Anomaly Detection with the Vera Rubin Observatory

The large, wide-field, multipurpose LSST survey conducted with the Vera C. Rubin Observatory will bring unprecedented opportunities for scientific discovery. A census of the solar system will result in the discovery of millions of new small bodies, the transient survey will find thousands of new supernovae and other transients each year and deep images will show the low surface brightness universe for the first time.

In each of these domains, there exists the opportunity for finding the truly unexpected and unusual.

Such anomaly detection requires the development of new machine learning methods capable of highlighting objects or events worthy of attention. Amongst the data, there exists the possibility of technosignatures - evidence of intelligent life in the Universe - which is a particular focus for this work.

This DPhil, which overlaps with the first data releases from the survey, will develop such techniques for a broad set of science cases. With the opportunity to follow up objects of interest via facilities around the world, this project will involve a broad range of precursor astronomical data including: ZTF, TESS and the Euclid and DESI samples of galaxies. It builds on the methodology in Rogers, Lintott et al. 2024 (AJ, 167, 3, 188).

This paper made use of a deep autoencoder to derive a latent space within which objects included in the latest simulation of the Vera Rubin Observatory solar system can be placed, training on colour and orbital parameters. Those objects which are separate from the bulk of the population can be identified as anomalies, and in addition to those which are extreme in brightness, or which have unconstrained orbits, we find clusters of interstellar objects, those with unusual orbital parameters and those whose colour is unusual for a given location in the Solar System.

As solar system catalogues are expected to be available early in the LSST survey, we will also apply this method to real data.

The large, wide-field, multipurpose LSST survey conducted with the Vera C. Rubin Observatory will bring unprecedented opportunities for scientific discovery. A census of the solar system will result in the discovery of millions of new small bodies, the transient survey will find thousands of new supernovae and other transients each year and deep images will show the low surface brightness universe for the first time.

In each of these domains, there exists the opportunity for finding the truly unexpected and unusual.

Such anomaly detection requires the development of new machine learning methods capable of highlighting objects or events worthy of attention. Amongst the data, there exists the possibility of technosignatures - evidence of intelligent life in the Universe - which is a particular focus for this work.

This DPhil, which overlaps with the first data releases from the survey, will develop such techniques for a broad set of science cases. With the opportunity to follow up objects of interest via facilities around the world, this project would suit a student with a broad interest in observational astronomy and a desire to develop skills in citizen science, machine learning and beyond.