TRANSFORMING SOURCE DISTRIBUTION TO TARGET DISTRIBUTION USING SOBOLEV DESCENT

• US 20200134399A1
• Filed: 10/30/2018
• Published: 04/30/2020
• Est. Priority Date: 10/30/2018
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Mroueh, Youssef, Sercu, Tom
• Granted Patent: US 10,860,900 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 18/217 Validation; Performance evaluation; Active pattern learning techniques
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06V 10/82 using neural networks

Abstract

Systems, computer-implemented methods, and computer program products for transforming a source distribution to a target distribution. A system can comprise a memory that stores computer executable components and a processor that executes the computer executable components stored in the memory. The computer executable components can comprise a sampling component that receives a source distribution having a source sample and a target distribution having a target sample. The computer executable components can further comprise an optimizer component that employs a neural network to find a critic that dynamically discriminates between the source sample and the target sample, while constraining a gradient of the neural network. The computer executable components can further comprise a morphing component that generates a first product distribution by morphing the source distribution along the gradient of the neural network to the target distribution.

Claims

1. A system, comprising:

a memory that stores computer executable components;

a processor that executes the computer executable components stored in the memory, wherein the computer executable components comprise;

a sampling component that receives a source distribution having a source sample and a target distribution having a target sample;

an optimizer component that employs a neural network to find a critic that dynamically discriminates between the source sample and the target sample, while constraining a gradient of the neural network; and

a morphing component that generates a first product distribution by morphing the source distribution along the gradient of the neural network to the target distribution.

2. The system of claim 1, wherein the first product distribution comprises a color distribution morphed from respective a target color distribution of the target distribution.

3. The system of claim 1, wherein the first product distribution comprises a shape distribution morphed from respective a target shape distribution of the target distribution.

4. The system of claim 1, wherein the neural network comprises a multi-layer perceptron having at least three layers.

5. The system of claim 1, wherein the morphing component generates a first intermediate distribution from the first product distribution by morphing the source distribution along the gradient of the target distribution.

6. The system of claim 5, wherein the morphing component generates a second intermediate distribution from the first intermediate distribution by morphing the first intermediate distribution along the gradient of the target distribution.

7. The system of claim 1, wherein the neural network comprises domain specific architecture.

8. The system of claim 1, wherein the neural network is a convolution neural network for image applications.

9. A computer-implemented method, comprising:

receiving, by a system operatively coupled to a processor, a source distribution having a source sample, and a target distribution having a target sample;

employing, by the system operatively coupled to the processor, a neural network to find a critic that dynamically discriminates between the source sample and the target sample, while constraining a gradient of the neural network; and

generating, by the system operatively coupled to the processor, a first product distribution by morphing the source distribution along the gradient of the neural network to the target distribution.

10. The computer-implemented method of claim 9, wherein the first product distribution comprises a color distribution morphed from respective a target color distribution of the target distribution.

11. The computer-implemented method of claim 9, wherein the first product distribution comprises a shape distribution morphed from respective a target shape distribution of the target distribution.

12. The computer-implemented method of claim 9, wherein the neural network comprises a multi-layer perceptron having three layers.

13. The computer-implemented method of claim 9, further comprising:

generating, by the system operatively coupled to the processor, a first intermediate distribution from the source distribution by morphing the source distribution along the gradient of the target distribution.

14. The computer-implemented method of claim 13, further comprising:

generating, by the system operatively coupled to the processor, a second intermediate distribution from the first intermediate distribution by morphing the first intermediate distribution along the gradient of the target distribution.

15. The computer-implemented method of claim 9, wherein the neural network comprises domain specific architecture.

16. The computer-implemented method of claim 9, wherein the neural network is a convolution neural network for image applications.

17. A computer program product for transforming a source distribution to a target distribution using Sobolev Descent, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:

receive, by the processor, the source distribution having a source sample and the target distribution having a target sample;

employ, by the processor, a neural network to find a critic that dynamically discriminates between the source sample and the target sample, while constraining a gradient of the neural network; and

generate, by the processor, a first product distribution by morphing the source distribution along the gradient of the neural network to the target distribution.

18. The computer program product of claim 17, wherein the program instructions are further executable by the processor to cause the processor to:

generate, by the processor, a first intermediate distribution from the source distribution by morphing the first product distribution along the gradient of the target distribution.

19. The computer program product of claim 18, wherein the program instructions are further executable by the processor to cause the processor to:

generate, by the processor, a second intermediate distribution from the first intermediate distribution by morphing the first intermediate distribution along the gradient of the target distribution.

20. A system, comprising:

a memory that stores computer executable components;

a processor that executes the computer executable components stored in the memory, wherein the computer executable components comprise;

a sampling component that receives a source distribution having a source sample and a target distribution having a target sample;

an optimizer component that optimizes a neural network, wherein the optimizer component;

determines a critic that dynamically discriminates between the source sample and the target sample,

iteratively optimizes the critic, and

updates one or more parameters of the critic; and

a morphing component that iteratively transports particles of the source distribution along a gradient of the target distribution to create a product distribution, wherein the product distribution is stored in the memory.

21. The system of claim 20, wherein the source distribution comprises a first image and the target distribution comprise a second image.

22. The system of claim 20, wherein the neural network comprises a multi-layer perceptron having at least three layers.

23. A computer-implemented method, comprising:

receiving, by a sampling component operatively coupled to a processor, a source distribution having a source sample and a target distribution having a target sample;

optimizing, by an optimizer component operatively coupled to the processor, a neural network, wherein optimizing comprises;

determining a critic that dynamically discriminates between the source sample and the target sample, and

iteratively updating the critic; and

iteratively transporting, by a morphing component operatively coupled to the processor, one or more particles of the source distribution along a gradient of the target distribution to create a product distribution that is stored in a memory.

24. The computer-implemented method of claim 23, wherein the source distribution comprises a first image and the target distribution comprise a second image.

25. The computer-implemented method of claim 23, wherein the neural network comprises a multi-layer perceptron having at least three layers.