Robotic process automation architectures and processes for hosting, monitoring, and retraining machine learning models
Robotic process automation (RPA) architectures and processes for hosting, monitoring, and retraining ML machine learning (ML) models are disclosed. Retraining is an important part of the ML model lifecycle. The retraining may depend on the type of the ML model and the data on which the ML model will be trained. A secure storage layer may be used to store data from RPA robots for retraining. This retraining may be performed automatically, remotely, and without user involvement.
1. A system, comprising:
one or more robotic process automation (RPA) robots configured to:
call one or more machine learning (ML) models by respective activities of respective RPA workflows of the one or more RPA robots, the RPA workflows defining a controllable execution order over time and a relationship between a set of activities comprising the respective activities that call the one or more ML models by an expression of the respective activities, at least one activity of the set of activities configured to utilize at least one driver that facilitates interaction between the one or more RPA robots and software of respective computing systems,
receive results from the call to the one or more ML models, and
complete execution of the respective activities that call the one or more ML models; and
an artificial intelligence (AI) center comprising a plurality of physical servers configured to store the one or more ML models and to execute the one or more ML models upon a call from a respective RPA robot of the one or more RPA robots, wherein
the plurality of physical servers are configured to:
store a plurality of datasets for the ML models, each dataset comprising similar types of data in a logical or physical grouping, and
retrain an ML model of the one or more ML models responsive to a training condition being met or responsive to retraining being manually requested using a subset of the plurality of datasets specified in a training configuration for the ML model and deploy the retrained ML model to be called by the one or more RPA robots, wherein the ML model is retrained based on a confidence threshold.
2. The system of claim 1 , wherein the plurality of physical servers comprise:
a training cluster configured to run one or more retraining pipelines;
a core microservices cluster configured to perform create/read/update/delete (CRUD) operations for the one or more ML models and to cause a retraining pipeline of the training cluster to initiate retraining of the one or more ML models; and
an ML services cluster that is configured to execute the one or more ML models and provide results of the execution to calling RPA robots of the one or more RPA robots.
3. The system of claim 1 , wherein an RPA workflow of at least one of the one or more RPA robots comprises a data push activity that comprises a location of storage for the respective RPA robot to upload data for retraining.
4. The system of claim 1 , wherein at least one RPA robot of the one or more RPA robots is configured to iterate through local data, data stored in a remote database, or both, and send the data to storage of the AI center for retraining.
5. The system of claim 1 , wherein the plurality of datasets are presented as pointers to ML models during retraining, the ML model being retrained accessing the subset of the plurality of datasets for retraining via the respective pointers.
6. The system of claim 1 , wherein the plurality of physical servers are configured to:
receive one or more performance measures for the ML model being retrained; and
generate one or more scores for the one or more performance measures during the retraining.
7. The system of claim 6 , wherein changes to the one or more scores are tracked over time as retraining is performed multiple times.
8. The system of claim 1 , wherein the plurality of physical servers are configured to adjust a number of data points for retraining, a frequency with which retraining data is collected, an amount of retraining data that is collected, one or more triggers for retraining, or a combination thereof.
9. The system of claim 1 , wherein an RPA workflow of at least one of the one or more RPA robots is configured to adjust a number of data points for retraining, a frequency with which retraining data is collected, an amount of retraining data that is collected, one or more triggers for retraining, or a combination thereof.
10. The system of claim 1 , wherein the plurality of physical servers permit the one or more RPA robots to call the one or more ML models and receive results of execution of the ML models, but does not permit the one or more RPA robots to access the one or more ML models themselves.
11. A non-transitory computer-readable medium storing a computer program, the computer program configured to cause at least one processor to:
store a plurality of machine learning (ML) models and a plurality of datasets for the ML models, each dataset comprising similar types of data in a logical or physical grouping;
retrain an ML model of the plurality of ML models responsive to a training condition being met or responsive to retraining being manually requested using a subset of the plurality of datasets specified in a training configuration for the ML model, wherein the ML model is retrained based on a confidence threshold; and
deploy the retrained ML model by making the retrained ML model available to be called by one or more robotic process automation (RPA) robots by respective activities of respective RPA workflows of the one or more RPA robots, the RPA workflows defining a controllable execution order over time and a relationship between a set of activities comprising the respective activities that call the one or more ML models by an expression of the respective activities, at least one activity of the set of activities configured to utilize at least one driver that facilitates interaction between the one or more RPA robots and software of respective computing systems.
12. The non-transitory computer-readable medium of claim 11 , wherein the plurality of datasets are presented as pointers to ML models during retraining, the ML model being retrained accessing the subset of the plurality of datasets for retraining via the respective pointers.
13. The non-transitory computer-readable medium of claim 11 , wherein the computer program is further configured to cause the at least one processor to:
receive one or more performance measures for the ML model being retrained; and
generate one or more scores for the one or more performance measures during the retraining.
14. The non-transitory computer-readable medium of claim 13 , wherein changes to the one or more scores are tracked over time as retraining is performed multiple times.
15. The non-transitory computer-readable medium of claim 11 , wherein the computer program is further configured to cause the at least one processor to:
adjust a number of data points for retraining, a frequency with which retraining data is collected, an amount of retraining data that is collected, one or more triggers for retraining, or a combination thereof.
16. The non-transitory computer-readable medium of claim 11 , wherein the computer program is further configured to cause the at least one processor to:
permit the one or more RPA robots to call the plurality of ML models and receive results of execution of the ML models; and
prevent the one or more RPA robots from accessing at least one of the plurality of ML models themselves.
17. An artificial intelligence (AI) center, comprising:
a training cluster comprising at least one physical server configured to run one or more retraining pipelines, wherein the retraining pipelines are training jobs running in the training cluster for different tenants;
a core microservices cluster comprising at least one physical server configured to perform create/read/update/delete (CRUD) operations for one or more machine learning (ML) models and to cause a retraining pipeline of the training cluster to initiate retraining of the one or more ML models, wherein one or more of the ML models are retrained based on a confidence threshold; and
an ML services cluster comprising at least one physical server that is configured to execute the one or more ML models and provide results of the execution to calling robotic process automation (RPA) robots by respective activities of respective RPA workflows of the calling RPA robots, the RPA workflows defining a controllable execution order over time and a relationship between a set of activities comprising the respective activities that call the one or more ML models by an expression of the respective activities, at least one activity of the set of activities configured to utilize at least one driver that facilitates interaction between the calling RPA robots and software of respective computing systems.
18. The AI center of claim 17 , wherein the at least one physical server of the training cluster is configured to retrain an ML model of the one or more ML models responsive to a training condition being met or responsive to retraining being manually requested using a subset of the plurality of datasets specified in a training configuration for the ML model.
19. The AI center of claim 18 , wherein the at least one physical server of the core microservices cluster is configured to deploy the retrained ML model to the at least one physical server of the ML services cluster to be called by the RPA robots.
20. The AI center of claim 18 , wherein the plurality of datasets are presented as pointers to ML models during retraining, the ML model being retrained accessing the subset of the plurality of datasets for retraining via the respective pointers.
21. The AI center of claim 17 , wherein the at least one physical server of the training cluster is configured to adjust a number of data points for retraining, a frequency with which retraining data is collected, an amount of retraining data that is collected, one or more triggers for retraining, or a combination thereof.
22. The AI center of claim 17 , wherein the at least one physical server of the ML services cluster permits the one or more RPA robots to call the one or more ML models and receive results of execution of the ML models, but does not permit the one or more RPA robots to access the one or more ML models themselves.