Communication-Efficient Learning of Deep Networks from Decentralized Data

TLDR

Mobile devices generate abundant, often privacy‑sensitive data that could improve on‑device models, yet conventional centralized training is impeded by data size and privacy concerns. The authors propose to keep data on devices and learn a shared model by aggregating locally computed updates. They introduce Federated Learning, a practical method that iteratively averages model updates across devices, evaluated on five architectures and four datasets. Experiments show the method is robust to unbalanced, non‑IID data and cuts communication rounds by 10–100× compared to synchronized SGD.

Abstract

Modern mobile devices have access to a wealth of data suitable for learning models, which in turn can greatly improve the user experience on the device. For example, language models can improve speech recognition and text entry, and image models can automatically select good photos. However, this rich data is often privacy sensitive, large in quantity, or both, which may preclude logging to the data center and training there using conventional approaches. We advocate an alternative that leaves the training data distributed on the mobile devices, and learns a shared model by aggregating locally-computed updates. We term this decentralized approach Federated Learning. We present a practical method for the federated learning of deep networks based on iterative model averaging, and conduct an extensive empirical evaluation, considering five different model architectures and four datasets. These experiments demonstrate the approach is robust to the unbalanced and non-IID data distributions that are a defining characteristic of this setting. Communication costs are the principal constraint, and we show a reduction in required communication rounds by 10-100x as compared to synchronized stochastic gradient descent.