Generating realistic synthetic data with adversarial nets

Abstract

A generative network may be learned in an adversarial setting with a goal of modifying synthetic data such that a discriminative network may not be able to reliably tell the difference between refined synthetic data and real data. The generative network and discriminative network may work together to learn how to produce more realistic synthetic data with reduced computational cost. The generative network may iteratively learn a function that synthetic data with a goal of generating refined synthetic data that is more difficult for the discriminative network to differentiate from real data, while the discriminative network may be configured to iteratively learn a function that classifies data as either synthetic or real. Over multiple iterations, the generative network may learn to refine the synthetic data to produce refined synthetic data on which other machine learning models may be trained.

1 .- 20 . (canceled) 
     
     
         21 . A computer implemented method, comprising:
 modifying synthetic data that approximates real data to produce refined synthetic data, wherein the synthetic data is modified according to a refiner neural network and based at least in part on one or more refinement parameters;   classifying the refined synthetic data as either synthetic or real according to a neural network and based in part on one or more parameters; and   adjusting, based on said classifying, at least one of the one or more refinement parameters and at least one of the one or more discriminator parameters to improve realism of the synthetic data and to improve the classifying by the discriminator module.   
     
     
         22 . The method of  claim 21 , wherein said adjusting comprises applying an adversarial cost value to the refinement function, wherein the adversarial cost value is based at least in part on said classifying. 
     
     
         23 . The method of  claim 21 , wherein said adjusting comprises applying a self-regularization term to the refinement function, wherein the self-regularization term comprises a difference value based on a comparison between the synthetic data and refined synthetic data. 
     
     
         24 . The method of  claim 21 , further comprising training the refiner neural network, wherein said training comprises repeating said receiving, said modifying, said classifying and said adjusting. 
     
     
         25 . The method of  claim 24 , wherein said training comprises repeating said receiving, said modifying, said classifying and said adjusting until the refined synthetic data is classified as real. 
     
     
         26 . The method of  claim 21 , further comprising applying the refined synthetic data as training input to a machine learning system, wherein the machine learning system performs better when trained using the refined synthetic data than when trained using the unrefined synthetic data. 
     
     
         27 . The method of  claim 21 , further comprising generating the synthetic data based on one or more key attributes describing the synthesized data. 
     
     
         28 . A system, comprising one or more processors and a memory coupled to the processors, wherein the memory comprises program instructions configured to cause the processors to:
 modify synthetic data that approximates real data to produce refined synthetic data, wherein the synthetic data is modified according to a refiner neural network and based at least in part on one or more refinement parameters;   classify the refined synthetic data as either synthetic or real according to a neural network and based in part on one or more parameters; and   adjust, based on said classify, at least one of the one or more refinement parameters and at least one of the one or more discriminator parameters to improve realism of the synthetic data and to improve the classifying by the discriminator module.   
     
     
         29 . The system of  claim 28 , wherein to adjust the at least one of the one or more refinement parameters, the memory further comprising program instructions configured to cause the processors to apply an adversarial cost value to the refinement function, wherein the adversarial cost value is based at least in part on said classifying of the refined synthetic data. 
     
     
         30 . The system of  claim 28 , wherein to adjust the at least one of the one or more refinement parameters, the memory further comprising program instructions configured to cause the processors to apply a self-regularization term to the refinement function, wherein the self-regularization term comprises a difference value based on a comparison between the synthetic data and refined synthetic data. 
     
     
         31 . The system of  claim 28 , wherein the memory further comprising program instructions configured to cause the processors to repeat said receive, said modify, said classify and said adjust. 
     
     
         32 . The system of  claim 31 , wherein memory further comprising program instructions configured to cause the processors to repeat said receive, said modify, said classify and said adjust until the refined synthetic data is classified as real. 
     
     
         33 . The system of  claim 28 , wherein the memory further comprises program instructions configured to cause the processors to a synthesizer module configured to generate the synthetic data based on one or more key attributes describing the synthesized data. 
     
     
         34 . A non-transitory, computer-readable storage medium storing program instructions that when executed on one or more computers cause the one or more computers to perform:
 modifying synthetic data that approximates real data to produce refined synthetic data, wherein the synthetic data is modified according to a refiner neural network and based at least in part on one or more refinement parameters;   classifying the refined synthetic data as either synthetic or real according to a neural network and based in part on one or more parameters; and   adjusting, based on said classifying, at least one of the one or more refinement parameters and at least one of the one or more discriminator parameters to improve realism of the synthetic data and to improve the classifying by the discriminator module.   
     
     
         35 . The non-transitory, computer-readable storage medium of  claim 34 , wherein said adjusting comprises applying an adversarial cost value to the refinement function, wherein the adversarial cost value is based at least in part on said classifying. 
     
     
         36 . The non-transitory, computer-readable storage medium of  claim 34 , wherein said adjusting comprises applying a self-regularization term to the refinement function, wherein the self-regularization term comprises a difference value based on a comparison between the synthetic data and refined synthetic data. 
     
     
         37 . The non-transitory, computer-readable storage medium of  claim 34 , wherein the program instructions further cause the one or more computers to perform: training the refiner neural network, wherein said training comprises repeating said receiving, said modifying, said classifying and said adjusting. 
     
     
         38 . The non-transitory, computer-readable storage medium of  claim 37 , wherein said training comprises repeating said receiving, said modifying, said classifying and said adjusting until the refined synthetic data is classified as real. 
     
     
         39 . The non-transitory, computer-readable storage medium of  claim 34 , wherein the program instructions further cause the one or more computers to perform: applying the refined synthetic data as training input to a machine learning system, wherein the machine learning system performs better when trained using the refined synthetic data than when trained using the unrefined synthetic data. 
     
     
         40 . The non-transitory, computer-readable storage medium of  claim 34 , wherein the program instructions further cause the one or more computers to perform:
 generating the synthetic data based on one or more key attributes describing the synthesized data.