1 paper
Alex Aizman, Abhishek Gaikwad, Piotr Żelasko
Machine learning training pipelines consume data in batches. A single training step may require thousands of samples drawn from shards distributed across a storage cluster. Issuing…