Hierarchical system and method for generating intercorrelated datasets

1. A system for generating synthetic intercorrelated data, the system comprising:

one or more memory units storing instructions; and

one or more processors configured to execute the instructions to perform operations comprising:

training a parent model by iteratively performing steps comprising:

generating, using the parent model, first latent-space data and second latent-space data, the first and second latent space data being configured to generate intercorrelated datasets when passed as inputs to child models;

generating, using a first child model of the child models, a first synthetic dataset based on the first latent-space data;

generating, using a second child model of the child models, a second synthetic dataset based on the second latent-space data;

comparing the first synthetic dataset and second synthetic dataset to training data, the comparing including determining a test correlation metric of the first synthetic dataset and second synthetic dataset; and

based on the test correlation metric, adjusting a parameter of the parent model or terminating training of the parent model.

2. The system of claim 1 , wherein:

the training data comprises a first training dataset and a second training dataset; and

comparing the first synthetic dataset and second synthetic dataset to the training data comprises:

determining a reference correlation metric between the first training dataset and the second training dataset;

comparing the test correlation metric to the reference correlation metric.

3. The system of claim 1 , wherein:

the parent model is a generative adversarial network model; and

comparing the first synthetic dataset and the second synthetic dataset to the training data comprises using a discriminator of the parent model.

4. The system of claim 1 , where the first child model and the second child model are generative adversarial network models.

5. The system of claim 1 , the operations further comprising:

training the first child model to generate synthetic data; and

training the second child model to generate synthetic data.

6. The system of claim 1 , the operations further comprising:

generating the correlated data by combining the first synthetic data and the second synthetic data; and

transmitting the correlated data to a user device.

7. The system of claim 1 , wherein generating the first latent-space data is based on randomized input data.

8. The system of claim 1 , the operations further comprising using the parent model to generate synthetic data based on a correlation between a subset of the training data and the synthetic data.

9. The system of claim 1 , wherein the first child model is a first instance of a template model and the second child model is a second instance of the template model.

10. The system of claim 9 , wherein the template model is trained on a subset of the training data.

11. The system of claim 1 , wherein the first latent-space data and the second latent-space data partially overlap.

12. The system of claim 1 , the operations further comprising generating a synthetic database using the first child model, the second child model, and the parent model.

13. The system of claim 1 , wherein:

the parent model is a first parent model;

generating the first latent-space data and the second latent-space data is based on first input data; and

the operations further comprise training a second parent model by iteratively performing steps comprising:

generating, using the second parent model, third latent-space data and fourth latent-space data based on second input data, the second input data at least partially overlapping with the first input data;

generating, using the first child model, a third synthetic dataset based on the third latent-space data;

generating, using the second child model, a fourth synthetic dataset based on the fourth latent-space data;

comparing the third synthetic dataset and the fourth synthetic dataset to the training data; and

based on the comparison of the third synthetic dataset and the fourth synthetic dataset to the training data, adjusting a parameter of the second parent model or terminating training of the second parent model.

14. The system of claim 13 , the operations further comprising generating the correlated data by combining the first synthetic dataset, the second synthetic dataset, the third synthetic dataset, and the fourth synthetic dataset.

15. The system of claim 13 , wherein comparing the third synthetic data and fourth synthetic data to the training data comprises comparing a test correlation metric to a reference correlation metric.

16. The system of claim 1 , wherein the parent model is a recurrent neural network model.

17. The system of claim 1 , the operations further comprising:

generating third latent-space data using the parent model; and

generating synthetic dataset based on the third latent-space data using a substitute child model.

18. The system of claim 1 , wherein the synthetic data comprises at least one of audio data, financial data, or demographic data.

19. A method for generating synthetic intercorrelated data, the method comprising:

training a parent model by iteratively performing steps comprising:

generating, using the parent model, first latent-space data and second latent-space data, the first and second latent space data being configured to generate intercorrelated datasets when passed as inputs to one or more child models;

generating, using a first child model, a first synthetic dataset based on the first latent-space data;

generating, using a second child model, a second synthetic dataset based on the second latent-space data;

comparing the first synthetic dataset and second synthetic dataset to training data, the comparing including determining a test correlation metric of the first synthetic dataset and second synthetic dataset; and

based on the test correlation metric, adjusting a parameter of the parent model or terminating training of the parent model.

20. A system for generating synthetic intercorrelated data, the system comprising:

one or more memory units storing instructions; and

one or more processors configured to execute the instructions to perform operations comprising:

training a first child model to generate synthetic data;

training a second child model to generate synthetic data; and

training a parent model by iteratively performing steps comprising:

generating, using the parent model, first latent-space data and second latent-space data, the first and second latent space data being configured to generate intercorrelated datasets when passed as inputs to child models;

generating, using the first child model, a first synthetic dataset based on the first latent-space data;

generating, using the second child model, a second synthetic dataset based on the second latent-space data;

determining a reference correlation metric between a first training dataset and a second training dataset;

determining a test correlation metric between the first synthetic dataset and second synthetic dataset;

comparing the test correlation metric to the reference correlation metric, the comparing including determining a test correlation metric of the first synthetic dataset and second synthetic dataset; and

based on the test correlation metric, adjusting a parameter of the parent model or terminating training of the parent model; and

generating the synthetic intercorrelated data by combining the first synthetic dataset and the second synthetic dataset.