Fetching the paper…

DataLens: Scalable Privacy Preserving Training via Gradient Compression and Aggregation · Around