GxP artificial intelligence / machine learning (AI/ML) platform
A GxP (good practice) platform is implemented to enable artificial intelligence (AI) algorithms to be tracked from creation through training and into production. Deployed algorithms are assigned a GxP chain ID that enables identification of production details associated with respective algorithms. Trained algorithms, each of which are respectively associated with a GxP chain ID, are containerized and can be utilized through an application programing interface (API) to provide a service. The GxP chain ID is linked to production details stored within a database, in which the production details can include information such as data used to train the algorithm, a history version, a date/time stamp when the algorithm was validated, software and hardware on which the algorithm was developed and trained, among other details. Changes to the algorithm can be tracked using an immutable ledger facilitated by the implementation of blockchain.
1. A method performed by a computing device to track algorithms throughout their life cycle, comprising:
determining, by a good practice (GxP) platform, whether to train an artificial intelligence (AI) algorithm using a machine learning platform;
if determined to train the baseline AI algorithm, applying training data to the developed AI algorithm to derive a trained AI algorithm;
attaching a GxP chain identifier (ID) to the trained AI algorithm to enable tracking of production details associated with the trained AI algorithm; and
exposing the trained AI algorithm to a remote or local service for deployment, in which the GxP chain ID follows the trained AI algorithm on any service on which it is exposed.
2. The method of claim 1 , in which the applied training data is unique for a given use scenario.
3. The method of claim 2 , in which the given use scenario is for a customer or a customer's implementation requirements.
4. The method of claim 1 , further comprising placing the trained AI algorithm in a container for deployment in an application programming interface (API), wherein the container includes multiple trained AI algorithms each of which are respectively assigned a unique GxP chain ID.
5. The method of claim 1 , in which the production details include a history version.
6. The method of claim 1 , in which the production details include legal details.
7. The method of claim 1 , in which the production details include software on which the trained AI algorithm was developed and trained.
8. The method of claim 1 , in which the production details include hardware on which the trained AI algorithm was developed and trained.
9. The method of claim 1 , in which the production details include a timestamp as to when the trained AI algorithm was validated, the timestamp including a date and time.
10. The method of claim 1 , in which the production details include an identification of data utilized to train the AI algorithm.
11. The method of claim 1 , in which the production details include an identification of the developed AI algorithm on which the trained AI algorithm is based.
12. The method of claim 1 , further comprising utilizing an immutable ledger to track changes to the trained AI algorithm after deployment, in which a new GxP chain ID is assigned to the changed AI algorithm.
13. The method of claim 1 , wherein the trained AI algorithm's ability to interact with humans is enhanced.
14. The method of claim 1 , wherein the GxP platform determines whether to test the AI algorithm.
15. A computer system to track and identify an artificial intelligence (AI)-driven application operating using an algorithm, comprising:
one or more processors; and
one or more hardware-based non-transitory computer-readable memory devices storing instructions which, when executed by the one or more processors, cause the computer system to:
determine, by a good practice (GxP) platform, whether to train a baseline AI algorithm using a machine learning platform;
if determined to train the baseline AI algorithm, train the baseline AI algorithm using one or more sets of training data;
derive a new algorithm based on the trained baseline AI algorithm;
validate the new algorithm using testing data that is distinct from the training data utilized to derive the new algorithm; and
assign a GxP chain identifier (ID) to the validated new algorithm, in which the GxP chain ID enables identification of production details about the new algorithm including at least a version number of the new algorithm and available history versions for the new algorithm.
16. The computer system of claim 15 , in which the sets of training data used to train the baseline AI algorithm include in-memory training data which is streamed in real time from a source that gathers the data.
17. The computer system of claim 16 , in which the source of the in-memory training data is derived from one or more sensors which gather real-time data of a subject or environment.
18. The computer system of claim 15 , in which the sets of training data used to train the baseline AI algorithm include batch training data.
19. The computer system of claim 15 , in which the sets of training data used to train the baseline AI algorithm include online training data derived from one or more external sources.
20. The computer system of claim 18 , in which an external source for the training data is from a remote service associated with a customer for whom the new algorithm is designed.
21. One or more hardware-based non-transitory computer-readable memory devices storing computer-readable instructions which, when executed by one or more processors operatively coupled to one or more memory devices inside a computing device, cause the computing device to:
determine, by a good practice (GxP) platform, whether to train a baseline artificial intelligence (AI) algorithm using a machine learning platform, based on which, future versions are built;
if determined to train the baseline AI algorithm, apply unique training data from distinct sources to the baseline AI algorithm to derive trained algorithms, in which each trained algorithm is derived from a unique set of training data from one or more sources of the distinct sources;
assign a GxP chain identifier (ID) to each trained algorithm to enable identification and tracking of each respective trained algorithm when deployed, in which the GxP chain ID provides a link to stored production details about the trained algorithms including the respective training data from which they were derived; and
utilize an immutable ledger to track changes to each trained algorithm after deployment, in which each change to a trained algorithm is assigned a new GxP chain ID.
22. The one or more hardware-based non-transitory computer-readable memory devices of claim 21 , in which the executed instructions further cause the computing device to containerize a trained algorithm with one or more additional algorithms for deployment in an application programming interface (API), in which the assigned GxP chain ID tracks the containerized trained algorithm throughout its life cycle.