The SMC Data Challenge offers a unique opportunity to fuse cutting edge analytics and machine learning with curated, high-value datasets from the United States Department of Energy. Participants are challenged to generate new insights from DoE datasets using novel techniques, and report them in an understandable manner.
7pod Technologies is proud to announce that two submissions to the Data Challenge were accepted as finalists this year, led by co-founder Dr. Max Grossman and Staff Engineer Nav Ravindranath.
The first submission, Massive Visualization of Application Codes, applied graph clustering and other graph analysis techniques to code metadata generated from the DoE’s E3SM (Energy Exascale Earth System Model) application. With an emphasis on leveraging the insights gained to port the application to novel, heterogeneous, and other parallel architectures, the 7pod team focused on uncovering correlations or couplings of procedures in the FORTRAN code base, as well as identifying profitable targets for parallelization or accleration.
The second submission, Impact of Urban Weather on Energy Use, focused on modeling energy output and consumption of urban buildings given local weather patterns. Using both K-Nearest Neighbor regressor models and deep neural networks, the 7pod team was able to achieve R2 accuracies of greater than 0.9. Further work on hyperparameter tuning of the deep neural net, as well as longer and more computationally intensive training on GPUs may yield even better results.
The 7pod team will present these results at the Smoky Mountain Conference in August 2018.