# US 2023/0054467 A1 · Prediction of classification of an unknown input by a trained neural network

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2023/0054467 A1
- Document ID: US20230054467A1
- Status: Published application
- Kind: A1
- Application number: 17/409,222
- Filed: 2021-08-23
- Published: 2023-02-23
- Earliest claimed: 2021-08-23
- Assignees: Raytheon Company
- Applicants: Raytheon Company
- Inventors: John E. Mixter
- Web page: https://patentgrep.com/patent/US20230054467A1

## CPC

- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A system and method are provided for classifying of an unknown object by a trained neural network. The method includes receiving unknown input data that is not classified. For each class of a set of candidate classes for which the neural network has been trained, the method further includes retraining the neural network from its trained state until a prediction can be made for classifying the input data to the class by an inference procedure and determining an amount of effort exerted for retraining the class. The method further includes selecting a class of the group of classes for which the least amount of effort was exerted and outputting the selected class as the predicted class to which the input data is predicted to be classified.

## Claims

### Claim 1

A method of classifying an unknown object by a trained neural network, the method comprising:
receiving unknown input data that is not classified;
for each class of a set of candidate classes for which the neural network has been trained:
retraining the neural network from its trained state until a prediction can be made for classifying the input data to the class by an inference procedure; and
determining an amount of effort exerted for retraining the class;
selecting a class of the group of classes for which the least amount of effort was exerted; and
outputting the selected class as the predicted class to which the input data is predicted to be classified.

### Claim 2 (depends on 1)

The method of claim 1, wherein retraining the neural network comprises:
accessing a copy of the trained neural network as trained; and
iteratively performing backpropagation in the trained neural network as copied using the class using the class as the target class until the prediction can be made for classifying the input data to the class.

### Claim 3 (depends on 2)

The method of claim 2, wherein determining the amount of effort exerted for retraining the class includes counting the number of iterations performed.

### Claim 4 (depends on 3)

The method of claim 3, wherein selecting the class of the group of classes for which the least amount of effort was exerted includes selecting the class for which the least number of iterations were performed.

### Claim 5 (depends on 1)

The method of claim 1, wherein:
retraining the trained neural network further comprises determining when the prediction can be made for classifying the input data to the class;
the inference procedure is a forward propagation procedure that outputs a predicted class, and
it is determined the prediction can be made for classifying the input data to the class when the inference procedure outputs the class as the predicted class.

### Claim 6 (depends on 5)

The method of claim 5, wherein the inference procedure further outputs a prediction strength associated with the predicted class that indicates a level of confidence in the predicted class, and wherein determining when the prediction can be made for classifying the input data to the class includes determining whether the output prediction strength satisfies a threshold criteria.

### Claim 7 (depends on 6)

The method of claim 6, wherein the predicted class is associated with a neuron of the trained neural network after retraining for which a highest value was output by the forward propagation and the prediction strength is a difference between the highest value and a second highest value output for another neuron of the trained neural network after retraining by the forward propagation, and the threshold criteria is satisfied when the different exceeds a predetermined threshold.

### Claim 8 (depends on 2)

The method of claim 2, wherein the copy of the trained neural network includes a copy of weights associated with the trained neural network once trained and before weights are changed by backpropagation during retraining.

### Claim 9

A computer system for classifying an unknown object by a trained neural network, the computer system comprising:
a memory configured to store a plurality of programmable instructions; and
a processing device in communication with the memory, wherein the processing device, upon execution of the plurality of programmable instructions is configured to:
receive unknown input data that is not classified;
or each class of a set of candidate classes for which the neural network has been trained:
retrain the neural network from its trained state until a prediction can be made for classifying the input data to the class by an inference procedure; and
determine an amount of effort exerted for retraining the class;
select a class of the group of classes for which the least amount of effort was exerted; and
output the selected class as the predicted class to which the input data is predicted to be classified.

### Claim 10 (depends on 9)

The computer system of claim 9, wherein retraining the neural network comprises:
accessing a copy of the trained neural network as trained; and
iteratively performing backpropagation in the trained neural network as copied using the class using the class as the target class until the prediction can be made for classifying the input data to the class.

### Claim 11 (depends on 10)

The computer system of claim 10, wherein determining the amount of effort exerted for retraining the class includes counting the number of iterations performed.

### Claim 12 (depends on 11)

The computer system of claim 11, wherein selecting the class of the group of classes for which the least amount of effort was exerted includes selecting the class for which the least number of iterations were performed.

### Claim 13 (depends on 9)

The computer system of claim 9, wherein:
retraining the trained neural network further comprises determining when the prediction can be made for classifying the input data to the class;
the inference procedure is a forward propagation procedure that outputs a predicted class, and
it is determined the prediction can be made for classifying the input data to the class when the inference procedure outputs the class as the predicted class.

### Claim 14 (depends on 13)

The computer system of claim 13, wherein the inference procedure further outputs a prediction strength associated with the predicted class that indicates a level of confidence in the predicted class, and wherein determining when the prediction can be made for classifying the input data to the class includes determining whether the output prediction strength satisfies a threshold criteria.

### Claim 15 (depends on 14)

The computer system of claim 14, wherein the predicted class is associated with a neuron of the trained neural network after retraining for which a highest value was output by the forward propagation and the prediction strength is a difference between the highest value and a second highest value output for another neuron of the trained neural network after retraining by the forward propagation, and the threshold criteria is satisfied when the different exceeds a predetermined threshold.

### Claim 16 (depends on 10)

The computer system of claim 10, wherein the copy of the trained neural network includes a copy of weights associated with the trained neural network once trained and before weights are changed by backpropagation during retraining.

### Claim 17

A non-transitory computer readable storage medium and one or more computer programs embedded therein, the computer programs comprising instructions, which when executed by a computer system, cause the computer system to:
receive unknown input data that is not classified;
for each class of a set of candidate classes for which the neural network has been trained:
retrain the neural network from its trained state until a prediction can be made for classifying the input data to the class by an inference procedure; and
determine an amount of effort exerted for retraining the class;
select a class of the group of classes for which the least amount of effort was exerted; and
output the selected class as the predicted class to which the input data is predicted to be classified.

### Claim 18 (depends on 17)

The non-transitory computer readable storage medium of claim 17, wherein retraining the network comprises:
accessing a copy of the trained neural network as trained; and
iteratively performing backpropagation in the trained neural network as copied using the class using the class as the target class until the prediction can be made for classifying the input data to the class.

### Claim 19 (depends on 18)

The non-transitory computer readable storage medium of claim 18, wherein determining the amount of effort exerted for retraining the class includes counting the number of iterations performed.

### Claim 20 (depends on 19)

The non-transitory computer readable storage medium of claim 19, wherein selecting the class of the group of classes for which the least amount of effort was exerted includes selecting the class for which the least number of iterations were performed.
