Experimenting with Machine Learning/AI in Science Operations

Operational efficiency and reliability are two performance indicators that modern astronomical observatories try to maximize. These depend on multiple factors like readiness of equipment (telescope, instruments, auxiliary systems), program scheduling practices, the ratio of calibration/system health measurements versus the amount of time spent on the scientific targets, identification and troubleshooting of problems, etc. In many of these areas the decisions and actions taken by astronomers and operators are crucial, and they are based on an increasingly large amount of information and judgement. By nature, these decisions are subject to errors, personal impressions, inexperience, bad habits or misconceptions not always compatible with achieving the highest efficiency or reliability. We believe Machine Learning (ML) and Artificial Intelligence (AI) techniques can help in this direction and therefore in our APEX Science Operations team we have started learning and experimenting with some ML/AI concepts. The final goal would be developing a small prototype system to assist astronomers and operators taking better decisions on specific operational aspects.