Generating data based on pre-trained models using generative adversarial models

Patent metadata

Field	Value
Publication number	US-12307377-B2
Application number	US-202017110629-A
Country	US
Kind code	B2
Filing date	Dec 3, 2020
Priority date	Dec 3, 2020
Publication date	May 20, 2025
Grant date	May 20, 2025

Abstract

Techniques for generator model training are provided. A classifier model that was trained using one or more data samples in a target class is received, and a generative adversarial network (GAN) is trained to generate simulated data samples for the target class, comprising: generating a first simulated data sample using a generator model, computing a first discriminator loss by processing the first simulated data sample using a discriminator model, computing a classifier loss by processing the first simulated data sample using the classifier model, and refining the generator model based on the first discriminator loss and the classifier loss.

First claim

What is claimed is:

1. A method, comprising:
receiving a classifier model that was trained using one or more data samples in belonging to a plurality of classes;
determining a target class of the plurality of classes; and
training a generative adversarial network (GAN) to generate simulated data samples belonging to the target class, comprising:
generating a random input vector;
generating a first simulated data sample based on processing only the random input vector using a generator model;
generating a first discriminator output by processing the first simulated data using a discriminator model, wherein the first discriminator output predicts whether the first simulated data sample was generated by the generator model;
computing a first discriminator loss based on whether the first discriminator output accurately classifies the first simulated data sample as having been generated by the generator model;
generating a second discriminator output by processing a first data samples from one or more non-target classes of the plurality of classes using the discriminator model, wherein the second discriminator output predicts whether the first data sample was generated by the generator model;
computing a second discriminator loss based on whether the second discriminator output accurately classifies the first data sample as not generated by the generator model;
generating a classifier output by processing the first simulated data sample using the classifier model, wherein the classifier output predicts whether the first simulated data sample belongs to the target class;
computing a classifier loss based on comparing the classifier output and the target class; and
refining the generator model based on the first discriminator loss, the second discriminator loss, and the classifier loss.

2. The method of claim 1 , wherein the GAN is trained without processing any data samples in the target class.

3. The method of claim 1 , wherein training the GAN further comprises:
generating a second simulated data sample using the generator model;
computing a third discriminator loss by processing the second simulated data sample using the discriminator model; and
refining the discriminator model based on the third discriminator loss.

4. The method of claim 1 , wherein the discriminator model is trained to differentiate between the simulated data samples generated by the generator model and data samples not generated by the generator model.

5. The method of claim 1 , the method further comprising:
generating an evaluation data sample for the target class by providing the generator model with a randomized input vector.

6. The method of claim 5 , the method further comprising:
determining that the classifier model was trained with insufficient data samples in the target class based at least in part on the evaluation data sample, wherein determining that the classifier model was trained with insufficient data comprises determining that accuracy of the classifier model with respect to the target class is below a threshold.

7. The method of claim 5 , the method further comprising:
determining that the data samples in the target class used to train the classifier model included one or more suspicious features based at least in part on the evaluation data sample.

8. One or more computer-readable storage media collectively containing computer program code that, when executed by operation of one or more computer processors, performs an operation comprising:
receiving a classifier model that was trained using one or more data samples in belonging to a plurality of classes;
determining a target class of the plurality of classes; and
training a generative adversarial network (GAN) to generate simulated data samples belonging to the target class, comprising:
generating a random input vector;
generating a first simulated data sample based on processing only the random input vector using a generator model;
generating a first discriminator output by processing the first simulated data using a discriminator model, wherein the first discriminator output predicts whether the first simulated data sample was generated by the generator model;
computing a first discriminator loss based on whether the first discriminator output accurately classifies the first simulated data sample as having been generated by the generator model;
generating a second discriminator output by processing a first data samples from one or more non-target classes of the plurality of classes using the discriminator model, wherein the second discriminator output predicts whether the first data sample was generated by the generator model;
computing a second discriminator loss based on whether the second discriminator output accurately classifies the first data sample as not generated by the generator model;
generating a classifier output by processing the first simulated data sample using the classifier model, wherein the classifier output predicts whether the first simulated data sample belongs to the target class;
computing a classifier loss based on comparing the classifier output and the target class; and
refining the generator model based on the first discriminator loss, the second discriminator loss, and the classifier loss.

9. The computer-readable storage media of claim 8 , wherein the GAN is trained without processing any data samples in the target class.

10. The computer-readable storage media of claim 8 , wherein training the GAN further comprises:
generating a second simulated data sample using the generator model;
computing a third discriminator loss by processing the second simulated data sample using the discriminator model; and
refining the discriminator model based on the third discriminator loss.

11. The computer-readable storage media of claim 8 , wherein the discriminator model is trained to differentiate between the simulated data samples generated by the generator model and data samples not generated by the generator model.

12. The computer-readable storage media of claim 8 , the operation further comprising:
generating an evaluation data sample for the target class by providing the generator model with a randomized input vector.

13. The computer-readable storage media of claim 12 , the operation further comprising:
determining that the classifier model was trained with insufficient data samples in the target class based at least in part on the evaluation data sample, wherein determining that the classifier model was trained with insufficient data comprises determining that accuracy of the classifier model with respect to the target class is below a threshold.

14. The computer-readable storage media of claim 12 , the operation further comprising:
determining that the data samples in the target class used to train the classifier model included one or more suspicious features based at least in part on the evaluation data sample.

15. A system comprising:
one or more computer processors; and
one or more memories collectively containing one or more programs which when executed by the one or more computer processors performs an operation, the operation comprising:
receiving a classifier model that was trained using one or more data samples belonging to a plurality of classes;
determining a target class of the plurality of classes; and
training a generative adversarial network (GAN) to generate simulated data samples belonging to the target class, comprising:
generating a random input vector;
generating a first simulated data sample based on processing only the random input vector using a generator model;
gener

Assignees

• IBM

Inventors

• Zhou Wang
• Zhang Yang
• Chang Shiyu

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/094 Primary

  Adversarial learning · CPC title

• G06N3/0475

  Generative networks · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/047

  Probabilistic or stochastic networks · CPC title