Data augmentation simulator

1 . A method of simulating data augmentation, the method comprising, by a computer system:

receiving input data;

collecting data attributes from the input data based, at least in part, on a type of the input data;

classifying the input data using the data attributes, the classifying comprising partitioning the input data into a plurality of strata that each represent a subpopulation of the input data;

performing class-imbalance removal on the classified input data by executing at least a first stage and a second stage of stratified sampling, the performing yielding a balanced dataset, the performing comprising:

at the first stage of stratified sampling, executing proportionate stratified sampling of the classified input data;

responsive to the executing, determining, for each stratum of the plurality of strata, whether a sample size of the stratum exceeds a defined threshold; and

responsive to a determination that the sample size of at least one stratum of the plurality of strata exceeds the defined threshold, at the second stage of stratified sampling, executing disproportionate stratified sampling of the classified input data;

creating integrated data by cyclically performing data processing, modeling, and integration on the balanced dataset, the integrated data comprising the balanced dataset integrated with data-attribute intelligence comprising at least one recommended data attribute of the balanced dataset for extrapolation;

representing at least a portion of the integrated data as structured knowledge for execution of a particular task;

applying a plurality of extrapolation algorithms to the structured knowledge to yield raw augmented data, wherein the applying is at least partly based on the at least one recommended data attribute for extrapolation;

generating an extrapolated dataset using the raw augmented data;

validating the extrapolated dataset by verifying the extrapolated dataset against stored patterns that ensure acceptable data proportions;

responsive to a determination that the extrapolated dataset fails to satisfy the stored patterns, reapplying the plurality of extrapolation algorithms to the structured knowledge;

constructing a synthetic dataset based on the extrapolated dataset, wherein the constructing is performed responsive to a determination that the extrapolated dataset satisfies the stored patterns; and

using the synthetic dataset for artificial intelligence-based data processing in relation to the particular task.

2 . The method of claim 1 , the using comprising using the synthetic dataset as a training dataset for an artificial intelligence model.

3 . The method of claim 1 , further comprising outputting the integrated data as a continuous data feed to a staging database.

4 . The method of claim 1 , wherein the creating the integrated data comprises cyclically performing data processing, modeling and integration of the balanced dataset until desired data is attained.

5 . The method of claim 4 , wherein the creating the integrated data comprises recommending the at least one recommended data attribute of the balanced dataset.

6 . The method of claim 4 , wherein the creating the integrated data comprises predicting one or more common attributes of the balanced dataset, the data-attribute intelligence comprising the predicted one or more common attributes.

7 . The method of claim 1 , wherein the plurality of extrapolation algorithms comprise local Gaussian process regression, linear extrapolation, Richardson's extrapolation, and quadratic extrapolation.

8 . The method of claim 1 , the generating the extrapolated dataset comprising:

applying a neural accumulator to at least a portion of the raw augmented data;

responsive to the applying the neural accumulator, applying an equation learner to at least a portion of the raw augmented data;

applying a division algorithm to at least a portion of the raw augmented data; and

executing analysis using outputs from the neural accumulator, the equation learner and the division algorithm.

9 . The method of claim 8 , wherein the generating produces data that is split between a declarative memory and a procedural memory, the declarative memory storing data related to a cognitive model and the procedural memory storing data related to facts.

10 . The method of claim 1 , comprising exporting the input data to a learning resource.

11 . The method of claim 1 , comprising performing a statistical analysis of the raw augmented data, the performing the statistical analysis comprising:

executing verification of the raw augmented data; and

comparing outputs of the plurality of extrapolation algorithms to an output of a target extrapolation algorithm.

12 . The method of claim 11 , wherein the executing the verification comprises calculating a standard deviation, error, and confidence interval.

13 . The method of claim 1 , comprising generating pattern summaries that indicate relationships between a plurality of variables in the raw augmented data.

14 . The method of claim 13 , the generating the pattern summaries comprising computing correlation coefficients that indicate strength of relationship between variable pairs.

15 . The method of claim 1 , wherein the constructing comprises publishing the extrapolated dataset to a synthetic data repository.

16 . The method of claim 1 , comprising collecting health and performance metrics from the data attributes.

17 . A computer system comprising a processor and memory, wherein the processor and the memory in combination are operable to implement a method comprising:

receiving input data;

collecting data attributes from the input data based, at least in part, on a type of the input data;

classifying the input data using the data attributes, the classifying comprising partitioning the input data into a plurality of strata that each represent a subpopulation of the input data;

performing class-imbalance removal on the classified input data by executing at least a first stage and a second stage of stratified sampling, the performing yielding a balanced dataset, the performing comprising:

at the first stage of stratified sampling, executing proportionate stratified sampling of the classified input data;

responsive to the executing, determining, for each stratum of the plurality of strata, whether a sample size of the stratum exceeds a defined threshold; and

responsive to a determination that the sample size of at least one stratum of the plurality of strata exceeds the defined threshold, at the second stage of stratified sampling, executing disproportionate stratified sampling of the classified input data;

creating integrated data by cyclically performing data processing, modeling, and integration on the balanced dataset, the integrated data comprising the balanced dataset integrated with data-attribute intelligence comprising at least one recommended data attribute of the balanced dataset for extrapolation;

representing at least a portion of the integrated data as structured knowledge for execution of a particular task;

applying a plurality of extrapolation algorithms to the structured knowledge to yield raw augmented data, wherein the applying is at least partly based on the at least one recommended data attribute for extrapolation;

generating an extrapolated dataset using the raw augmented data;

validating the extrapolated dataset by verifying the extrapolated dataset against stored patterns that ensure acceptable data proportions;

responsive to a determination that the extrapolated dataset fails to satisfy the stored patterns, reapplying the plurality of extrapolation algorithms to the structured knowledge;

constructing a synthetic dataset based on the extrapolated dataset, wherein the constructing is performed responsive to a determination that the extrapolated dataset satisfies the stored patterns; and

using the synthetic dataset for artificial intelligence-based data processing in relation to the particular task.

18 . A computer-program product comprising a non-transitory computer-usable medium having computer-readable program code embodied therein, the computer-readable program code adapted to be executed to implement a method comprising:

receiving input data;

collecting data attributes from the input data based, at least in part, on a type of the input data;

classifying the input data using the data attributes, the classifying comprising partitioning the input data into a plurality of strata that each represent a subpopulation of the input data;

performing class-imbalance removal on the classified input data by executing at least a first stage and a second stage of stratified sampling, the performing yielding a balanced dataset, the performing comprising:

at the first stage of stratified sampling, executing proportionate stratified sampling of the classified input data;

responsive to the executing, determining, for each stratum of the plurality of strata, whether a sample size of the stratum exceeds a defined threshold; and

responsive to a determination that the sample size of at least one stratum of the plurality of strata exceeds the defined threshold, at the second stage of stratified sampling, executing disproportionate stratified sampling of the classified input data;

creating integrated data by cyclically performing data processing, modeling, and integration on the balanced dataset, the integrated data comprising the balanced dataset integrated with data-attribute intelligence comprising at least one recommended data attribute of the balanced dataset for extrapolation;

representing at least a portion of the integrated data as structured knowledge for execution of a particular task;

applying a plurality of extrapolation algorithms to the structured knowledge to yield raw augmented data, wherein the applying is at least partly based on the at least one recommended data attribute for extrapolation;

generating an extrapolated dataset using the raw augmented data;

validating the extrapolated dataset by verifying the extrapolated dataset against stored patterns that ensure acceptable data proportions;

responsive to a determination that the extrapolated dataset fails to satisfy the stored patterns, reapplying the plurality of extrapolation algorithms to the structured knowledge;

constructing a synthetic dataset based, at least in part, on the extrapolated dataset; and

using the synthetic dataset for artificial intelligence-based data processing in relation to the particular task.