Astronomaly: Personalised active anomaly detection in astronomical data

Survey telescopes such as the Vera C. Rubin Observatory and Square Kilometre Array will discover billions of static dynamic astronomical sources. Properly mined, these enormous datasets likely be wellsprings rare or unknown astrophysical phenomena. The challenge is that are so large most data never seen by human eyes; currently robust instrument we have to detect relevant anomalies. Machine learning a useful tool for anomaly detection in this regime. However, it struggles distinguish between interesting anomalies irrelevant instrumental artefacts sources simply not interest particular scientist. Active combines flexibility intuition brain with raw processing power machine learning. By strategically choosing specific objects expert labelling, minimises amount scientists look through while maximising potential scientific return. Here introduce Astronomaly: general framework novel active approach designed provide personalised recommendations. Astronomaly can operate on types data, including images, light curves spectra. We use Galaxy Zoo dataset demonstrate effectiveness Astronomaly, well simulated thoroughly test our new approach. find both datasets, roughly doubles number found first 100 viewed user. easily extendable include feature extraction techniques, algorithms even different approaches. code publicly available at https://github.com/MichelleLochner/astronomaly.