Tell Me Why: Interpretable Machine Learning for Space Exploration and Beyond
The growing deployment of machine learning (ML) systems in medicine, finance, advertising, and even space exploration has raised new, urgent questions about how these systems make their decisions and why. This is especially relevant when ML is used to aid in the discovery of new objects, phenomena, mineralogies, and species. Interpretable ML methods explicitly acknowledge that there is a human involved in the process, and they seek to provide human-understandable reasons for their decisions. I will describe an approach to interpretable discovery in which the algorithm identifies specific attributes that make an observation novel. These machine explanations can help us take the first step in understanding a novel discovery. I will share our experiences applying this method in support of scientific investigations such as Mars exploration and biodiversity monitoring in freshwater streams on Earth.
Technical Talk: Mars Target Encyclopedia: Text-based Machine Learning for Planetary Science
Every day, rovers on Mars send back data for new observation targets (e.g., rocks, soils, layers). Some of these targets yield new discoveries that are published in the scientific literature. Yet there is currently no accessible link between data (or targets) and their subsequent publications. We constructed the Mars Target Encyclopedia (MTE) to enable users to ask questions such as "What do we know about target Epworth?" and "What are all of the Mars targets that contain hematite?" We used information extraction and machine learning methods to mine the steadily growing body of scientific publications and extract compositional knowledge about Mars surface targets. The MTE benefits Mars mission planners, planetary scientists, and the interested public by condensing relevant knowledge into a central resource in an accessible way. More than just a literature search, the MTE allows us to ask new questions that previously could not be answered.
Bio:
Dr. Kiri L. Wagstaff is a principal researcher in artificial intelligence and machine learning at the Jet Propulsion Laboratory. Her research focuses on developing new machine learning and data analysis methods for spacecraft. She holds a Ph.D. in Computer Science from Cornell University, an M.S. in Geological Sciences from the University of Southern California, and an MLIS from San Jose State University. She received a 2008 Lew Allen Award for Excellence in Research for work on the sensitivity of machine learning methods to high-radiation space environments and a 2012 NASA Exceptional Technology Achievement award for work on transient detection methods in radio astronomy data. She is passionate about keeping machine learning relevant to real-world problems.