Dynamic Gradient Aggregation for Federated Domain Adaptation

TL;DR

This paper introduces a novel federated learning algorithm with weighted gradient aggregation that significantly accelerates convergence and improves speech recognition performance in supervised and unsupervised settings.

Contribution

A new federated learning scheme using weighted gradient aggregation with two-step optimization, enhancing convergence speed and model accuracy in speech recognition tasks.

Findings

7x faster convergence on LibriSpeech

6% WER reduction in supervised SR

20% WER improvement in session adaptation

Abstract

In this paper, a new learning algorithm for Federated Learning (FL) is introduced. The proposed scheme is based on a weighted gradient aggregation using two-step optimization to offer a flexible training pipeline. Herein, two different flavors of the aggregation method are presented, leading to an order of magnitude improvement in convergence speed compared to other distributed or FL training algorithms like BMUF and FedAvg. Further, the aggregation algorithm acts as a regularizer of the gradient quality. We investigate the effect of our FL algorithm in supervised and unsupervised Speech Recognition (SR) scenarios. The experimental validation is performed based on three tasks: first, the LibriSpeech task showing a speed-up of 7x and 6% word error rate reduction (WERR) compared to the baseline results. The second task is based on session adaptation providing 20% WERR over a powerful LAS model. Finally, our unsupervised pipeline is applied to the conversational SR task. The proposed FL system outperforms the baseline systems in both convergence speed and overall model performance.