An Approach to Tuning Hyperparameters in Parallel: An Approach to Tuning Hyperparameters in Parallel
Loading...
Links to Files
Permanent Link
Author/Creator
Author/Creator ORCID
Date
2019-01-01
Type of Work
Department
Mathematics and Statistics
Program
Mathematics and Statistics
Citation of Original Publication
Rights
Access limited to the UMBC community. Item may possibly be obtained via Interlibrary Loan thorugh a local library, pending author/copyright holder's permission.
This item may be protected under Title 17 of the U.S. Copyright Law. It is made available by UMBC for non-commercial research and education. For permission to publish or reproduce, please see http://aok.lib.umbc.edu/specoll/repro.php or contact Special Collections at speccoll(at)umbc.edu
This item may be protected under Title 17 of the U.S. Copyright Law. It is made available by UMBC for non-commercial research and education. For permission to publish or reproduce, please see http://aok.lib.umbc.edu/specoll/repro.php or contact Special Collections at speccoll(at)umbc.edu
Abstract
Predicting violent storms and dangerous weather conditions, for instance predicting tornados, is an important application for public safety. Using numerical weather simulations to classify a weather pattern as tornadic or not tornadic can take a long time due to the immense complexity associated with current models. Machine learning has the potential to classify tornadic weather patterns much more quickly, since a trained model can classify a storm in seconds when provided data. This allows for alerts to be issued more rapidly such that there is more time for the public to respond to the alert. Preprocessing methods must be applied to the natural data prior to training to prevent inaccuracies. Neural networks must have an equal balance of all classifiable cases for accurate prediction of each case. Different data augmentation approaches have been proposed to fix data imbalances. With our specialized HPC enabled framework, we examine the wall time difference of live data augmentation method versus the use of preaugmented data, when used with a convolutional neural network under various hyperparameter configurations. We also compare CPU and GPU based training over varying sizes of augmented data sets. Then we examine the wall time impact associated with varying the number of GPUs used for training a convolutional neural network. Finally, we create a new data augmentation system by implementing a generative adversarial network and qualitatively compare its output to natural data.