Probabilistic Federated Prompt-Tuning with Non-IID and Imbalanced Data

TL;DR

This paper introduces a probabilistic prompt-tuning method for federated learning that effectively handles non-IID and imbalanced data, outperforming existing aggregation techniques in diverse computer vision tasks.

Contribution

It proposes a novel probabilistic prompt aggregation approach that enhances federated prompt-tuning, addressing data heterogeneity challenges in federated learning.

Findings

Outperforms baseline federated aggregation methods

Effective in non-IID and imbalanced data scenarios

Improves model performance on computer vision datasets

Abstract

Fine-tuning pre-trained models is a popular approach in machine learning for solving complex tasks with moderate data. However, fine-tuning the entire pre-trained model is ineffective in federated data scenarios where local data distributions are diversely skewed. To address this, we explore integrating federated learning with a more effective prompt-tuning method, optimizing for a small set of input prefixes to reprogram the pre-trained model's behavior. Our approach transforms federated learning into a distributed set modeling task, aggregating diverse sets of prompts to globally fine-tune the pre-trained model. We benchmark various baselines based on direct adaptations of existing federated model aggregation techniques and introduce a new probabilistic prompt aggregation method that substantially outperforms these baselines. Our reported results on a variety of computer vision datasets confirm that the proposed method is most effective to combat extreme data heterogeneity in federated learning.