# Machine Learning applied to asteroid dynamics

**Authors:** V. Carruba, S. Aljbaae, R. C. Domingos, M. Huaman, W. Barletta  
**arXiv:** [2110.06611](https://arxiv.org/abs/2110.06611)  
**Date:** 2021-10-13  
**Summary page:** https://www.summarizepaper.com/en/paper/2110.06611v4/

## Key points

- Machine Learning (ML) is a branch of computer science focused on developing algorithms that learn from data
- ML techniques include supervised learning and unsupervised learning
- Supervised learning uses labeled examples to learn patterns and make predictions
- Unsupervised learning discovers patterns and structures in unlabeled data
- Deep learning is a subset of ML that uses artificial neural networks with multiple layers to process complex information
- Deep learning models, like the Transformer model, have shown promise in various applications
- The Transformer model relies on attention processes instead of recurrent or convolutional layers for superior results with faster training times
- ML methods are increasingly used in asteroid dynamics for tasks like identifying asteroid families and detecting resonant arguments among asteroids in three-body resonances
- ML applications in asteroid dynamics are considered emerging compared to other astronomical subfields
- Large observational surveys like those at the Vera C. Rubin Observatory will generate substantial datasets for further development of ML techniques in asteroid dynamics
- ML methods are expected to play a crucial role in analyzing data from next-generation astronomical surveys, such as GAIA and the Zwicky Transient Facility (ZTF)
- The Vera C. Rubin Observatory is projected to increase the known number of small bodies in the Solar System by a factor of 10-100, leading to the discovery of millions of new asteroids within its first year of operation
- Handling and analyzing massive amounts of data from these surveys have spurred the development of new methods and software packages

## Summary

Machine Learning (ML) is a branch of computer science that focuses on developing algorithms that can learn from data. It encompasses various techniques, including supervised learning and unsupervised learning. In supervised learning, the computer is provided with labeled examples to learn patterns and make predictions, while in unsupervised learning, the algorithm discovers patterns and structures in unlabeled data. Deep learning is a subset of machine learning that utilizes artificial neural networks with multiple layers to process complex information. Inspired by the biological neural networks found in animal brains, deep learning models have shown great promise in various applications. One notable architecture is the Transformer model, which relies on attention processes rather than recurrent or convolutional layers. This approach has been proven to be highly effective, producing superior results with faster training times. In the field of asteroid dynamics, machine learning methods have been increasingly utilized for tasks such as identifying members of asteroid families and detecting resonant arguments among asteroids in three-body resonances. However, compared to other astronomical subfields like Solar System bodies imaging and spectrophotometry, ML applications in asteroid dynamics are still considered emerging. The research community and methodologies are not as established, resulting in fewer papers producing significant discoveries. The advent of large observational surveys like those conducted at the Vera C. Rubin Observatory will generate substantial datasets containing orbital and physical properties of asteroids. These datasets present an opportunity for ML techniques such as clustering, image identification, and anomaly detection to be developed further and provide valuable insights into asteroid dynamics. Furthermore, ML applications are expected to play a crucial role in analyzing data produced by next-generation astronomical surveys. For instance, space missions like GAIA have already generated vast databases that require ML methods to fully leverage the available information. Ongoing robotic surveys such as the Zwicky Transient Facility (ZTF) have also employed ML approaches for tasks like identifying fast-moving near-Earth objects and comets. However, it is estimated that ZTF's data output is only a fraction of what is expected from the Vera C. Rubin Observatory. With its advanced telescope design, this observatory is projected to increase the known number of small bodies in the Solar System by a factor of 10-100, potentially leading to the discovery of millions of new asteroids within its first year of operation. The handling and analysis of such massive amounts of data have already spurred the development of new methods and software packages. In conclusion, this review highlights the current state and potential impact of ML applications in asteroid dynamics and other astronomical subfields.