Measuring the effects of augmentation artifacts on a machine learning network

A system comprising:

• one or more processing units;
• and
• one or more memory units storing instructions that, when executed by the one or more processing units, cause the one or more processing units to execute operations comprising:
• training a neural network using training input data corresponding to a first perspective;
• generating first testing data corresponding to a second perspective, the second perspective defined by at least one value of at least one parameter being offset from the first perspective in a first direction;
• generating second testing data corresponding to a third perspective, the second perspective defined by the at least one value of the at least one parameter being offset from the first perspective in a second direction;
• applying the first testing data and the second testing data to the neural network;
• evaluating a difference in performance of the neural network between the first testing data and the second testing data, the difference in performance indicating whether the neural network relies on a presence of one or more artifacts to perform inferences;
• and
• displaying data indicating the difference in performance of the neural network.

A processor comprising:

• one or more circuits to
• train a machine learning model (MLM) using training input data generated using a first value of a parameter that defines at least one value associated with at least one characteristic of the training input data,
• apply first testing data to the MLM, the first testing data being generated using a second value of the parameter that shifts the at least one value in a first direction;
• apply second testing data to the MLM, the second testing data being generated using a third value of the parameter that shifts the at least one value in a second direction;
• determine a difference in performance of the MLM between the first testing data and the second testing data, the difference in performance indicating whether the MLM is trained to rely on one or more artifacts, and
• perform one or more computer operations based at least on the determining of the difference in performance.