Features for black-box machine-learning models

A feature mapper creation system for enabling a transformation system comprising a transformation model built for a first setting using first input values to incorporate second feature values present in a second setting, the feature mapper creation system comprising:

• a processor and a memory, communicatively coupled to the processor, wherein the memory stores program code portions that, when executed, enable the processor to:
• provide training input data comprising second feature values relating to the second features as well as expected second results in the second setting;
• provide a feature mapper comprising a machine-learning model, wherein output signals of the feature mapper are used as input signals for the transformation system, thereby building a combination of the feature mapper and the transformation model, wherein internal details of the transformation system, including architecture, parameters, and first training data used to build the transformation model are unavailable during training of the feature mapper, and the transformation model remains unmodified throughout training;
• create a reinforcement learning agent, wherein the reinforcement learning agent uses first input values and correct output values to create predictions of the combination of the feature mapper and the transformation model, wherein a training loop uses expected results from the second setting as a reward function;
• train the machine-learning model of the feature mapper using the training input data as input for the feature mapper and using the expected second results as expected output data of the transformation system, the training being performed based solely on observed output differences of the transformation system,
• wherein the feature mapper learns to generate input signals that emulate input values of the first setting based on the second feature values, thereby enabling the transformation system to operate in the second setting;
• and
• deploy the combination of the feature mapper and the transformation system as a super machine-learning system comprising a super machine-learning model usable in the second setting.

A computer program product for enabling a transformation system, comprising a transformation model build for a first setting using first input values, to incorporate second feature values present in a second setting, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions being executable by one or more computing systems or controllers to cause the one or more computing systems to:

• provide training input data comprising second feature values relating to the second features as well as expected second results in the second setting;
• provide a feature mapper comprising a machine-learning model, wherein output signals of the feature mapper are used as input signals for the transformation system, thereby building a combination of the feature mapper and the transformation model, wherein internal details of the transformation system, including architecture, parameters, and first training data used to build the transformation model are unavailable during training of the feature mapper, and the transformation model remains unmodified throughout training;
• create a reinforcement learning agent, wherein the reinforcement learning agent uses first input values and correct output values to create predictions of the combination of the feature mapper and the transformation model, wherein a training loop uses expected results from the second setting as a reward function;
• train the machine-learning model of the feature mapper using the training input data as input for the feature mapper and using the expected second results as expected output data of the transformation system, the training being performed based solely on observed output differences of the transformation system,
• wherein the feature mapper learns to generate input signals that emulate input values of the first setting based on the second feature values, thereby enabling the transformation system to operate in the second setting;
• and
• deploy the combination of the feature mapper and the transformation system as a super machine-learning system comprising a super machine-learning model usable in the second setting.