Annotation pipeline for machine learning algorithm training and optimization

1. A system, comprising:

a memory that stores computer executable components; and

a processor that executes the computer executable components stored in the memory, wherein the computer executable components comprise:

a collection component that collects unannotated data samples and stores the unannotated data samples in an annotation queue;

a priority evaluation component that determines annotation priority levels for respective unannotated data samples of the unannotated data samples based on attributes associated with the respective unannotated data samples and correlations between the attributes and accuracy of performance of a machine learning model on previous data samples comprising the attributes;

an annotation management component that selects, from amongst different annotation techniques, one or more of the different annotation techniques for annotating the respective unannotated data samples based the annotation priority levels associated with the respective unannotated data samples;

an annotation component that receives annotations applied to the respective unannotated data samples via the one or more different annotation techniques, thereby transforming the unannotated data samples into annotated data samples; and

a model development module that trains the machine learning model using at least some of the annotated data samples.

2. The system of claim 1 , wherein the different annotation techniques are selected from a group consisting of, a manual annotation technique, a supervised learning annotation technique, and a metadata extraction annotation technique.

3. The system of claim 1 , wherein the priority evaluation component determines the annotation priority levels based on estimated confidence in the accuracy of inferences that would be generated based on application of the machine learning model to respective unannotated data samples of the unannotated data samples.

4. The system of claim 3 , wherein the annotation management component further selects a subset of the unannotated data samples for annotating based on the estimated degrees of confidence associated with the respective unannotated data samples.

5. The system of claim 1 , wherein the different annotation techniques comprise a first annotation technique and a second annotation technique, and wherein the annotation component management component selects the first annotation technique for a first subset of the unannotated data samples based on association of the first subset with a first annotation priority level of the annotation priority levels, and selects the second annotation technique for a second subset of the unannotated data samples based on association of the second subset with a second annotation priority level of the annotation priority levels.

6. The system of claim 1 , wherein the priority evaluation component further determines the annotation priority levels based on a quantity of annotated training data samples used to train the machine learning model that correspond to the respective unannotated data samples.

7. The system of claim 1 , wherein the computer executable components further comprise:

an active learning component that employs one or more machine learning techniques to learn the correlations.

8. The system of claim 1 , wherein the computer executable components further comprise:

an active learning component that learns one or more attributes of the attributes that are associated with a degree of accuracy of the performance of the machine learning model that is below a threshold degree of accuracy, and wherein the priority evaluation component assigns a high annotation priority level to a subset of the unannotated data samples based on the subset comprising the one or more attributes.

9. The system of claim 1 , wherein the computer executable components further comprise:

an active learning component that learns one or more attributes of the attributes that are associated with a degree of accuracy of the performance of the machine learning model that is above a threshold degree of accuracy, and wherein the priority evaluation component assigns a low annotation priority level to a subset of the unannotated data samples based on the subset comprising the one or more attributes.

10. The system of claim 1 , wherein the an annotation component that facilitates applying the one or more different annotation techniques to the unannotated data samples to receive the annotations for the respective unannotated data samples.

11. The system of claim 1 , wherein the computer executable components further comprise:

an annotation accuracy evaluation component that evaluates the annotations and determines levels of confidence in the annotations.

12. The system of claim 11 , wherein the annotation accuracy evaluation applies the machine learning model to the respective unannotated data samples to determine the levels of confidence in the annotations.

13. The system of claim 11 , wherein the computer executable components further comprise:

an active learning component that identifies a subset of the annotated data samples for reannotating based on the annotations associated being associated with a level of confidence that is below a threshold level of confidence and sends the subset back to the annotation queue.

14. The system of claim 13 , wherein the computer executable components further comprise:

a feedback component that generates feedback information regarding the subset and facilitates rendering the feedback information at a device associated with an entity responsible for reviewing the subset.

15. The system of claim 11 , wherein the active learning component further identifies a subset of the annotated data samples associated with annotations with a confidence level that is above a threshold level of confidence and adds the subset to a set of annotated training data samples for training or updating the machine learning model.

16. The system of claim 15 , wherein the model development module employs the set of annotated training data samples to train or update the machine learning model.

17. The system of claim 1 , wherein the data samples comprise medical images and wherein the machine learning model is configured to generate medical inferences regarding a medical condition or disease based on the medical images.

18. A method, comprising:

collecting, by a system comprising a processor, unannotated data samples in an annotation queue;

determining, by the system, annotation priority levels for respective unannotated data samples of the unannotated data samples based on attributes associated with the respective unannotated data samples and correlations between the attributes and accuracy of performance of a machine learning model on previous data samples comprising the attributes;

selecting, by the system from amongst different annotation techniques, one or more of the different annotation techniques for annotating the respective unannotated data samples based the annotation priority levels associated with the respective unannotated data samples

receiving, by the system, annotations applied to the respective unannotated data samples via the one or more different annotation techniques, thereby transforming the unannotated data samples into annotated data samples; and

training, by the system, the machine learning model using at least some of the annotated data samples.

19. The method of claim 18 , wherein the different annotation techniques are selected from a group consisting of, a manual annotation technique, a supervised learning annotation technique, and a metadata extraction annotation technique.

20. The method of claim 18 , wherein the determining comprises determining the annotation priority levels based on estimated confidence in the accuracy of inferences that would be generated based on application of the machine learning model to respective unannotated data samples of the unannotated data samples.

21. The method of claim 18 , further comprising:

selecting, by the system, a subset of the unannotated data samples for annotating based on the annotation priority levels associated with the respective unannotated data samples.

22. The method of claim 18 , wherein the different annotation techniques comprise a first annotation technique and a second annotation technique, and wherein the selecting further comprises:

selecting, by the system, the first annotation technique for a first subset of the unannotated data samples based on association of the first subset with a first annotation priority level of the annotation priority levels; and

selecting, by the system, the second annotation technique for a second subset of the unannotated data samples based on association of the second subset with a second annotation priority level of the annotation priority levels.

23. The method of claim 18 , further comprising:

employing, by the system, one or more active machine learning techniques to learn the correlations.

24. The method of claim 18 , further comprising:

learning, by the system, one or more attributes of the attributes that are associated with a degree of accuracy of the performance of the machine learning model that is below a threshold degree of accuracy; and

assigning, by the system, a high annotation priority level to a subset of the unannotated data samples based on the subset comprising the one or more attributes.

25. The method of claim 18 , further comprising:

learning, by the system, one or more attributes of the attributes that are associated with a degree of accuracy of the performance of the machine learning model that is above a threshold degree of accuracy; and

assigning, by the system, a low annotation priority level to a subset of the unannotated data samples based on the subset comprising the one or more attributes.

26. The method of claim 18 , further comprising:

controlling, by the system, application of the one or more different annotation techniques to the unannotated data samples to generate the annotations for the respective unannotated data samples.

27. The method of claim 18 , further comprising:

determining, by the system, levels of confidence in the annotations.

28. The method of claim 27 , further comprising:

identifying, by the system, a subset of the annotated data samples for reannotating based on the annotations associated therewith having a level of confidence that is less than a threshold level of confidence; and

sending, by the system, by the subset back to the annotation queue based on the identifying.

29. The method of claim 27 , further comprising:

identifying, by the system, a subset of the annotated data samples associated with annotations having a level of confidence that is greater than a threshold level of confidence; and

employing, by the system, the subset to train or update the machine learning model based on the identifying.

30. A non-transitory machine-readable storage medium, comprising executable instructions that, when executed by a processor, facilitate performance of operations, comprising:

receiving unannotated data samples;

determining priority levels for respective unannotated data samples of the unannotated data samples based on attributes associated with the respective unannotated data samples and correlations between the attributes and accuracy of performance of a machine learning model on previous data samples comprising the attributes;

selecting one or more annotation techniques for annotating the respective unannotated data samples based on the annotation priority levels respectively associated therewith;

receiving annotations applied to the respective unannotated data samples via the one or more annotation techniques, thereby transforming the unannotated data samples into annotated data samples; and

training the machine learning model using at least some of the annotated data samples.

31. The non-transitory machine-readable storage medium of claim 30 , wherein the operations further comprise:

employing one or more active machine learning techniques to learn the correlations.

32. The non-transitory machine-readable storage medium of claim 30 , wherein the operations further comprise:

learning one or more attributes of the attributes that are associated with a degree of accuracy of the performance of the machine learning model that is below a threshold degree of accuracy; and

assigning a high annotation priority level to a subset of the unannotated data samples based on the subset comprising the one or more attributes.