Machine learning model updating
Techniques for updating a machine learning (ML) model are described. A device or system may receive input data corresponding to a natural or non-natural language (e.g., gesture) input. Using a first ML model, the device or system may determine the input data corresponds to a data category of a plurality of data categories. Based on the data category, the device or system may select a ML training type from among a plurality of ML training types. Using the input data, the device or system may perform the selected ML training type with respect to a runtime ML model to generate an updated ML model.
1. A computer-implemented method comprising:
processing, using a first machine learning (ML) model, first input data corresponding to a first user input to generate first output data representing the first user input;
determining, using the first output data and a second ML model, that the first input data corresponds to a first training technique of a plurality of training techniques, the first training technique being incremental learning;
performing, using the first input data and the first output data, incremental learning with respect to the first ML model to generate a first updated ML model; and
deleting the first input data after generating the first updated ML model.
2. The computer-implemented method of claim 1 , wherein determining that the first input data corresponds to the first training technique further comprises:
determining, using the second ML model, that the first input data corresponds to an intent indicator; and
determining that the first input data corresponds to the first training technique based at least in part on the second ML model determining that the first input data corresponds to the intent indicator.
3. The computer-implemented method of claim 1 , further comprising:
processing, using the first ML model, second input data corresponding to a second user input to generate second output data representing the second user input;
determining the second input data is associated with a user feedback indicator; and
determining, using the second ML model and based at least in part on the second input data being associated with the user feedback indicator, that the second input data corresponds to a second training technique of the plurality of training techniques, the second training technique being weakly supervised learning.
4. The computer-implemented method of claim 1 , further comprising:
processing, using the first ML model, second input data corresponding to a second user input to generate second output data representing the second user input;
determining, using the second ML model, that the second input data is unassociated with an intent indicator; and
determining, using the second ML model and based at least in part on the second input data being unassociated with the intent indicator, that the second input data corresponds to a second training technique of the plurality of training techniques, the second training technique being unsupervised learning.
5. The computer-implemented method of claim 1 , further comprising:
performing the incremental learning with respect to an automatic speech recognition (ASR) model to generate a first updated ASR model.
6. The computer-implemented method of claim 1 , further comprising:
performing the incremental learning with respect to a named entity recognition (NER) model to generate a first updated NER model.
7. The computer-implemented method of claim 1 , further comprising:
performing the incremental learning with respect to an intent classification (IC) model to generate a first updated IC model.
8. The computer-implemented method of claim 1 , further comprising:
performing the incremental learning with respect to a gesture detection model to generate a first updated gesture detection model.
9. The computer-implemented method of claim 1 , further comprising:
performing, by a first device, the incremental learning to generate first model data; and
sending, by the first device, the first model data to a second device, the second device configured to update a third ML model using the first model data.
10. A computing system comprising:
at least one processor; and
at least one memory comprising instructions that, when executed by the at least one processor, cause the computing system to:
process, using a first machine learning (ML) model), first input data corresponding to a first user input to generate first output data representing the first user input;
determine, using a second ML model, that the first input data corresponds to a first training technique of a plurality of training techniques, the first training technique being incremental learning;
perform, using the first input data and the first output data, incremental learning with respect to the first ML model to generate a first updated ML model; and
delete the first input data after generating the first updated ML model.
11. The computing system of claim 10 , wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:
determine, using the second ML model, that the first input data corresponds to an intent indicator; and
determine the first input data corresponds to the first training technique based at least in part on the second ML model determining the first input data corresponds to the intent indicator.
12. The computing system of claim 10 , wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:
process, using the first ML model, second input data corresponding to a second user input to generate second output data representing the second user input;
determine the second input data is associated with a user feedback indicator; and
determine, using the second ML model and based at least in part on the second input data being associated with the user feedback indicator, that the second input data corresponds to a second training technique of the plurality of training techniques, the second training technique being weakly supervised learning.
13. The computing system of claim 10 , wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:
process, using the first ML model, second input data corresponding to a second user input to generate second output data representing the second user input;
determine, using the second ML model, that the second input data is unassociated with an intent indicator; and
determine, using the second ML model and based at least in part on the second input data being unassociated with the intent indicator, that the second input data corresponds to a second training technique of the plurality of training techniques, the second training technique being unsupervised learning.
14. The computing system of claim 10 , wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:
perform the incremental learning with respect to an automatic speech recognition (ASR) model to generate a first updated ASR model.
15. The computing system of claim 10 , wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:
perform the incremental learning with respect to an named entity recognition (NER) model to generate a first updated NER model.
16. The computing system of claim 10 , wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:
perform the incremental learning with respect to an intent classification (IC) model to generate a first updated IC model.
17. The computing system of claim 10 , wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:
perform, by a first device, the incremental learning to generate first model data; and
send, by the first device, the first model data to a second device, the second device configured to update a third ML model using the first model data.
18. The computer-implemented method of claim 1 , wherein determining that the first input data corresponds to the first training technique further comprises:
determining, using the first output data and the second ML model, that the first input data corresponds to a first indicator from a plurality of indicators, the plurality of indicators including at least one of an intent indicator, an entity indicator, or a domain indicator;
determining first data indicating a likelihood that the first indicator corresponds to the first input data;
determining that the first data satisfies a condition; and
determining that the first input data corresponds to the first training technique based on the second ML model determining that the first input data corresponds to the first indicator and the first data satisfying the condition.
19. The computer-implemented method of claim 1 , wherein determining that the first input data corresponds to the first training technique further comprises:
determining, using the first output data and the second ML model, that the first input data corresponds to a first indicator from a plurality of indicators, the plurality of indicators including at least one of an intent indicator, an entity indicator, or a domain indicator; and
determining that the first input data corresponds to the first training technique based at least in part on the second ML model determining that the first input data corresponds to the first indicator, and
the method further comprises:
processing, using the first ML model, second input data corresponding to a second user input to generate second output data representing the second user input;
determining, using the second output data and the second ML model, that the second input data corresponds to a second indicator from the plurality of indicators;
determining that the second input data is associated with a user feedback indicator; and
determining, based at least in part on the second input data being associated with the second indicator and the user feedback indicator, that the second input data corresponds to a second training technique of the plurality of training techniques, the second training technique being weakly supervised learning.
20. The computer-implemented method of claim 1 , wherein determining that the first input data corresponds to the first training technique further comprises:
determining, using the first output data and the second ML model, that the first input data corresponds to a first indicator from a plurality of indicators, the plurality of indicators including at least one of an intent indicator, an entity indicator, or a domain indicator; and
determining that the first input data corresponds to the first training technique based at least in part on the second ML model determining that the first input data corresponds to the first indicator, and
the method further comprises:
processing, using the first ML model, second input data corresponding to a second user input to generate second output data representing the second user input;
determining, using the second output data and the second ML model, that the second input data is unassociated with the plurality of indicators; and
determining, based at least in part on the second input data being unassociated with the plurality of indicators, that the second input data corresponds to a second training technique of the plurality of training technique, the second training technique being unsupervised learning.
21. A computer-implemented method comprising:
processing, using a first machine learning (ML) model, first input data corresponding to a first user input to generate first output data representing the first user input;
performing, based on the first output data, an action responsive to the first user input;
processing, using the first ML model, second input data corresponding to a second user input to generate second output data representing the second user input;
determining, using the second output data, that the second input data corresponds to user feedback responsive to the action;
associating the first input data with a user feedback indicator;
determining, using a second ML model and based at least in part on the first input data being associated with the user feedback indicator, that the first input data corresponds to a first training technique of a plurality of training techniques, the first training technique being weakly supervised learning;
performing, using the first input data and the first output data, weakly supervised learning with respect to the first ML model to generate a first updated ML model; and
deleting the first input data after generating the first updated ML model.