Creating deep learning models using feature augmentation

1. A computer implemented method for reducing computation processing time, computation load and required storage space in media objects classification, by automatically creating a classification function trained with augmented representation of features extracted from a plurality of sample media objects, comprising;

reducing computation processing time, computation load and required storage space in creating a classification function, of at least one hardware processor by causing said at least one hardware processor to execute a code comprising code instructions for performing the following:

extracting a plurality of features from a plurality of sample media objects;

upon completion of said extraction, generating a plurality of feature samples for each one of said plurality of features by augmenting said plurality of features;

training said classification function with said plurality of features samples; and

outputting said classification function for classifying at least one new media object.

2. The computer implemented method of claim 1 , wherein said classification function may perform as at least one of: a regression function and a clustering function.

3. The computer implemented method of claim 1 , wherein said classification function is a convolutional neural network (CNN).

4. The computer implemented method of claim 1 , wherein said plurality of features is extracted by submitting said plurality of sample media objects to a CNN during a learning process of said CNN.

5. The computer implemented method of claim 1 , further comprising a topology of said classification function is optimized during said extraction of said plurality of features according to a size of each of said plurality of sample media objects.

6. The computer implemented method of claim 1 , wherein said augmentation includes applying at least one transformation to each of said plurality of features.

7. The computer implemented method of claim 6 , further comprising said at least one transformation is a member selected from a group consisting of: a rotation, a shift, a translation, a scaling, a reflection, a geometric transformation, a pixel flip, a noise injection, a speed perturbation, a pitch shifting, a time stretching, a gain induction and a frequency spectrum warpage.

8. A system for reducing computation processing time, computation load and required storage space in media objects classification, by creating a classification function trained with augmented representations of features extracted from a plurality of sample media objects, comprising:

an interface module;

a program store storing a code, wherein said code, when executed by a processor, causes said processor a reduction in computation processing time, computation load and required storage space in creating a classification function; and

at least one processor coupled to said interface and said program store for executing said stored code, said code comprising:

code instructions to extract a plurality of features from a plurality of sample media objects;

code instructions to generate, upon completion of said extraction, a plurality of features samples for each one of said plurality of features by augmenting said plurality of features;

code instructions to train said classification function with said plurality of features samples; and

code instructions to output said classification function for classifying at least one new media object.

9. The system of claim 8 , wherein said classification function may perform as at least one of: a regression function and a clustering function.

10. The system of claim 8 , wherein said classification function is a CNN.

11. The system of claim 8 , wherein said plurality of features is extracted by submitting said plurality of sample media objects to a CNN during a learning process of said CNN.

12. The system of claim 8 , further comprising a topology of said classification function is optimized during said extraction of said plurality of features according to a size of each of said plurality of sample media objects.

13. The system of claim 8 , wherein said augmentation includes applying at least one transformation to each of said plurality of features.

14. A computer program product for reducing computation processing time, computation load and required storage space in media objects classification, by creating a classification function trained with augmented representations of features extracted from a plurality of sample media objects, comprising:

a non-transitory computer readable storage medium;

first program instructions to extract a plurality of features from a plurality of sample media objects;

second program instructions to generate a plurality of features samples for each one of said plurality of features by augmenting said plurality of features;

third program instructions to train a classification function with said plurality of features samples; and

fourth program instructions to output said classification function for classifying at least one new media object;

wherein said first, second, third and fourth program instructions are executed by at least one processor from said non-transitory computer readable storage medium and wherein said execution of said first, second, third and fourth program instructions causes a reduction in computation processing time, computation load and required storage space in creating said classification function.

15. The computer program product of claim 14 , wherein said classification function may perform as at least one of: a regression function and a clustering function.

16. The computer program product of claim 14 , wherein said classification function is a CNN.

17. The computer program product of claim 14 , wherein said plurality of features is extracted by submitting said plurality of sample media objects to a CNN during a learning process of said CNN.

18. The computer program product of claim 14 , further comprising a topology of said classification function is optimized during said extraction of said plurality of features according to a size of each of said plurality of sample media objects.

19. The computer program product of claim 14 , wherein said augmentation includes applying at least one transformation to each of said plurality of features.

20. The computer program product of claim 19 , further comprising said at least one transformation is a member selected from a group consisting of: a rotation, a shift, a translation, a scaling, a reflection, a geometric transformation, a pixel flip, a noise injection, a speed perturbation, a pitch shifting, a time stretching, a gain induction and/or a frequency spectrum warpage.