# US 2021/0342631 A1 · Information processing method and information processing system

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

- Number: US 2021/0342631 A1
- Document ID: US20210342631A1
- Status: Published application
- Kind: A1
- Application number: 17/375,288
- Filed: 2021-07-14
- Published: 2021-11-04
- Earliest claimed: 2019-06-25
- Assignees: Panasonic Intellectual Property Management Co., Ltd.
- Applicants: Panasonic Intellectual Property Management Co., Ltd.
- Inventors: Masaki TAKAHASHI; Kazunobu ISHIKAWA; Yusuke TSUKAMOTO; Yasunori ISHII
- Web page: https://patentgrep.com/patent/US20210342631A1

## CPC

- G06V 20/56: exterior to a vehicle by using sensors mounted on the vehicle
- G06F 18/2411: based on the proximity to a decision surface, e.g. support vector machines
- G06F 18/2413: based on distances to training or reference patterns
- G06F 18/2431: Multiple classes
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/08: Learning methods
- G06N 3/082: modifying the architecture, e.g. adding, deleting or silencing nodes or connections
- G06N 3/09: Supervised learning
- G06T 7/00: Image analysis
- G06V 10/764: using classification, e.g. of video objects
- G06V 10/776: Validation; Performance evaluation
- G06V 10/82: using neural networks
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

An information processing method is executed by a computer and includes acquiring a first recognition result that is output as a result of inputting sensing data to a first recognition model trained through machine learning, acquiring reference data for the sensing data, determining a difference in class of a recognition target between the first recognition result and the reference data, generating an additional class for the first recognition model when the difference satisfies a predetermined condition, and outputting the sensing data or processed data obtained by processing the sensing data as training data for the additional class.

## Claims

### Claim 1

An information processing method that is executed by a computer, the information processing method comprising:
acquiring a first recognition result that is output as a result of inputting sensing data to a first recognition model that is trained through machine learning;
acquiring reference data for the sensing data;
determining a difference in class of a recognition target between the first recognition result and the reference data;
when the difference satisfies a predetermined condition, generating an additional class for the first recognition model; and
outputting the sensing data or processed data obtained by processing the sensing data as training data for the additional class.

### Claim 2 (depends on 1)

The information processing method according to claim 1, further comprising:
acquiring a second recognition result that is output as a result of inputting the sensing data to a second recognition model different from the first recognition model; and
generating the additional class in accordance with the second recognition result.

### Claim 3 (depends on 2)

The information processing method according to claim 2,
wherein the second recognition result includes one or more candidates for the class of the recognition target, and
the additional class is generated based on the one or more candidates.

### Claim 4 (depends on 3)

The information processing method according to claim 3, further comprising:
acquiring a plurality of second recognition results from a plurality of sensing data, each of the plurality of second recognition results being the second recognition result, and each of the plurality of sensing data being the sensing data; and
when at least part of the one or more candidates is the same or similar among the plurality of second recognition results, generating the additional class in accordance with the at least part of the one or more candidates.

### Claim 5 (depends on 3)

The information processing method according to claim 3, further comprising:
acquiring a plurality of second recognition results from a plurality of sensing data, each of the plurality of second recognition results being the second recognition result, and each of the plurality of sensing data being the sensing data; and
when the one or more candidates are the same or similar in probability distribution among the plurality of second recognition results, generating the additional class in accordance with the probability distribution of the one or more candidates.

### Claim 6 (depends on 2)

The information processing method according to claim 2,
wherein the second recognition result includes an intermediate product of the second recognition model, and
the additional class is generated based on the intermediate product.

### Claim 7 (depends on 1)

The information processing method according to claim 1, further comprising:
generating the additional class in accordance with the difference.

### Claim 8 (depends on 1)

The information processing method according to claim 1,
wherein the first recognition model is a neural network model, and
the generating of the additional class includes modifying a network configuration or a parameter of the neural network model.

### Claim 9 (depends on 1)

The information processing method according to claim 1,
wherein the predetermined condition includes a condition that the first recognition result includes misdetection or undetection.

### Claim 10

An information processing system, comprising:
a first acquirer that acquires a first recognition result that is output as a result of inputting sensing data to a first recognition model trained through machine learning;
a second acquirer that acquires reference data for the sensing data;
a determiner that determines a difference in class of a recognition target between the first recognition result and the reference data; and
a class adder that, when the difference satisfies a predetermined condition, generates an additional class for the first recognition model and outputs the sensing data or processed data obtained by processing the sensing data as training data for the additional class.
