Contrastive explanations for images with monotonic attribute functions
In an embodiment, a method for generating contrastive information for a classifier prediction comprises receiving image data representative of an input image, using a deep learning classifier model to predict a first classification for the input image, evaluating the input image using a plurality of classifier functions corresponding to respective high-level features to identify one or more of the high-level features absent from the input image, and identifying, from among the high-level features absent from the input image, a pertinent-negative feature that, if added to the input image, will result in the deep learning classifier model predicting a second classification for the modified input image, the second classification being different from the first classification. In an embodiment, the method includes creating a pertinent-positive image that is a modified version of the input image that has the first classification and fewer than all superpixels of the input image.
1. A computer implemented method for generating contrastive information for a classifier prediction, the computer implemented method comprising:
receiving, by one or more processors, image data representative of an original input image;
predicting, by one or more processors, using a deep learning classifier model, a first classification for the input image;
creating, by one or more processors, a set of modified input images by adding each of a set of high-level features to a respective copy of the original image;
predicting, by one or more processors, using the deep learning classifier model, classifications for respective modified input images to identify one or more modified input images classified differently from the original input image; and
identifying, by one or more processors, from among the one or more modified input images classified differently from the original input image, a pertinent-negative image that is most similar to the original input image.
2. The computer implemented method of claim 1 , further comprising:
generating a report that includes a visual representation of the pertinent-negative image.
3. The computer implemented method of claim 1 , further comprising:
creating, by one or more processors, a pertinent-positive image that is a modified version of the input image that includes fewer than all superpixels of the input image while still being predicted by the deep learning classifier model to be in the same first classification as the input image.
4. The computer implemented method of claim 3 , further comprising:
comparing a plurality of modified images to identify a candidate modified image that has the amount of the input image remaining while still being classified in the first classification with the input image, wherein the modified images are modified versions of the input image having fewer than all portions of the input image.
5. The computer implemented method of claim 1 , wherein the identifying of the pertinent-negative image includes determining that pertinent-negative image is a result with a smallest change to the input image from among other potential modifications to the original input image.
6. The computer implemented method of claim 1 , wherein the input image is a multiple-color image.
7. The computer implemented method of claim 1 , wherein the identifying includes identifying that the pertinent-negative image will be classified by the deep learning classifier model as a second classification different from the first classification.
8. A computer usable program product for generating contrastive information for a classifier prediction, the computer usable program product comprising a computer-readable storage medium, and program instructions stored on the storage medium, the stored program instructions comprising: program instructions to receive, by one or more processors, image data representative of an original input image; program instructions to predict, by one or more processors, using a deep learning classifier model, a first classification for the input image; program instructions to create, by one or more processors, a set of modified input images by adding each of a set of high-level features to a respective copy of the original image; program instructions to predict, by one or more processors, using the deep learning classifier model, classifications for respective modified input images to identify one or more modified input images classified differently from the original input image; and program instructions to identify, by one or more processors, from among the one or more modified input images classified differently from the original input image, a pertinent-negative image that is most similar to the original input image.
9. A computer usable program product of claim 8 , further comprising program instructions to generate, by one or more processors, a report that includes a visual representation of the pertinent-negative image.
10. A computer usable program product of claim 8 , further comprising program instructions to create, by one or more processors, a pertinent-positive image that is a modified version of the input image that includes fewer than all superpixels of the input image while still being predicted by the deep learning classifier model to be in the same first classification as the input image.
11. A computer usable program product of claim 10 , further comprising program instructions to compare a plurality of modified images to identify a candidate modified image that has the amount of the input image remaining while still being classified in the first classification with the input image, wherein the modified images are modified versions of the input image having fewer than all portions of the input image.
12. A computer usable program product of claim 8 , wherein the identifying of the pertinent-negative image includes determining that pertinent-negative image is a result with a smallest change to the input image from among other potential modifications to the original input image.
13. A computer usable program product of claim 8 , wherein the input image is a multiple-color image.
14. A computer usable program product of claim 8 , wherein the identifying includes identifying that the pertinent-negative image will be classified by the deep learning classifier model as a second classification different from the first classification.
15. A computer system comprising a processor, a computer-readable memory, and a computer-readable storage device, and program instructions stored on the storage device for execution by the processor via the memory, the stored program instructions comprising:
program instructions to receive, by one or more processors, image data representative of an original input image;
program instructions to predict, by one or more processors, using a deep learning classifier model, a first classification for the input image;
program instructions to create, by one or more processors, a set of modified input images by adding each of a set of high-level features to a respective copy of the original image;
program instructions to predict, by one or more processors, using the deep learning classifier model, classifications for respective modified input images to identify one or more modified input images classified differently from the original input image; and
program instructions to identify, by one or more processors, from among the one or more modified input images classified differently from the original input image, a pertinent-negative image that is most similar to the original input image.
16. The computer system of claim 15 , further comprising program instructions to generate, by one or more processors, a report that includes a visual representation of the pertinent-negative image.
17. The computer system of claim 15 , further comprising program instructions to create, by one or more processors, a pertinent-positive image that is a modified version of the input image that includes fewer than all superpixels of the input image while still being predicted by the deep learning classifier model to be in the same first classification as the input image.
18. The computer system of claim 17 , further comprising program instructions to compare a plurality of modified images to identify a candidate modified image that has the amount of the input image remaining while still being classified in the first classification with the input image, wherein the modified images are modified versions of the input image having fewer than all portions of the input image.
19. The computer system of claim 15 , wherein the identifying of the pertinent-negative image includes determining that pertinent-negative image is a result with a smallest change to the input image from among other potential modifications to the original input image.
20. The computer system of claim 15 , wherein the identifying includes identifying that the pertinent-negative image will be classified by the deep learning classifier model as a second classification different from the first classification.