AOPT-FL: A Communication-Efficient Federated Learning Method with Clusterd and Sparsification

Federated Learning is a distributed machine learning technique that allows multiple devices to learn a shared model collaboratively without exchanging their data. It can be used to improve model accuracy while preserving user privacy. But traditional Federated Learning incurs significant communication overhead and does not perform well when the training data are not independent and identically distribute (Non-IID). Therefore, a Federated Learning algorithm based on adaptive Top-k sparsification and OPTICS method is proposed, which solves the problem that Federated Learning has low accuracy and high communication overhead on Non-IID data. Compared to existing Federated Learning algorithm, our algorithm has improved the accuracy of the model and reduced communication overhead.