Drift mode acquisition using successive model training

A non-transitory storage medium having stored therein instructions that are executable by one or more hardware processors to perform operations comprising:

• defining a time window during which a drift mode of a machine learning model will be evaluated, and the time window begins at t0;
• training, beginning at the time t0, a reduced reference model R0 without ground truth data but with a data sample d0 that is a subset of a training dataset that was used to train the machine learning model, wherein the reduced reference model R0 is distinct from the machine learning model and is trained to a point of overfitting;
• for ‘n’ iterations:
• at a time ti, when new data samples di are available, training a reduced reference model Ri distinct from the machine learning model and from the reduced reference model R0, wherein the reduced reference model Ri is trained with the data samples di and is trained to a point of overfitting;
• after Ri is trained, collecting data samples vi from a time when the new data samples di are acquired to a time when training the reduced reference model Ri is complete;
• and
• comparing respective inference values generated by the reduced reference model R0 and Ri, using the data sample vi, to the reduced reference models R0 and Ri and storing an absolute difference between the inference values;
• defining a drift mode curve using the absolute differences;
• based on the drift mode curve, determining a drift mode of the machine learning model;
• and
• based on the drift mode, determining a moment at which the machine learning model is retrained or replaced.