High-Performance Hybrid Computation Platform for Astronomy Data Classification

Robert J. Brunner, University of Illinois at Urbana-Champaign

Usage Details

In this proposal, we are requesting a Blue Waters exploratory allocation to translate our new proof-of-concept to hybrid computation framework on astronomical imaging data classification into a Blue Waters application. Our new approach combines the domain experience from human computing together with computational power from high-efficiency Deep Neural Networks implemented on GPU, and optimizes the strategic computational flow, to facilitate high-performance scientific data classification. This is the first exploratory attempt in artificial intelligence and machine learning community to optimally combine the computational power from both human intelligence and computer algorithms to the best of our knowledge. Once successful, this project will eventually scale to Petabytes of imaging data and data from other scientific researches, and will benefit a broad range of scientific data processing problems. Our request, if approved, will allow us to confidently pursue follow-on private and Federal funding and additional Blue Waters allocations to process imaging data from the new Petascale Astronomical surveys like PAN-STARRs or the Illinois led Dark Energy Survey (DES).