Communication-efficient distributed eigenspace estimation with arbitrary node failures

Neural Information Processing Systems 

Modern machine learning has seen the proliferation of heterogeneous distributed environments for training and deploying data science pipelines.