Systems and methods for replacing sensitive data

Abstract

A model optimizer is disclosed for managing training of models with automatic hyperparameter tuning. The model optimizer can perform a process including multiple steps. The steps can include receiving a model generation request, retrieving from a model storage a stored model and a stored hyperparameter value for the stored model, and provisioning computing resources with the stored model according to the stored hyperparameter value to generate a first trained model. The steps can further include provisioning the computing resources with the stored model according to a new hyperparameter value to generate a second trained model, determining a satisfaction of a termination condition, storing the second trained model and the new hyperparameter value in the model storage, and providing the second trained model in response to the model generation request.

1 - 20 . (canceled) 
     
     
         21 . A system comprising:
 at least one processor; and   at least one non-transitory memory storing instructions that, when executed by the at least one processor, cause the system to perform operations comprising:
 receiving actual data having at least one sensitive data portion; 
 determining a class associated with the at least one sensitive data portion; 
 accessing a synthetic data generation model trained using a data space having data of the class; 
 generating, using the synthetic data generation model, at least one synthetic data portion; and 
 replacing the at least one sensitive data portion with the at least one synthetic data portion. 
   
     
     
         22 . The system of  claim 21 , wherein the synthetic data generation model is trained to generate synthetic data satisfying a similarity criterion. 
     
     
         23 . The system of  claim 22 , wherein the similarity criterion is based on at least one of a statistical correlation score, a data similarity score, or a data quality score. 
     
     
         24 . The system of  claim 22 , wherein the synthetic data generation model is a generative adversarial network (GAN). 
     
     
         25 . The system of  claim 21 , wherein determining the class associated with the at least one sensitive data portion comprises applying a recurrent neural network (RNN) to the actual data. 
     
     
         26 . The system of  claim 25 , wherein the RNN is trained to distinguish between classes of data. 
     
     
         27 . The system of  claim 21 , wherein the at least one sensitive data portion is a first text string and the at least one synthetic data portion is a second text string. 
     
     
         28 . The system of  claim 21 , wherein the at least one sensitive data portion comprises at least one of a social security number or a financial service account number. 
     
     
         29 . The system of  claim 21 , wherein the actual data comprises personnel records or patient medical records. 
     
     
         30 . The system of  claim 21 , wherein the actual data comprises unstructured data, the unstructured data comprising at least one of character strings or tokens. 
     
     
         31 . The system of  claim 21 , wherein the actual data comprises structured data, the structured data comprising at least one of a key-value pair, a relational database file, or a spreadsheet. 
     
     
         32 . The system of  claim 21 , the operations further comprising selecting the synthetic data generation model based on the determined class. 
     
     
         33 . The system of  claim 21 , the operations further comprising:
 determining a subclass associated with the at least one sensitive data portion; and   selecting the synthetic data generation model based on the determined subclass.   
     
     
         34 . The system of  claim 33 , wherein determining the subclass comprises using a distribution model associated with the class. 
     
     
         35 . A method for replacing sensitive data, the method comprising:
 receiving actual data having at least one sensitive data portion;   determining a class associated with the at least one sensitive data portion;   accessing a synthetic data generation model trained using a data space having data of the class;   generating, using the synthetic data generation model, at least one synthetic data portion; and   replacing the at least one sensitive data portion with the at least one synthetic data portion.   
     
     
         36 . The method of  claim 35 , wherein the synthetic data generation model is trained to generate synthetic data satisfying a similarity criterion. 
     
     
         37 . The method of  claim 36 , wherein the similarity criterion is based on at least one of a statistical correlation score, a data similarity score, or a data quality score. 
     
     
         38 . The method of  claim 35 , further comprising selecting the synthetic data generation model based on the determined class. 
     
     
         39 . The method of  claim 35 , further comprising:
 determining a subclass associated with the at least one sensitive data portion; and   selecting the synthetic data generation model based on the determined subclass.   
     
     
         40 . A non-transitory computer readable medium containing instructions that, when executed by one or more processors, cause a computing system to perform operations comprising:
 receiving actual data having at least one sensitive data portion;   determining, using a neural network classifier configured to distinguish classes of sensitive data, a class associated with the at least one sensitive data portion;   accessing a synthetic data generation model trained using training data associated with the class;   generating, using the synthetic data generation model, at least one synthetic data portion; and   replacing the at least one sensitive data portion with the at least one synthetic data portion.