The documents distributed by this server have been provided by the contributing authors as a means to ensure timely dissemination of scholarly and technical work on a noncommercial basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, notwithstanding that they have offered their works here electronically. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.
Publications of SPCL
|High-Performance Communication for Machine Learning|
(Presentation - presented in Aachen, Germany, Jan. 2019, )
AbstractOne of the main drivers behind the rapid recent advances in machine learning has been the availability of efficient system support. Despite existing progress, scaling compute-intensive machine learning workloads to a large number of compute nodes is still a challenging task. In this talk, we provide an overview of communication aspects in deep learning. We address the communication challenge, by proposing SparCML, a general, scalable communication layer for machine learning applications. SparCML is built on the observation that many distributed machine learning algorithms either have naturally sparse communication patterns, or have updates which can be sparsified in a structured way for improved performance, without loss of convergence or accuracy. To exploit this insight, we analyze, design, and implement a set of communication-efficient protocols for sparse input data, in conjunction with efficient machine learning algorithms which can leverage these primitives. Our communication protocols generalize standard collective operations, by allowing processes to contribute sparse input data vectors, of heterogeneous sizes. Our generic communication layer is enriched with additional features, such as support for non-blocking (asynchronous) operations and support for low-precision data representations. We validate our algorithmic results experimentally on a range of large-scale machine learning applications and target architectures, showing that we can leverage sparsity for order-of-magnitude runtime savings, compared to existing methods and frameworks.
Recorded talk (best effort)