@inproceedings{141bc875c2094c8a85221f01cd185ed4,
title = "Improving strong-scaling of CNN training by exploiting finer-grained parallelism",
abstract = "Scaling CNN training is necessary to keep up with growing datasets and reduce training time. We also see an emerging need to handle datasets with very large samples, where memory requirements for training are large. Existing training frameworks use a data-parallel approach that partitions samples within a mini-batch, but limits to scaling the minibatch size and memory consumption makes this untenable for large samples. We describe and implement new approaches to convolution, which parallelize using spatial decomposition or a combination of sample and spatial decomposition. This introduces many performance knobs for a network, so we develop a performance model for CNNs and present a method for using it to automatically determine efficient parallelization strategies. We evaluate our algorithms with microbenchmarks and image classification with ResNet-50. Our algorithms allow us to prototype a model for a mesh-tangling dataset, where sample sizes are very large. We show that our parallelization achieves excellent strong and weak scaling and enables training for previously unreachable datasets.",
keywords = "Algorithms, Convolution, Deep learning, HPC, Performance modeling",
author = "Nikoli Dryden and Naoya Maruyama and Tom Benson and Tim Moon and Marc Snir and {Van Essen}, Brian",
note = "Prepared by LLNL under Contract DE-AC52-07NA27344 (LLNL-CONF-759919). Funding provided by LDRD #17-SI-003. Some testing/development support work funded by the LLNL Sierra Institutional Center of Excellence. Experiments were performed at the Livermore Computing facility. The authors thank the LBANN and Alkemi teams for their assistance.; 33rd IEEE International Parallel and Distributed Processing Symposium, IPDPS 2019 ; Conference date: 20-05-2019 Through 24-05-2019",
year = "2019",
month = may,
doi = "10.1109/IPDPS.2019.00031",
language = "English (US)",
series = "Proceedings - 2019 IEEE 33rd International Parallel and Distributed Processing Symposium, IPDPS 2019",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
pages = "210--220",
booktitle = "Proceedings - 2019 IEEE 33rd International Parallel and Distributed Processing Symposium, IPDPS 2019",
address = "United States",
}