Dynamic Gradient Aggregation for Federated Domain 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.

arXiv commentarXiv admin note: substantial text overlap with arXiv:2008.02452

References in corpus (12)

• TensorFlow: A system for large-scale machine learning
• A Survey on Federated Learning Systems: Vision, Hype and Reality for Data Privacy and Protection
• Sequence Transduction with Recurrent Neural Networks
• Federated Optimization:Distributed Optimization Beyond the Datacenter
• Horovod: fast and easy distributed deep learning in TensorFlow
• Re-evaluating Continual Learning Scenarios: A Categorization and Case for Strong Baselines
• Adaptive Federated Optimization
• Variance Reduction in SGD by Distributed Importance Sampling
• Online Learning to Sample
• Federated Transfer Learning with Dynamic Gradient Aggregation
• Federated Learning for Keyword Spotting
• Asynchronous Decentralized Learning of a Neural Network