C2A: Client-Customized Adaptation for Parameter-Efficient Federated Learning

TL;DR

This paper proposes Client-Customized Adaptation (C2A), a novel hypernetwork-based FL framework that generates client-specific adapters by conditioning the client information that can maximize the utility of shared model parameters while minimizing the divergence caused by client heterogeneity.

Abstract

Despite the versatility of pre-trained language models (PLMs) across domains, their large memory footprints pose significant challenges in federated learning (FL), where the training model has to be distributed between a server and clients. One potential solution to bypass such constraints might be the use of parameter-efficient fine-tuning (PEFT) in the context of FL. However, we have observed that typical PEFT tends to severely suffer from heterogeneity among clients in FL scenarios, resulting in unstable and slow convergence. In this paper, we propose Client-Customized Adaptation (C2A), a novel hypernetwork-based FL framework that generates client-specific adapters by conditioning the client information. With the effectiveness of the hypernetworks in generating customized weights through learning to adopt the different characteristics of inputs, C2A can maximize the utility of shared model parameters while minimizing the divergence caused by client heterogeneity. To verify the efficacy of C2A, we perform extensive evaluations on FL scenarios involving heterogeneity in label and language distributions. Comprehensive evaluation results clearly support the superiority of C2A in terms of both efficiency and effectiveness in FL scenarios.

Figures (5)

• Figure 1: Conceptual illustration of the existing PEFT modules ($\mathcal{A}$) and the client-customized adaptation ($\mathcal{H}$). The proposed method learns to generate the client-customized PEFT modules rather than fitting a single global module to all clients.
• Figure 2: Sensitivity analysis of PEFT methods in federated context in terms of data heterogeneity and divergence from the global model. PT indicates prompt-tuning prompt, and the dotted line indicates full fine-tuning method.
• Figure 3: Overview of the proposed framework, denoted as Client-Customized Adaptation (C2A). To perform customized adaptation, C2A takes into account the client information as a form of label and context. Based on the client embeddings, the factorized hypernetworks generate adapters that are specialized for each client.
• Figure 4: Evaluation results of the test accuracy with different numbers of local epochs.
• Figure 5: Evaluation results of test accuracy on NC dataset with the different number of dimensions for client embedding and factorization.