Automated explanation of machine learning predictions using classification score targeting

1. A method comprising:

obtaining: an input, a machine learning model, and a classification score produced by application of the machine learning model to the input;

optimizing iterative application of the machine learning model to perturbations of the input, wherein a target of said optimizing is at least a predefined non-zero classification score; and

generating, based on one of the perturbations which achieved the predefined non-zero classification score, an explanation to the classification score produced by the application of the machine learning model to the input.

2. The method according to claim 1 , wherein said optimizing comprises performing the following actions iteratively:

generating the perturbations;

applying the machine learning model to each of the perturbations, to produce a classification score for each of the perturbations; and

determining which one of the perturbations achieved a classification score equal or closest to the predefined non-zero classification score.

3. The method according to claim 1 , wherein the explanation comprises an indication of one or more aspects of the input which had the most contribution to the classification score produced by the application of the machine learning model to the input.

4. The method according to claim 3 , wherein:

the input is an image, and

said generating of the explanation comprises generating a saliency map in which the one or more aspects are one or more regions of the image.

5. The method according to claim 4 , wherein the predefined non-zero classification score is between 0.1 and the classification score produced by application of the machine learning model to the input.

6. The method according to claim 4 , wherein the predefined non-zero classification score is between 0.5 and the classification score produced by application of the machine learning model to the input.

7. The method according to claim 4 , wherein the predefined non-zero classification score is between 10-90% of the classification score produced by application of the machine learning model to the input.

8. The method according to claim 4 , wherein the predefined non-zero classification score is between 50-90% of the classification score produced by application of the machine learning model to the input.

9. The method according to claim 1 , wherein the input is selected from the group consisting of: an audio file, a video file, and a vector of values.

10. The method according to claim 1 , performed by at least one hardware processor.

11. A system comprising:

(a) at least one hardware processor; and

(b) a non-transitory computer-readable storage medium having program code embodied therewith, the program code executable by said at least one hardware processor to:

obtain: an input, a machine learning model, and a classification score produced by application of the machine learning model to the input,

optimize iterative application of the machine learning model to perturbations of the input, wherein a target of said optimizing is at least a predefined non-zero classification score, and

generate, based on one of the perturbations which achieved the predefined non-zero classification score, an explanation to the classification score produced by the application of the machine learning model to the input.

12. The system according to claim 11 , wherein the optimizing comprises performing the following actions iteratively:

generating the perturbations;

applying the machine learning model to each of the perturbations, to produce a classification score for each of the perturbations; and

determining which one of the perturbations achieved a classification score equal or closest to the predefined non-zero classification score.

13. The system according to claim 11 , wherein the explanation comprises an indication of one or more aspects of the input which had the most contribution to the classification score produced by the application of the machine learning model to the input.

14. The system according to claim 13 , wherein:

the input is an image, and

said generating of the explanation comprises generating a saliency map in which the one or more aspects are one or more regions of the image.

15. The system according to claim 14 , wherein the predefined non-zero classification score is between 0.1 and the classification score produced by application of the machine learning model to the input.

16. The system according to claim 14 , wherein the predefined non-zero classification score is between 0.5 and the classification score produced by application of the machine learning model to the input.

17. The system according to claim 14 , wherein the predefined non-zero classification score is between 10-90% of the classification score produced by application of the machine learning model to the input.

18. The system according to claim 14 , wherein the predefined non-zero classification score is between 50-90% of the classification score produced by application of the machine learning model to the input.

19. The system according to claim 11 , wherein the input is selected from the group consisting of: an audio file, a video file, and a vector of values.

20. A computer program product comprising a non-transitory computer-readable storage medium having program code embodied therewith, the program code executable by at least one hardware processor to:

obtain: an input, a machine learning model, and a classification score produced by application of the machine learning model to the input;

optimize iterative application of the machine learning model to perturbations of the input, wherein a target of said optimizing is at least a predefined non-zero classification score; and

generate, based on one of the perturbations which achieved the predefined non-zero classification score, an explanation to the classification score produced by the application of the machine learning model to the input.