Method and system for enhancing training data and improving performance for neural network models

1. A computer-implemented method for improving performance of a neural network model, the computer-implemented method comprising:

receiving, at a novelty based machine learning system with a processor, the neural network model and a training data associated with the neural network model;

examining, at the novelty based machine learning system with the processor, a first plurality of neuron activations inside the neural network model for the training data;

creating, at the novelty based machine learning system with the processor, a statistical profile of the first plurality of neuron activations inside the neural network model for the training data;

receiving, at the novelty based machine learning system with the processor, a new set of data samples to improve the neural network model;

examining, at the novelty based machine learning system with the processor, a second plurality of neuron activations of each new sample of the new set of data samples inside the neural network model;

measuring, at the novelty based machine learning system with the processor, novelty of each new sample of the new set of data samples by comparing difference between the second plurality of neuron activations of each new sample of the new set of data samples inside the neural network model with the statistical profile of the first plurality of neuron activations inside the neural network model for the training data;

extracting, at the novelty based machine learning system with the processor, one or more samples from the new set of data samples with largest novelty measurements; wherein the extracted one or more samples are labelled in case of supervised learning, wherein the extracted one or more samples are not labelled in case of unsupervised learning;

adding, at the novelty based machine learning system with the processor, the extracted one or more samples with the largest novelty measurements to the training data of the neural network model, wherein the addition of the extracted one or more samples is performed for re-training the neural network model; and

repeating, at the novelty based machine learning system with the processor, above steps until the novelty of the new set of data samples is below a pre-defined threshold, wherein these iterations are performed for improving performance of the neural network model.

2. The computer-implemented method as recited in claim 1 , wherein the training data is received in one or more formats, wherein the one or more formats comprising at least one of metric data, text, images, videos, gif and animation.

3. The computer-implemented method as recited in claim 1 , wherein the new set of data samples is received in one or more formats, wherein the one or more formats comprising at least one of metric data, text, images, videos, gif and animation.

4. The computer-implemented method as recited in claim 1 , wherein the training data is received from one or more sources, wherein the one or more sources comprising at least one of an enterprise data source, IOT devices, an application, a third-party database, one or more online knowledgebase, one or more offline knowledgebase, an input device, a scanner, and a hardware computing device.

5. The computer-implemented method as recited in claim 1 , wherein the training data is received using random sampling from a dataset, wherein the random sampling comprising simple random sampling, stratified sampling, cluster sampling, multistage sampling, and systematic random sampling.

6. The computer-implemented method as recited in claim 1 , further comprising including, at the novelty based machine learning system with the processor, a subset of the second plurality of neuron activations in the novelty measurements of each new sample of the new set of data samples, wherein the subset is included if difference between the second plurality of neuron activations with the first plurality of neuron activations exceeds the pre-defined threshold, wherein the pre-defined threshold for activations of different neurons are different, wherein the pre-defined threshold is a statistical measure such as a multiple of standard deviation measured in the statistical profile of the first plurality of neuron activations of the training data.

7. The computer-implemented method as recited in claim 1 , further comprising including, at the novelty based machine learning system with the processor, a subset of the second plurality of neuron activations in the novelty measurements of each new sample of the new set of data samples, wherein the subset is included if shortest distance of neurons from any output of the neural network model is below a pre-defined threshold.

8. The computer-implemented method as recited in claim 1 , further comprising measuring, at the novelty based machine learning system with the processor, the novelty of each new sample of the new set of data samples, wherein the novelty is measured by statistically aggregating distance of each neuron activation in the second plurality of neuron activations of the new sample from the statistical profile of the first plurality of neuron activations of the training data.

9. A computer system comprising:

one or more processors; and

a memory coupled to the one or more processors, the memory for storing instructions which, when executed by the one or more processors, cause the one or more processors to perform a method for improving performance of a neural network model, the method comprising:

receiving, at a novelty based machine learning system, the neural network model and a training data associated with the neural network model;

examining, at the novelty based machine learning system, a first plurality of neuron activations inside the neural network model for the training data;

creating, at the novelty based machine learning system, a statistical profile of the first plurality of neuron activations inside the neural network model for the training data;

receiving, at the novelty based machine learning system, a new set of data samples to improve the neural network model;

examining, at the novelty based machine learning system, a second plurality of neuron activations of each new sample of the new set of data samples inside the neural network model;

measuring, at the novelty based machine learning system, novelty of each new sample of the new set of data samples by comparing difference between the second plurality of neuron activations of each new sample of the new set of data samples inside the neural network model with the statistical profile of the first plurality of neuron activations inside the neural network model for the training data;

extracting, at the novelty based machine learning system, one or more samples from the new set of data samples with largest novelty measurements, wherein the extracted one or more samples are labelled in case of supervised learning, wherein the extracted one or more samples are unlabeled in case of unsupervised learning;

adding, at the novelty based machine learning system, the extracted one or more samples with the largest novelty measurements to the training data of the neural network model, wherein the addition of the extracted one or more samples is performed for re-training the neural network model; and

repeating, at the novelty based machine learning system, above steps until the novelty of the new set of data samples is below a pre-defined threshold, wherein these iterations are performed for improving performance of the neural network model.

10. The computer system as recited in claim 9 , wherein the training data is received using random sampling from a dataset, wherein the random sampling comprising simple random sampling, stratified sampling, cluster sampling, multistage sampling, and systematic random sampling.

11. The computer system as recited in claim 9 , wherein the training data is received in one or more formats, wherein the one or more formats comprising at least one of metric data, text, images, videos, gif and animation.

12. The computer system as recited in claim 9 , wherein the new set of data samples is received in one or more formats, wherein the one or more formats comprising metric data, text, images, videos, gif and animation.

13. The computer system as recited in claim 9 , wherein the training data is received from one or more sources, wherein the one or more sources comprising at least one of an enterprise data source, IOT devices, an application, a third-party database, one or more online knowledgebase, one or more offline knowledgebase, an input device, a scanner, and a hardware computing device.

14. The computer system as recited in claim 9 , further comprising including, at the novelty based machine learning system, a subset of the second plurality of neuron activations in the novelty measurements of each new sample of the new set of data samples, wherein the subset is included if shortest distance of neurons is below the pre-defined threshold from any output of the neural network model.

15. The computer system as recited in claim 9 , further comprising including, at the novelty based machine learning system, a subset of the second plurality of neuron activations in the novelty measurements of each new sample of the new set of data samples, wherein the subset is included if difference between the second plurality of neuron activations with the first plurality of neuron activations exceeds the pre-defined threshold, wherein the pre-defined threshold for activations of different neurons are different, wherein the pre-defined threshold is statistical measure such as a multiple of standard deviation measured in the statistical profile of the first plurality of neuron activations of the training data.

16. A non-transitory computer-readable storage medium encoding computer executable instructions that, when executed by at least one processor, performs a method for improving performance of a neural network model, the method comprising:

receiving, at a computing device, the neural network model and a training data associated with the neural network model;

examining, at the computing device, a first plurality of neuron activations inside the neural network model for the training data;

creating, at the computing device, a statistical profile of the first plurality of neuron activations inside the neural network model for the training data;

receiving, at the computing device, a new set of data samples to improve the neural network model;

examining, at the computing device, a second plurality of neuron activations of each new sample of the new set of data samples inside the neural network model;

measuring, at the computing device, novelty of each new sample of the new set of data samples by comparing difference between the second plurality of neuron activations of each new sample of the new set of data samples inside the neural network model with the statistical profile of the first plurality of neuron activations inside the neural network model for the training data;

extracting, at the computing device, one or more samples from the new set of data samples with largest novelty measurements, wherein the extracted one or more samples are labelled in case of supervised learning, wherein the extracted one or more samples are unlabeled in case of unsupervised learning;

adding, at the computing device, the extracted one or more samples with the largest novelty measurements to the training data of the neural network model, wherein the addition of the extracted one or more samples is performed for re-training the neural network model; and

repeating, at the computing device, above steps until the novelty of the new set of data samples is below a pre-defined threshold, wherein these iterations are performed for improving performance of the neural network model.

17. The non-transitory computer-readable storage medium as recited in claim 16 , wherein the training data is received using random sampling from a dataset, wherein the random sampling comprising simple random sampling, stratified sampling, cluster sampling, multistage sampling, and systematic random sampling.

18. The non-transitory computer-readable storage medium as recited in claim 16 , wherein the training data is received in one or more formats, wherein the one or more formats comprising at least one of metric data, text, images, videos, gif and animation.

19. The non-transitory computer-readable storage medium as recited in claim 16 , wherein the new set of data samples is received in one or more formats, wherein the one or more formats comprising at least one of metric data, text, images, videos, gif and animation.

20. The non-transitory computer-readable storage medium as recited in claim 16 , wherein the training data is received from one or more sources, wherein the one or more sources comprising at least one of an enterprise data source, IOT devices, an application, a third-party database, one or more online knowledgebase, one or more offline knowledgebase, an input device, a scanner, and a hardware computing device.