Federated Learning Over Images: Vertical Decompositions and Pre-Trained Backbones Are Difficult to Beat
This work provides practical guidance for implementing federated learning in image domains, though it appears incremental in nature.
The paper systematically evaluates federated learning algorithms for image classification, finding that vertical decomposition of neural networks consistently outperforms standard reconciliation methods across diverse settings.
We carefully evaluate a number of algorithms for learning in a federated environment, and test their utility for a variety of image classification tasks. We consider many issues that have not been adequately considered before: whether learning over data sets that do not have diverse sets of images affects the results; whether to use a pre-trained feature extraction "backbone"; how to evaluate learner performance (we argue that classification accuracy is not enough), among others. Overall, across a wide variety of settings, we find that vertically decomposing a neural network seems to give the best results, and outperforms more standard reconciliation-used methods.