# US 2021/0174142 A1 · Artificial intelligence apparatus for generating training data for artificial intelligence model and method thereof

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

- Number: US 2021/0174142 A1
- Document ID: US20210174142A1
- Status: Published application
- Kind: A1
- Application number: 16/730,669
- Filed: 2019-12-30
- Published: 2021-06-10
- Earliest claimed: 2019-12-09
- Assignees: LG Electronics Inc.
- Applicants: LG Electronics Inc.
- Inventors: Jongwoo HAN
- Web page: https://patentgrep.com/patent/US20210174142A1

## CPC

- G06N 20/10: using kernel methods, e.g. support vector machines [SVM]
- G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06N 3/045: Combinations of networks
- G06N 3/0499: Feedforward networks
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- G06N 5/04: Inference or reasoning models
- G06V 10/764: using classification, e.g. of video objects
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

An embodiment of the present disclosure provides an artificial intelligence apparatus for generating training data including a memory configured to store an artificial intelligence model, an input interface including a microphone or a camera, and a processor configured to receive, via the input interface, input data, generate an inference result corresponding to the input data by using the artificial intelligence model, receive feedback corresponding to the inference result, determine suitability of the input data and the feedback for updating the artificial intelligence model, and generate training data based on the input data and the feedback if the input data and the feedback are determined as data suitable for updating of the artificial intelligence model.

## Claims

### Claim 1

An artificial intelligence apparatus for generating training data, comprising:
a memory configured to store an artificial intelligence model;
an input interface including a microphone or a camera; and
a processor configured to:
receive, via the input interface, input data,
generate an inference result corresponding to the input data by using the artificial intelligence model,
receive feedback corresponding to the inference result,
determine suitability of the input data and the feedback for updating the artificial intelligence model, and
generate training data based on the input data and the feedback if the input data and the feedback are determined as data suitable for updating of the artificial intelligence model.

### Claim 2 (depends on 1)

The artificial intelligence apparatus of claim 1,
wherein the processor is configured to determine the suitability of the input data and the feedback using at least one of a plurality of suitability determination methods.

### Claim 3 (depends on 2)

The artificial intelligence apparatus of claim 2,
wherein the processor is configured to determine the input data and the feedback as data suitable which updates the artificial intelligence model if a suitability determination ratio to the plurality of suitability determination results according to the plurality of suitability determination methods is equal to or greater than a reference ratio.

### Claim 4 (depends on 2)

The artificial intelligence apparatus of claim 2,
wherein the plurality of suitability determination methods include a first suitability determination method for determining the input data and the feedback as data unsuitable for updating the artificial intelligence model if the input data is an outlier, and
wherein the processor is configured to:
if the first suitability determination method is used,
calculate similarity between a first training data corresponding to a class corresponding to the feedback among the training data used for the training of the artificial intelligence model and the input data, and
determine the input data as an outlier if the similarity is less than a first reference value.

### Claim 5 (depends on 4)

The artificial intelligence apparatus of claim 4,
wherein the processor is configured to:
if the first suitability determination method is used,
map the input data and the first training data to a feature space by using the artificial intelligence model,
calculate a distance between the input data and the first training data on the feature space, and
calculate the similarity to be high as the calculated distance is short.

### Claim 6 (depends on 2)

The artificial intelligence apparatus of claim 2,
wherein the plurality of suitability determination methods include a second suitability determination method which determines the input data and the feedback as data unsuitable for updating the artificial intelligence model if the sensitivity of the artificial intelligence model for the input data is greater than a second reference value corresponding to a class corresponding to the feedback.

### Claim 7 (depends on 6)

The artificial intelligence apparatus of claim 6,
wherein the processor is configured to, if the second suitability determination method is used, calculate a Jacobian Norm Score of the artificial intelligence model for the input data as the sensitivity, and
wherein the Jacobian Norm score is an expected value of Frobenius Norm of the Jacobian matrix of the artificial intelligence model for the input data.

### Claim 8 (depends on 2)

The artificial intelligence apparatus of claim 2,
wherein the plurality of suitability determination methods include a third suitability determination method which generates a plurality of temporary inference results to the input data using the artificial intelligence model, generates an ensemble inference result from the plurality of temporary inference results, and determines the input data and the feedback as data unsuitable for updating the artificial intelligence model if the ensemble inference result is inconsistent with the feedback.

### Claim 9 (depends on 8)

The artificial intelligence apparatus of claim 8,
wherein the processor is configured to:
if the third suitability determination method is used,
apply a dropout to the artificial intelligence model to generate the plurality of temporary inference results, and
generate the temporary inference result having the highest frequency among the plurality of temporary inference results as the ensemble inference result.

### Claim 10 (depends on 2)

The artificial intelligence apparatus of claim 2,
wherein the plurality of suitability determination methods include a fourth suitability determination method which determines whether the input data and the feedback are unsuitable for updating the artificial intelligence model based on a kernel score in a class corresponding to the feedback through a kernel analysis, and
wherein the kernel score is a value obtained by dividing the sum of eigenvalues obtained through eigen decomposition by the maximum eigenvalue.

### Claim 11 (depends on 10)

The artificial intelligence apparatus of claim 10,
wherein the processor is configured to:
if the fourth suitability determination method is used,
calculate a first kernel score in a class corresponding to the feedback excluding the input data,
calculate a second kernel score in a class corresponding to the feedback including the input data, and
determine the input data and the feedback as data unsuitable for updating the artificial intelligence model if an increasing amount from the first kernel score to the second kernel score exceeds a third reference value.

### Claim 12 (depends on 1)

The artificial intelligence apparatus of claim 1,
wherein the feedback includes feedback generated from user input or feedback generated from operation information of the artificial intelligence apparatus.

### Claim 13 (depends on 1)

The artificial intelligence apparatus of claim 1,
wherein the processor is configured to update the artificial intelligence model using the generated training data.

### Claim 14

A method for generating training data comprising:
receiving input data through an input interface including a microphone or a camera;
generating an inference result corresponding to the input data using an artificial intelligence model;
receiving feedback corresponding to the inference result;
determining suitability of the input data and the feedback for updating the artificial intelligence model; and
generating training data based on the input data and the feedback if it is determined that the input data and the feedback are suitable for updating the artificial intelligence model.

### Claim 15

A recording medium having recorded thereon a program for performing a method for generating training data, the method comprising:
receiving input data through an input interface including a microphone or a camera;
generating an inference result corresponding to the input data using an artificial intelligence model;
receiving feedback corresponding to the inference result;
determining suitability of the input data and the feedback for updating the artificial intelligence model; and
generating training data based on the input data and the feedback if it is determined that the input data and the feedback are suitable for updating the artificial intelligence model.
