Abstract:

Machine learning (ML), coupled with hardware innovations, has enabled significant progress in the analysis of large volumes of scientific data. Analysis of scientific datasets is particularly challenging, often requiring a re-design of existing ML models to solve classification or regression problems accurately and efficiently. This talk will specifically focus on application of deep learning to galaxy morphology classification. As ML model and dataset sizes continue to increase, parallel hardware and parallel algorithms become critical in reducing training and inference times. This talk will also present algorithm design techniques that enhance training performance without degrading model accuracy, especially when deployed at multi-GPU and supercomputing scales.

**Refreshments will be served beginning at 3:30pm in the Olin Lobby

See Flyer