ProtoPFL that adaptively perturbs prototypes based on discriminative feature variance, improving privacy-utility balance.

Contribution

We propose VPP and DCR methods that allocate noise adaptively and regularize feature norms, enhancing privacy preservation in federated learning without losing discriminability.

Findings

VPDR outperforms isotropic Gaussian perturbation in privacy-utility trade-offs.

Theoretical analysis confirms privacy guarantees are at least as strong as baseline.

Experiments show improved robustness and semantic separability in multi-domain benchmarks.

Abstract

Prototype-based Personalized Federated Learning (ProtoPFL) enables efficient multi-domain adaptation by communicating compact class prototypes, but directly sharing them poses privacy risks. A common defense involves per-example $\ell_2$ clipping before prototype computation to bound sensitivity, followed by isotropic Gaussian noise to enforce Local Differential Privacy (LDP). However, Isotropic Gaussian Prototype Perturbation (IGPP) typically over-perturbs discriminative dimensions and struggles to balance the clipping threshold with representation fidelity. In this paper, we propose VPDR, a client-side privacy plug-in that seamlessly integrates into existing ProtoPFLs. Motivated by the observation that dimension-wise class variance reflects discriminability, we introduce Variance-adaptive Prototype Perturbation (VPP), which allocates less noise to discriminative subspaces, preserving semantic separability while ensuring privacy. We further develop Distillation-guided Clipping Regularization (DCR), which enables feature norms to adaptively concentrate near the predefined clipping threshold while maintaining prediction consistency. Theoretical analysis shows that our groupwise mechanism provides privacy guarantees no weaker than the isotropic baseline under the same privacy constraints. Extensive experiments on multi-domain benchmarks demonstrate that VPDR achieves a superior privacy-utility trade-off, outperforming IGPP in personalized federated fine-tuning without sacrificing robustness against realistic attacks.