At the recent privacy-preserving ML (PPML) workshop held by Apple, they announced the release of an interesting new dataset intended for federated ML experimentation and benchmarking. This dataset has ~430k images, spread across ~51k users. The announcement is here:
PPML workshop on Apple blog
Results are yet to be published using this dataset, but we hope our community and the wider privacy-preserving ML community will be able to use this dataset and run some interesting experiments!