Private Federated Learning with Autotuned Compression
This work addresses communication efficiency and privacy in federated learning for distributed data settings, representing an incremental improvement with automatic tuning.
The paper tackles the problem of reducing communication in private federated learning by proposing on-the-fly methods that automatically adjust compression rates based on training error, while maintaining provable privacy guarantees through secure aggregation and differential privacy. It demonstrates effectiveness on real-world datasets by achieving favorable compression rates without tuning.
We propose new techniques for reducing communication in private federated learning without the need for setting or tuning compression rates. Our on-the-fly methods automatically adjust the compression rate based on the error induced during training, while maintaining provable privacy guarantees through the use of secure aggregation and differential privacy. Our techniques are provably instance-optimal for mean estimation, meaning that they can adapt to the ``hardness of the problem" with minimal interactivity. We demonstrate the effectiveness of our approach on real-world datasets by achieving favorable compression rates without the need for tuning.