Understanding Neural Networks for Photometric Data

Greg Olmschenk  ✧  NASA Goddard Space Flight Center, MD, USA

Neural networks, arguably the most powerful machine learning methods, are becoming commonplace in photometric data analysis. This presentation will provide a (hopefully) intuitive understanding of neural networks, including an insight into what they learn, how they learn it, and, importantly, why they often fail to do what we expect them to do. Neural networks are particularly valuable in areas with large quantities of data; As data collection rates continue to increase, the nearly limitless capabilities of neural networks are sure to play an increasingly larger role in future works. The goal of this presentation is that you walk away with the ability to see where neural networks may be valuable in your own work and to conceptually understand how neural networks are being used by others.