The Machine Learning and Instrument Autonomy strives to research, develop, and infuse data driven solutions built with applied Machine Learning and related technologies to support the exploration of Earth & Space and the advancement of science. MLIA is enabling data science approach for JPL's past and future missions, such as OCO-2/3, SBG, Europa Clipper, NISR, and M2020.
▸ Content-based Search
▸ Data Triage and Down-link Prioritization
▸ Knowledge Compression and Summarization
▸ Change and Novelty Detection
Science Serving Data Science
▸ Science Understanding / Insight Generation
▸ Interpretable Models
▸ ML Uncertainty Awareness
▸ Data Exploration & Discovery
How we work
MLIA works in tandem with instrument and mission PI's and scientists across JPL. Typically, collaborations start with short, non-committal brain storming sessions to explore new challenges, determine whether data-driven solutions would be appropriate, and recommend approaches with the most promise . Next, we partner to draft a proposal targeting either internal (SRTD, TRTD, JNEXT) or external (ROSES, AIST, ACCESS) funding opportunities. Once funding is secured, collaborative research can begin. Together with our partners, we use data science approaches to explore and understand the data, develop simple interpretable models, and generate insight. As needed and only as much as necessary, we incorporate increasingly sophisticated technology to achieve the project goals. However, even when using cutting-edge approaches like deep learning, we strive to extract as much insight as feasible into the original data, the behavior and capability of the model, and the key drivers of uncertainty and bias.
04/15/22 - Mark Wronkiewicz was covered in this month’s TALENTS issue published by the Office of the Chief Technologist. The article gives an overview of the onboard science autonomy being built for the Ocean Worlds Life Surveyor by the OWLS-Autonomy team. Their onboard software suite is focused on summarizing and prioritizing data for return during future life detection missions to the outer solar system.
04/01/22 - Steven Lu and Kiri Wagstaff delivered version 2.0 of the Mars Target Encyclopedia (MTE) to the PDS. The MTE contains information about Mars mission surface targets that have appeared in scientific publications. Version 2.0 adds targets that were identified by the Mars Exploration Rover Spirit, adding to the MTE’s coverage of the Mars Pathfinder and Mars Phoenix Lander missions. Users can search for targets with particular composition (e.g., “calcium”, “perchlorate”) or properties such as “pitted” or “icy soil.”
02/01/22 - Dr. Lukas Mandrake gave a talk entitled “An Overview of Data Science for Physical Scientists”, to launch the JPL Science Understanding from Data Science (SUDS) Seminar series. The SUDS initiative seeks to bring together physical and data scientists to form a new, thriving community of practice, including educational opportunities, demonstrations of cutting-edge research collaboration, and strategic partnerships with academia and other NASA centers.
12/01/21 - The Visual Precision Targeting (VPT) library for landmark detection on the M2020 PIXL instrument was used onboard for the first time as part of science operations on Sol 269. VPT enabled precision placement (within ~1 mm) over a target of interest identified in a micro context camera image acquired on Sol 257 by visually aligning the Sol 269 image with the Sol 257 image, then mapping the target into the instrument’s current coordinate frame to accommodate instrument placement error.
Want to collaborate, or just do some quick brainstorming together? Please reach out!
Want to join our team? We are looking for new colleagues that know how to generate insight from complex data, have experience with UQ, or demonstrated how to merge physical and statistical models. If that sounds like you, please directly reach out to Lukas Mandrake.