Abstract: 

Spacecraft are exceptionally complex and expensive machines with thousands of telemetry channels detailing temperature, radiation, power, instrumentation, and computational activities. Current alarm-based systems for helping engineers monitor spacecraft health are insufficient, and increasing data rates and volumes exacerbate this problem. Fortunately, improvements in technology and deep learning now provide the tools to build improved anomaly detection systems. This talk will describe the development of our current LSTM-based system from research to deployment.

Bio: 

Kyle Hundman is a data scientist and group lead within the IT Chief Technology and Innovation Office at NASA’s Jet Propulsion Laboratory (JPL) and is also currently enrolled in the Machine Learning track of the Masters of Computer Science program at Columbia University. Prior to joining JPL he graduated from Northwestern University’s Masters of Science in Analytics (MSiA) program in 2014. In addition to developing spacecraft anomaly detection systems, he is working on adapting and extending search, natural language processing, and machine learning research from the DARPA Memex program for use by JPL’s Mission Formulation office.