Automated agent for data copies verification

1. A method for verifying data copies, comprising:

obtaining, for an original data set (ODS), a data copy set (DCS) associated with the ODS;

loading a deep learning model (DLM) optimized for the DCS;

identifying a first non-verified ODS copy (NOC) specified in the DCS;

segmenting the first NOC into a first plurality of NOC fragments;

predicting, using the DLM, a first plurality of target classes for the first plurality of NOC fragments; and

marking, based on the first plurality of target classes, the first NOC as verified.

2. The method of claim 1 , further comprising:

prior to loading the DLM optimized for the DCS:

identifying a verification policy for the DCS;

obtaining, from the verification policy, a training accuracy threshold (TAT);

identifying a plurality of verified ODS copies (VOCs) specified in the DCS;

for each VOC of the plurality of VOCs:

segmenting the VOC into a plurality of VOC fragments;

associating the plurality of VOC fragments with a valid target class, to create a plurality of valid data tuples;

generating a plurality of random fragments, wherein each random fragment comprises random data;

associating the plurality of random fragments with an invalid target class, to create a plurality of invalid data tuples;

generating a merged fragments superset using the plurality of valid data tuples and the plurality of invalid data tuples;

selecting a random subset of the merged fragments superset as a training set; and

identifying an optimal set of DLM parameters, defining the DLM optimized for the DCS, based on the training set for each VOC and the TAT.

3. The method of claim 2 , wherein each valid data tuple comprises a first element and a second element, wherein the first element is raw data contained in a VOC fragment, wherein the second element is a categorical variable representative of the valid target class.

4. The method of claim 2 , wherein each invalid data tuple comprises a first element and a second element, wherein the first element is the random data contained in a random fragment, wherein the second element is a categorical variable representative of the invalid target class.

5. The method of claim 2 , further comprising:

obtaining, from the verification policy, a validation accuracy threshold (VAT);

for each VOC of the plurality of VOCs:

selecting a remainder subset of the merged fragments superset as a validation set; and

identifying an optimal set of DLM hyper-parameters, further defining the DLM optimized for the DCS, based on the validation set for each VOC and the VAT.

6. The method of claim 1 , wherein marking the first NOC as verified based on the first plurality of target classes, comprises:

identifying a verification policy for the DCS;

obtaining, from the verification policy, a valid fragment percent (VFP) threshold;

computing a VFP for the first NOC using the first plurality of target classes, wherein the VFP quantifies a percentage of the first plurality of NOC fragments that have been classified into a valid target class;

determining that the VFP at least meets the VFP threshold; and

modifying, based on the determining, a metadata parameter associated with the first NOC to a verified state.

7. The method of claim 1 , further comprising:

identifying a second NOC specified in the DCS;

segmenting the second NOC into a second plurality of NOC fragments;

predicting, using the DLM, a second plurality of target classes for the second plurality of NOC fragments; and

triggering, based on the second plurality of target classes, at least one action specified in a verification policy for the DCS.

8. The method of claim 7 , wherein triggering at least one action based on the second plurality of target classes, comprises:

identifying the verification policy for the DCS;

obtaining, from the verification policy, a valid fragment percent (VFP) threshold;

computing a VFP for the second NOC using the second plurality of target classes, wherein the VFP quantifies a percentage of the second plurality of NOC fragments that have been classified into a valid target class;

determining that the VFP does not at least meet the VFP threshold; and

invoking, based on the determining, the at least one action specified in the verification policy.

9. A data backup system (DBS), comprising:

a first computer processor;

a first data repository operatively connected to the first computer processor; and

a copy verification agent (CVA) executing on the first computer processor, and programmed to:

obtain, for an original data set (ODS), a data copy set (DCS) from the first data repository, wherein the DCS is associated with the ODS;

load a deep learning model (DLM) optimized for the DCS;

identify a non-verified ODS copy (NOC) specified in the DCS;

segment the NOC into a plurality of NOC fragments;

predict, using the DLM, a plurality of target classes for the plurality of NOC fragments; and

mark, based on the plurality of target classes, the NOC as verified.

10. The DBS of claim 9 , further comprising:

an application program interface (API), and a storage semantics agent (SSA) executing on the first computer processor.

11. The DBS of claim 10 ,

wherein a second data repository is operatively connected to the DBS, wherein the second data repository stores the ODS.

12. The DBS of claim 11 ,

wherein a client computing system (CCS) is operatively connected to the DBS, and comprises:

a second computer processor;

a second data repository operatively connected to the second computer processor; and

a data copy agent (DCA) executing on the second computer processor.

13. A non-transitory computer readable medium (CRM) comprising computer readable program code, which when executed by a computer processor, enables the computer processor to:

obtain, for an original data set (ODS), a data copy set (DCS) associated with the ODS;

load a deep learning model (DLM) optimized for the DCS;

identify a first non-verified ODS copy (NOC) specified in the DCS;

segment the first NOC into a first plurality of NOC fragments;

predict, using the DLM, a first plurality of target classes for the first plurality of NOC fragments; and

mark, based on the first plurality of target classes, the first NOC as verified.

14. The non-transitory CRM of claim 13 , further comprising computer readable program code, which when executed by the computer processor, enables the computer processor to:

prior to loading the DLM optimized for the DCS:

identify a verification policy for the DCS;

obtain, from the verification policy, a training accuracy threshold (TAT);

identify a plurality of verified ODS copies (VOCs) specified in the DCS;

for each VOC of the plurality of VOCs:

segment the VOC into a plurality of VOC fragments;

associate the plurality of VOC fragments with a valid target class, to create a plurality of valid data tuples;

generate a plurality of random fragments, wherein each random fragment comprises random data;

associate the plurality of random fragments with an invalid target class, to create a plurality of invalid data tuples;

generate a merged fragments superset using the plurality of valid data tuples and the plurality of invalid data tuples;

select a random subset of the merged fragments superset as a training set; and

identify an optimal set of DLM parameters, defining the DLM optimized for the DCS, based on the training set for each VOC and the TAT.

15. The non-transitory CRM of claim 14 , wherein each valid data tuple comprises a first element and a second element, wherein the first element is raw data contained in a VOC fragment, wherein the second element is a categorical variable representative of the valid target class.

16. The non-transitory CRM of claim 14 , wherein each invalid data tuple comprises a first element and a second element, wherein the first element is the random data contained in a random fragment, wherein the second element is a categorical variable representative of the invalid target class.

17. The non-transitory CRM of claim 14 , further comprising computer readable program code, which when executed by the computer processor, enables the computer processor to:

obtain, from the verification policy, a validation accuracy threshold (VAT);

for each VOC of the plurality of VOCs:

select a remainder subset of the merged fragments superset as a validation set; and

identify an optimal set of DLM hyper-parameters, further defining the DLM optimized for the DCS, based on the validation set for each VOC and the VAT.

18. The non-transitory CRM of claim 13 , further comprising computer readable program code, which when executed by the computer processor, enables the computer processor to:

in order to mark the first NOC as verified based on the first plurality of target classes:

identify a verification policy for the DCS;

obtain, from the verification policy, a valid fragment percent (VFP) threshold;

compute a VFP for the first NOC using the first plurality of target classes, wherein the VFP quantifies a percentage of the first plurality of NOC fragments that have been classified into a valid target class;

make a determination that the VFP at least meets the VFP threshold; and

modify, based on the determination, a metadata parameter associated with the first NOC to a verified state.

19. The non-transitory CRM of claim 13 , further comprising computer readable program code, which when executed by the computer processor, enables the computer processor to:

identify a second NOC specified in the DCS;

segment the second NOC into a second plurality of NOC fragments;

predict, using the DLM, a second plurality of target classes for the second plurality of NOC fragments; and

trigger, based on the second plurality of target classes, at least one action specified in a verification policy for the DCS.

20. The non-transitory CRM of claim 19 , further comprising computer readable program code, which when executed by the computer processor, enables the computer processor to:

in order to trigger at least one action based on the second plurality of target classes:

identify the verification policy for the DCS;

obtain, from the verification policy, a valid fragment percent (VFP) threshold;

compute a VFP for the second NOC using the second plurality of target classes, wherein the VFP quantifies a percentage of the second plurality of NOC fragments that have been classified into a valid target class;

make a determination that the VFP does not at least meet the VFP threshold; and

invoke, based on the determination, the at least one action specified in the verification policy.