Managing an installed base of artificial intelligence modules

A computer-implemented method for managing an installed base of multiple artificial intelligence (AI) modules, each AI module being configured to transform input data into output data by an internal processing chain, the input data comprising data that characterize a state and/or a behavior of at least one physical plant, apparatus, or system, a behavior of the internal processing chain being determined by a set of configuration parameters trainable based both on training input data and corresponding reference data to which the internal processing chain is to map the training input data, the method comprising:

• obtaining, from multiple AI modules and domains in which each of the AI modules are applied to interface with a physical asset, pairs of input data and corresponding output data, and/or pairs of training input data and corresponding reference data, the pairs of input data and output data and/or pairs of training input data and reference data relating to same or sufficiently similar operating situations of the AI modules by filtering out the pairs of input data and corresponding output data, and/or the pairs of training input data and the corresponding reference data obtained from the multiple AI modules according to a predetermined quantitative similarity criterion;
• aggregating the pairs of input data and corresponding output data, and/or the pairs of training input data and corresponding reference data, to form augmented training data for training one or more of the AI modules;
• determining, based at least in part on the pairs of input data and corresponding output data, a quantitative indicator for a performance of at least one AI module according to a predetermined quality criterion;
• and
• determining that the quantitative indicator for the performance of at least one AI module meets a predetermined condition, and in response:
• transferring learning experience from at least one AI module for which the quantitative indicator for the performance is better to another AI module for which the quantitative indicator for the performance is worse;
• or
• replacing an AI module for which the quantitative indicator for the performance is worse with an AI module for which the quantitative indicator for the performance is better,
• wherein the quantitative similarity criterion is based at least in part on one or more of:
• at least one physical process that is controlled based on the output data produced by each AI module;
• a type of at least one physical quantity that forms part of the input data, the output data, the training input data, and/or the reference data of each AI module;
• and
• at least one environmental and/or geographical condition in the domain in which each AI module is applied.

A computer-implemented method for managing an installed base of multiple artificial intelligence (AI) modules, each AI module being configured to transform input data into output data by an internal processing chain, the input data comprising data that characterize a state and/or a behavior of at least one physical plant, apparatus, or system, a behavior of the internal processing chain being determined by a set of configuration parameters trainable based both on training input data and corresponding reference data to which the internal processing chain is to map the training input data, the method comprising:

• obtaining, from multiple AI modules and domains in which each of the AI modules are applied to interface with a physical asset, pairs of input data and corresponding output data, and/or pairs of training input data and corresponding reference data, and configuration parameters, the parameters and pairs of input data and output data and/or pairs of training input data and reference data relating to same or sufficiently similar operating situations of the AI modules by filtering out the pairs of input data and corresponding output data, and/or the pairs of training input data and the corresponding reference data, and the configuration parameters obtained from the multiple AI modules according to a predetermined quantitative similarity criterion;
• aggregating the pairs of input data and corresponding output data, and/or the pairs of training input data and corresponding reference data, to form augmented training data for training one or more of the AI modules;
• aggregating the configuration parameters to form augmented configuration parameters for configuring the internal processing chain of one or more AI modules;
• determining, based at least in part on the pairs of input data and corresponding output data, a quantitative indicator for a performance of at least one AI module according to a predetermined quality criterion;
• and
• determining that the quantitative indicator for the performance of at least one AI module meets a predetermined condition, and in response:
• transferring learning experience from at least one AI module for which the quantitative indicator for the performance is better to another AI module for which the quantitative indicator for the performance is worse;
• or
• replacing an AI module for which the quantitative indicator for the performance is worse with an AI module for which the quantitative indicator for the performance is better,
• wherein the quantitative similarity criterion is based at least in part on one or more of at least one physical process that is controlled based on the output data produced by each AI module;
• a type of at least one physical quantity that forms part of the input data, the output data, the training input data, and/or the reference data of each AI module;
• and
• at least one environmental and/or geographical condition in the domain in which each AI module is applied.