Machine Learning applied to Planetary Sciences
Course objectives
The learning outcome of the course is to teach students the basics of data science and machine learning by reviewing case studies in Planetary and Earth Science and performing a small research project. The course is intended for both graduate students in planetary science and computer science, as well as advanced undergraduates.
Prerequisites
Knowledge of a high level programming language (e.g., Matlab or Python) and access to a personal laptop where to install the toolboxes needed for the project. In every class, the instructor will provide the students with code templates for the different machine learning techniques.
Research project (60% of the grade)
The students will be asked to develop a small research project related to their research interests and to use machine learning techniques in their projects. The time commitment is expected to be about 3 hours a week. The students will be graded on the following criteria:
the "big picture" argument;
why do you need machine learning?
literature review (both on planetary science and machine learning);
quality of the data sources and project strategy;
algorithm coding, training metrics and computing times;
discussion of the challenges and the results.
The students will be encouraged to bring the project to publication after the course is over and, for graduate students, to include the project in their dissertation.
Example of final projects include the use of supervised machine learning for planetary surface mapping and characterization, surrogate modelling to reduce the computational time of complex models, automated crater counting, automated characterization of astronomical events (e.g., fireballs, exoplanet transits), forward and inverse mapping for remote sensing (e.g., radar mapping), clustering analysis (e.g., asteroid families identification), diversity, equity and inclusion research (e.g., identification of implicit biases in the planetary science community).
Homework (40% of the grade)
In almost all classes, we will do a "hot topics" discussion of a paper driven by one of the student (1/2 of class). For the "hot topics" discussion, the instructor will prepare a handout with a brief description of the paper(s) and some questions. Students will sign up the class prior for which question they want to report on during the in-class discussion. Each student is expected to lead a brief discussion on their question too (~5 minutes).
Tentative syllabus
Week 1: Introduction to machine learning and the goals of the course: syllabus reading and presentation of machine learning software tools
Week 2: Lecture on basics of machine learning: formalism, idea of learning; comparison with human learning; overview of the different types of machine learning algorithms. "Hot topics" discussion on machine learning in planetary sciences.
Azari, A.R. et al. Integrating Machine Learning for Planetary Science: Perspectives for the Next Decade (white paper submitted to the Planetary Science and Astrobiology Decadal Survey 2023-2032)
Week 3: Lecture on supervised machine learning: classification (e.g., Support Vector Machines). "Hot topics" discussion on classification of rock types on Mars.
Wagstaff, K. L., et al. (2013). Smart, texture-sensitive instrument classification for in situ rock and layer analysis. Geophys. Res. Lett., 40(16), pp.4188–4193.
Week 4: Lecture on supervised machine learning: regression (e.g., neural networks). "Hot topics" discussion on predicting orbital properties of planets using deep neural networks.
Lam, C., Kipping, D. (2018) A machine learns to predict the stability of circumbinary planets, MNRAS, Volume 476, Issue 4, June 2018, Pages 5692–5697
Week 5: Mini-symposium for the students to present the ideas for their projects. Make sure the presentation is about 5 minutes.
Week 6: Lecture on unsupervised machine learning: clustering analysis (e.g., k-means algorithms). "Hot topics" discussion on asteroid family identification.
Masiero, J.R. et al. (2013). Asteroid family identification using the hierarchical clustering method and WISE/NEOWISE physical properties. ApJ 770 7
Week 7: Lecture on Convolutional Neural Networks. "Hot topics" discussion on image processing for cratering/feature identification on planetary surfaces.
Palafox, L. F., Hamilton, C. W., Scheidt, S. P., Alvarez, A. M. Automated detection of geological landforms on Mars using Convolutional Neural Networks
Week 8: Lecture on Machine Learning applied to space exploration. "Hot topics" discussion on algorithms on engineering and scientific opportunities and challenges.
McGovern, A. and K. L. Wagstaff (2011). Machine learning in space: extending our reach. Mach Learn, 84, pp. 335–340.
Week 9: Guest lecture on ethical aspects, reproducibility and interpretability of the Machine Learning results. "Hot topics" discussion on "explainable AI".
Arrieta, A. B. et al. ( ). Explainable Artificial Intelligence (XAI): Concepts, Taxonomies, Opportunities and Challenges toward Responsible AI. arXiv:1910.10045
Week 10: Guest lecture on the application of machine learning to Diversity, Equity and Inclusion in the planetary science community. "Hot topics" discussion on recent findings of gender bias in astronomical publications.
Caplar, N., Tacchella S. and Birrer, S. (2017) Quantitative evaluation of gender bias in astronomical publications from citation counts, Nature Astronomy, 1, 0141
Week 11: Mini-symposium: final project presentation (1/2).
Week 12: Mini-symposium: final project presentation (2/2).