Paper No. 246-3
Presentation Time: 8:00 AM-5:30 PM
MACHINE LEARNING IN PREDICTING MULTI-COMPONENT MINERAL COMPOSITIONS FOR SPACECRAFT X-RAY DIFFRACTION INSTRUMENTS
ELEISH, Ahmed1, MORRISON, Shaunna2, PAN, Feifei1, PRABHU, Anirudh2, DOWNS, Robert T.3, RAMPE, Elizabeth4, CASTLE, Nicholas5, BLAKE, David6, BRISTOW, Thomas F.6, HAZEN, Robert M.2, VANIMAN, David T.5, HUANG, Fang7, MING, Douglas W.4, TU, Valerie8, MORRIS, Richard V.4, SARRAZIN, Philippe C.9, TREIMAN, Allan10, CRAIG, Patricia I.5, YEN, Albert S.11, CHIPERA, Steve5, ACHILLES, Cherie N.12 and FOX, Peter13, (1)Tetherless World Constellation, Rensselaer Polytechnic Institute, 110 Eighth Street, Troy, NY 12180, (2)Earth and Planets Laboratory, Carnegie Institution for Science, 5251 Broad Branch Road NW, Washington, DC 20015, (3)Department of Geosciences, University of Arizona, Tucson, AZ 85721, (4)NASA Johnson Space Center, 2101 NASA Pkwy, Houston, TX 77058, (5)Planetary Science Institute, Tucson, AZ 85719, (6)NASA, Ames Research Center, Moffett Field, CA 94035, (7)Earth and Environmental Science, Rensselaer Polytechnic Institute, 110 Eighth Street, Troy, NY 12180, (8)Jacobs, NASA Johnson Space Center, Houston, TX 77058, (9)SETI, 339 Bernardo Ave., Suite 200, Mountain View, CA 94043, (10)Lunar and Planetary Institute, Houston, TX 77058, (11)Jet Propulsion Laboratory, Pasadena, CA 91109, (12)NASA GSFC, USRA, Greenbelt, MD 20771, (13)Rensselaer Polytechnic Institute, 5 State St, Troy, NY 12180
To gain a deeper understanding of the mineral forming environments, geologic history, and potential of past or present habitability, the NASA Mars Science Laboratory CheMin X-ray diffractometer team and collaborators developed crystal-chemical methods for predicting limited chemical compositions of the minerals found in the CheMin samples [1,2]. In this study, we extend the methodology by leveraging a machine learning technique known as Label Distribution Learning (LDL) [3] to forecast multicomponent chemical compositions of mineral phases within Gale Crater and beyond. This application of LDL allows for a more detailed petrologic interpretation of the martian surface's geologic evolution, shedding light on the processes shaping the planet's history.
LDL is an innovative framework designed for classification problems with small datasets, widely recognized for its success in facial recognition tasks such as age estimation. In this research, we adapt the LDL algorithm to predict chemical elements (labels) and their corresponding abundances (degrees) for each martian mineral sample, utilizing crystallographic parameters. The model's performance is assessed based on the distance and similarity between label distributions, as well as mean square error, while also comparing results against traditional machine learning methods.
The adaptation of LDL for mineralogical predictions opens avenues for similar applications in future Martian missions and other planetary exploration endeavors. The capability to infer multicomponent mineral compositions from crystallographic data presents exciting opportunities for advancing our understanding of planetary geology.
[1] Morrison et al. (2018) Am Min, 103(6): 848-856
[2] Morrison et al. (2018) Am Min, 103(6): 857-871
[3] Geng (2016) IEEE Transactions on Knowledge and Data Engineering, 28(7), 1734-1748
[4] Morrison et al. (2018) Predicting Multi-Component Mineral Compositions in Gale crater, Mars with Label Distribution Learning, AGU, P21I-3438
[5] Morrison et al. (2019) Machine Learning in Predicting Multi-Component Mineral Compositions in Gale Crater, Mars, Goldschmidt, Abstract
![[Visit Client Website]](/img/gsa/banner.jpg)