IP Library › Granted Patent US 11,468,276
Granted Patent B2
US 11,468,276 · App. 16/850,816 · Granted Oct 11, 2022

System and method of a monotone operator neural network

Inventors: Ezra Winston (Pittsburgh, PA); Jeremy Kolter (Pittsburgh, PA); Anit Kumar Sahu (Pittsburgh, PA)
G06K9/6277G06N3/084G06N5/046G06N20/10
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 11,468,276
App. No.
16/850,816
Granted
Oct 11, 2022
Kind
B2
Abstract

A system for training a neural work that includes an input interface for accessing input data for the neural network and a processor in communication with the input interface. The processor is programmed to receive input at the neural network and output a trained neural networking utilizing a forward prorogation and a backward propagation, wherein the forward propagation includes utilizing a root-finding procedure to identify a fixed point associated with one or more parameters of the neural network, wherein the backward propagation includes identifying a derivative of a loss associated with the parameters of the network.

Claims (33)

1. A computer-implemented method for predicting a classification and training a neural network, comprising:

receiving input at the neural network;

classifying the input of the neural networking utilizing a forward propagation, wherein the forward propagation includes utilizing a root-finding procedure to identify a fixed point associated with one or more parameters of the neural network;

computing a vector by utilizing a linear transformation of the fixed point;

outputting the classification of the input in response to a maximum element of the vector; and

training the neural network utilizing a backward propagation, wherein the backward propagation includes identifying a derivative of a loss associated with the parameters of the neural network.

2. The computer-implemented method of claim 1 , wherein performing backwards propagation includes utilizing a Forward-Backward splitting method.

3. The computer-implemented method of claim 1 , wherein performing backwards propagation includes utilizing a Peaceman-Rachford splitting method.

4. The computer-implemented method of claim 1 , wherein the input includes image data.

5. The computer-implement method of claim 1 , wherein the forward propagation includes utilizing a Forward-Backward splitting method to identify the fixed point.

6. The computer-implement method of claim 1 , wherein the forward propagation includes utilizing the forward propagation until a convergence.

7. The computer-implemented method of claim 1 , wherein the forward propagation includes utilizing a Peaceman-Rachford splitting method to identify the fixed point.

8. The computer-implemented method of claim 1 , wherein the backward propagation includes multiplying a Jacobian of a partial derivative of the fixed point of a nonlinear function.

9. The computer-implemented method of claim 1 , wherein the backward propagation includes utilization of a hyperparameter.

10. A computer-implemented method for a classification of input to a neural network, comprising:

receiving input at the neural network;

identifying a classification of the input at the neural networking utilizing a forward propagation, wherein the forward propagation includes utilizing a root-finding procedure to identify a fixed point associated with a parameter of the neural network;

computing a prediction vector by utilizing a linear transformation of the fixed point; and

outputting the classification of the input in response to a maximum value of the prediction vector.

11. The computer-implemented method of claim 10 , wherein performing the forward propagation includes utilizing a Forward-Backward splitting method.

12. The computer-implemented method of claim 10 , wherein performing the forward propagation includes utilizing a Peaceman-Rachford splitting method.

13. The computer-implemented method of claim 10 , wherein the input at the neural network includes image data received from one or more cameras.

14. The computer-implemented method of claim 10 , wherein the fixed point is identified utilizing a non-linear function.

15. A system for training a neural network, comprising:

an input interface for accessing input data for the neural network; and

a processor in communication with the input interface, the processor programmed to:

receive the input data at the neural network; and

output a trained neural network utilizing a forward propagation and a backward propagation, wherein the forward propagation includes utilizing a root-finding procedure to identify a fixed point associated with one or more parameters of the neural network, wherein the backward propagation includes identifying a derivative of a loss associated with the parameters of the network.

16. The system of claim 15 , wherein performing the backward propagation includes utilizing a Forward-Backward splitting method.

17. The system of claim 15 , wherein performing the backward propagation includes utilizing a Peaceman-Rachford splitting method.

18. The system of claim 15 , wherein the processor is further programmed to adjust the parameters of the neural network utilizing a gradient associated with the parameters.

19. The system of claim 15 , wherein the processor is further programmed to run the forward propagation for each minibatch in response to an index.

20. The system of claim 15 , wherein the neural network is a convolutional neural network.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Apr 21, 2020
From: WINSTON, EZRA; KOLTER, JEREMY; SAHU, ANIT KUMAR
To: ROBERT BOSCH GMBH
Reel/Frame 052449/0282 →
Continuity (1)
Related Publication 20210326663A1 · Oct 21, 2021