Training anonymized machine learning models via generalized data generated using received trained machine learning models
An example system includes a processor to receive training data and predictions on the training data of a trained machine learning model to be anonymized. The processor is to generate generalized data from training data based on the predictions of the trained machine learning model on the training data. The processor is to train an anonymized machine learning model using the generalized data.
1. A system, comprising a processor to:
receive training data and predictions on the training data of a trained machine learning model to be anonymized;
generate generalized data based on the predictions of the trained machine learning model on the training data; and
train an anonymized machine learning model using the generalized data.
2. The system of claim 1 , wherein the processor is to generate an anonymizer model based on the predictions of the trained machine learning model on the training data and generate the generalized data from the training data using the anonymizer model.
3. The system of claim 1 , wherein the processor is to use the predictions of the trained machine learning model on the training data as input to generate groups of similar records based on similar outputs from the trained machine learning model and generalize the groups to generate an anonymizer model used to generate the generalized data.
4. The system of claim 1 , wherein the generalized data comprises representative values in the same domain as original features used to train the anonymized machine learning model.
5. The system of claim 1 , wherein the training data comprises unlabeled data and the processor is to label the unlabeled data based on the predictions of the trained machine learning model.
6. The system of claim 1 , wherein the predictions comprise outputs from a layer of the trained machine learning model that is prior to a final classification layer.
7. The system of claim 1 , wherein the trained machine learning model comprises a complex model, and the anonymized machine learning model comprises an anonymized part of the complex model.
8. A computer-implemented method, comprising:
receiving, via a processor, training data and predictions on the training data of a trained machine learning model to be anonymized;
generating, via the processor, an anonymizer model based on the predictions of the trained machine learning model on the training data;
anonymizing, via the processor, the training data via the anonymizer model to generate generalized data; and
retraining, via the processor, the trained machine learning model using the generalized data to generate an anonymized machine learning model.
9. The computer-implemented method of claim 8 , wherein generating the anonymizer model comprises training a decision tree using predictions of the trained machine learning model.
10. The computer-implemented method of claim 8 , wherein generating the anonymizer model comprises using a two phase clustering algorithm comprising a coarse clustering phase and a sub-clustering phase.
11. The computer-implemented method of claim 8 , wherein generating the anonymizer model comprises using the predictions of the trained machine learning model on the training data as input to generate groups of similar records and generalizing the groups to generate the anonymizer model.
12. The computer-implemented method of claim 8 , wherein anonymizing the training data comprises replacing data points in each cluster or bucket of similar inputs with a representative value for the cluster or the bucket.
13. The computer-implemented method of claim 8 , comprising receiving the trained machine learning model and generating the predictions, via the trained learning model, on the training data.
14. The computer-implemented method of claim 8 , wherein retraining the trained machine learning model comprises retraining parts of the trained machine learning model, wherein the trained machine learning model comprises a complex model.
15. The computer-implemented method of claim 8 , where the training data that is generalized for creating the anonymized machine learning model is different from the training data used to train the trained machine learning model.
16. A computer program product for anonymizing machine learning models, the computer program product comprising a computer-readable storage medium having program code embodied therewith, wherein the computer-readable storage medium is not a transitory signal per se, the program code executable by a processor to cause the processor to:
receive training data and predictions on the training data of a trained machine learning model to be anonymized;
generate an anonymizer model based on the predictions of the trained machine learning model on the training data;
anonymize the training data via the anonymizer model to generate generalized data; and
retrain the trained machine learning model using the generalized data to generate an anonymized machine learning model.
17. The computer program product of claim 16 , further comprising program code executable by the processor to train a decision tree using predictions of the trained machine learning model.
18. The computer program product of claim 16 , further comprising program code executable by the processor to execute a two phase clustering comprising a coarse clustering phase and a sub-clustering phase.
19. The computer program product of claim 16 , further comprising program code executable by the processor to use the predictions of the trained machine learning model on the training data as input to generate groups of similar records and generalize the groups to generate the anonymizer model.
20. The computer program product of claim 16 , further comprising program code executable by the processor to replace data points in each cluster or bucket of similar inputs with a representative value for the cluster or the bucket.