ANOMALY DETECTION

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for anomaly detection. One of the methods includes receiving training data for updating an updateable trained first predictive model. The method includes inputting the training data into a trained second predictive model and generating predictive output data. The method includes based on the predictive output data, detecting an anomaly in the training data as compared to previously received training data. The method includes generating a retrained first predictive model based on the updateable trained first predictive model, a training function and training data that includes the received training data and previously received training data, wherein the received training data is weighted differently than the previously received training data based on the detected anomaly.