Deep learning algorithms base their success on building high learning capacity models with millions of parameters that are tuned in a data-driven fashion. These models are trained by processing millions of examples, so that the development of more accurate algorithms is usually limited by the throughput of the computing devices on which they are trained. In this work, we explore how the training of a state-of-the-art neural network for computer vision can be parallelized on a distributed GPU cluster. The effect of distributing the training process is addressed from two different points of view. First, the scalability of the task and its performance in the distributed setting are analyzed. Second, the impact of distributed training methods on the final accuracy of the models is studied.

This work is partially supported by the Spanish Ministry of Economy and Competitivity under contract TIN2012-34557, by the BSC-CNS Severo Ochoa program (SEV-2011-00067), by the SGR programmes (2014-SGR-1051 and 2014-SGR-1421) of the Catalan Government and by the framework of the project BigGraph TEC2013-43935-R, funded by the Spanish Ministerio de Economia y Competitividad and the European Regional Development Fund (ERDF). We also would like to thank the technical support team at the Barcelona Supercomputing center (BSC) especially to Carlos Tripiana.

Peer Reviewed