Region of interest weighted anomaly detection

1. A computer-implemented method, comprising:

generating a reconstructed dataset from an original dataset by using a generative model;

calculating a differential dataset between the original dataset and the reconstructed dataset as a differential dataset;

determining a region of interest of the original dataset and a region of interest of the reconstructed dataset;

weighting the differential dataset by using the determined region of interest;

applying Gradient-weighted Class Activation Mapping (Grad-CAM) to obtain a Grad-CAM output as the region of interest of the original dataset and the region of interest of the reconstructed dataset; and

detecting an anomaly by using the weighted differential dataset.

2. The method of claim 1 , wherein the generative model is a Variational Auto Encoder.

3. The method of claim 1 , wherein the determining of a region of interest of the original dataset and a region of interest of the reconstructed dataset, comprises:

calculating an integration of the region of interest of the original dataset and the region of interest of the reconstructed dataset,

wherein the weighting a differential dataset with the determined region of interest, comprises:

weighting a differential dataset with the integration of the region of interest of the original dataset and the region of interest of the reconstructed dataset.

4. The method of claim 3 , wherein the determining a region of interest of the original dataset and a region of interest of the reconstructed dataset, further comprises:

classifying each of the original dataset and the reconstructed dataset as one of normal and anomalous by using a Convolutional Neural Network; and

applying the Gradient-weighted Class Activation Mapping (Grad-CAM) to the Convolutional Neural Network to obtain the Grad-CAM output as the region of interest of the original dataset and the region of interest of the reconstructed dataset.

5. The method of claim 4 , further comprising:

applying a first activating function, which returns a positive value in response to a negative value, to the region of interest of the original dataset to obtain a first output; and

applying a second activating function, which does not return a positive value to a negative value, to the region of interest of the reconstructed dataset to obtain a second output,

wherein the integration of the region of interest of the original dataset and the region of interest of the reconstructed dataset is a summation of the first output and the second output.

6. The method of claim 5 , wherein the first activating function is an absolute value (ABS) function, and the second activating function is a Rectified Linear Unit (ReLU) function.

7. The method of claim 4 , wherein the detecting the anomaly by using the weighted differential dataset, comprises: detecting the anomaly by using the weighted differential dataset and the classification of the original dataset.

8. The method of claim 1 , further comprising: normalizing the weighted differential dataset before the detecting the anomaly by using the determined region of interest.

9. The method of claim 1 , wherein the original dataset is an image data.

10. The method of claim 9 , wherein the image data is 2D image data or 3D image data.

11. The method of claim 1 , wherein the detecting the anomaly by using the weighted differential dataset, comprises:

calculating a summation of the weighted differential dataset; and

determining that the original dataset is anomalous in response to the summation exceeding a threshold.

12. The method of claim 1 , wherein the detecting the anomaly by using the weighted differential dataset, comprises:

classifying the weighted differential dataset using a neural network; and

determining that the original dataset is anomalous on condition that the neural network classifies the weighted differential dataset as anomalous.

13. An apparatus comprising:

a processor or a programmable circuitry; and

one or more computer readable mediums collectively including instructions that, when executed by the processor or the programmable circuitry, cause the processor or the programmable circuitry to perform operations including:

generating a reconstructed dataset from an original dataset by using a generative model;

calculating a differential dataset between the original dataset and the reconstructed dataset as a differential dataset;

determining a region of interest of the original dataset and a region of interest of the reconstructed dataset;

weighting the differential dataset by using the determined region of interest;

applying Gradient-weighted Class Activation Mapping (Grad-CAM) to obtain a Grad-CAM output as the region of interest of the original dataset and the region of interest of the reconstructed dataset; and

detecting an anomaly by using the weighted differential dataset.

14. The apparatus of claim 13 , wherein the generative model is a Variational Auto Encoder.

15. The apparatus of claim 13 , wherein the determining a region of interest of the original dataset and a region of interest of the reconstructed dataset, comprises:

calculating an integration of the region of interest of the original dataset and the region of interest of the reconstructed dataset,

wherein the weighting a differential dataset with the determined region of interest, comprises:

weighting a differential dataset with the integration of the region of interest of the original dataset and the region of interest of the reconstructed dataset.

16. The apparatus of claim 15 , wherein the determining a region of interest of the original dataset and a region of interest of the reconstructed dataset, further comprises:

classifying each of the original dataset and the reconstructed dataset as one of normal and anomaly by using a Convolutional Neural Network; and

applying the Gradient-weighted Class Activation Mapping (Grad-CAM) to the Convolutional Neural Network to obtain the Grad-CAM output as the region of interest of the original dataset and the region of interest of the reconstructed dataset.

17. A computer program product including one or more computer readable storage mediums collectively storing program instructions that are executable by a processor or programmable circuitry to cause the processor or programmable circuitry to perform operations comprising:

generating a reconstructed dataset from an original dataset by using a generative model;

calculating a differential dataset between the original dataset and the reconstructed dataset as a differential dataset;

determining a region of interest of the original dataset and a region of interest of the reconstructed dataset;

weighting the differential dataset by using the determined region of interest;

applying Gradient-weighted Class Activation Mapping (Grad-CAM) to obtain a Grad-CAM output as the region of interest of the original dataset and the region of interest of the reconstructed dataset; and

detecting an anomaly by using the weighted differential dataset.

18. The computer program product of claim 17 , wherein the generative model is a Variational Auto Encoder.

19. The computer program product of claim 17 , wherein the determining a region of interest of the original dataset and a region of interest of the reconstructed dataset, comprises:

calculating an integration of the region of interest of the original dataset and the region of interest of the reconstructed dataset,

wherein the weighting a differential dataset with the determined region of interest, comprises:

weighting a differential dataset with the integration of the region of interest of the original dataset and the region of interest of the reconstructed dataset.

20. The computer program product of claim 19 , wherein the determining a region of interest of the original dataset and a region of interest of the reconstructed dataset, further comprises:

classifying each of the original dataset and the reconstructed dataset as one of normal and anomaly by using a Convolutional Neural Network; and

applying the Gradient-weighted Class Activation Mapping (Grad-CAM) to the Convolutional Neural Network to obtain the Grad-CAM output as the region of interest of the original dataset and the region of interest of the reconstructed dataset.