Anomaly detection in time series is an active area of research in which multiple classic and machine learning algorithms were proposed. However, satellite telemetry is a special case of time series characterized by high dimensionality, missing data, and – most importantly – limited information about known anomalies. The leading approach for time series anomaly detection (TSAD) in satellite telemetry is to establish a normality model based on available nominal telemetry periods and search for significant deviations from that model in the test data. Majority of the existing (deep and shallow) TSAD methods are not able to leverage prior knowledge (i.e., a few labelled anomalies) when such information is available. This technology gap can be addressed with a new approach called few-shot learning.