# US 2023/0410488 A1 · Predictor creation device and predictor creation method

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

- Number: US 2023/0410488 A1
- Document ID: US20230410488A1
- Status: Published application
- Kind: A1
- Application number: 18/036,212
- Filed: 2021-11-22
- Published: 2023-12-21
- Earliest claimed: 2020-12-09
- Assignees: Hitachi, Ltd.
- Applicants: Hitachi, Ltd.
- Inventors: Masaki HAMAMOTO; Masashi EGI; Masakazu TAKAHASHI; Hiroyuki NAMBA
- Web page: https://patentgrep.com/patent/US20230410488A1

## CPC

- G06V 10/80: Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level (multimodal speaker identification or verification G10L17/10)
- G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06N 3/02: Neural networks
- G06N 3/08: Learning methods
- G06N 3/0985: Hyperparameter optimisation; Meta-learning; Learning-to-learn
- G06N 5/04: Inference or reasoning models
- G06N 5/045: Explanation of inference; Explainable artificial intelligence [XAI]; Interpretable artificial intelligence
- G06N 7/023: {Learning or tuning the parameters of a fuzzy system}
- G06N 20/00: Machine learning
- G06N 20/20: Ensemble learning
- G06V 10/40: Extraction of image or video features
- G06V 10/70: using pattern recognition or machine learning (optical pattern recognition or electronic computations therefor G06V10/88)
- G06V 10/764: using classification, e.g. of video objects
- G06V 10/77: Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V 10/774: Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
- G06V 10/7788: {the supervisor being a human, e.g. interactive learning with a human teacher}
- G06V 10/82: using neural networks
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- 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

A predictor creation device including a processor configured to execute a program and a storage device that stores the program acquires a calibration target ensemble predictor obtained by combining a plurality of predictors based on a training data set which is a combination of training data and ground truth data, calculates a prediction basis characteristic related to a feature of the training data for each of the plurality of predictors, acquires an expected prediction basis characteristic related to the feature based on the prediction basis characteristic related to the feature as a result of outputting the prediction basis characteristic related to the calculated feature, determines a combination coefficient of each of the plurality of predictors based on the prediction basis characteristic related to the feature and the expected prediction basis characteristic related to the feature, and calibrates the calibration target ensemble predictor based on the combination coefficient.

## Claims

### Claim 1

A predictor creation device comprising:
a processor configured to execute a program; and
a storage device that stores the program, wherein
the processor executes
first acquisition processing of acquiring a calibration target ensemble predictor obtained by combining a plurality of predictors based on a training data set which is a combination of training data and ground truth data,
calculation processing of calculating a prediction basis characteristic related to a feature of the training data for each of the plurality of predictors,
second acquisition processing of acquiring an expected prediction basis characteristic related to the feature based on the prediction basis characteristic related to the feature as a result of outputting the prediction basis characteristic related to the feature calculated by the calculation processing,
determination processing of determining a combination coefficient of each of the plurality of predictors based on the prediction basis characteristic related to the feature and the expected prediction basis characteristic related to the feature acquired in the second acquisition processing, and
calibration processing of calibrating the calibration target ensemble predictor based on the combination coefficient determined by the determination processing.

### Claim 2 (depends on 1)

The predictor creation device according to claim 1, wherein
in the determination processing, the processor calculates a loss function related to the prediction basis characteristic of the calibration target ensemble predictor based on the prediction basis characteristic related to the feature and the expected prediction basis characteristic related to the feature, and determines a combination coefficient of the plurality of predictors based on the loss function.

### Claim 3 (depends on 2)

The predictor creation device according to claim 2, wherein
in the second acquisition processing, the processor acquires a quantitative expected prediction basis characteristic related to the feature, and
in the determination processing, the processor calculates a loss function related to a quantitative prediction basis characteristic of the calibration target ensemble predictor based on the prediction basis characteristic related to the feature and the quantitative expected prediction basis characteristic related to the feature, and determines a combination coefficient of the plurality of predictors based on the loss function.

### Claim 4 (depends on 3)

The predictor creation device according to claim 3, wherein
in the second acquisition processing, the processor acquires the quantitative expected prediction basis characteristic related to the feature input by a user operation.

### Claim 5 (depends on 3)

The predictor creation device according to claim 3, wherein
in the second acquisition processing, the processor acquires the quantitative expected prediction basis characteristic related to the feature based on an effective range of the feature set by a user operation and the prediction basis characteristic related to the feature.

### Claim 6 (depends on 3)

The predictor creation device according to claim 3, wherein
in the second acquisition processing, the processor acquires the quantitative expected prediction basis characteristic related to the feature after correction as a result of outputting the quantitative expected prediction basis characteristic related to the feature, and
in the determination processing, the processor calculates the loss function based on the prediction basis characteristic related to the feature and the quantitative expected prediction basis characteristic related to the feature, and determines a combination coefficient of the plurality of predictors based on the loss function.

### Claim 7 (depends on 2)

The predictor creation device according to claim 2, wherein
in the second acquisition processing, the processor acquires a qualitative expected prediction basis characteristic related to the feature, and
in the determination processing, the processor calculates a loss function related to a qualitative prediction basis characteristic of the calibration target ensemble predictor based on the prediction basis characteristic related to the feature and the qualitative expected prediction basis characteristic related to the feature, and determines a combination coefficient of the plurality of predictors based on the loss function.

### Claim 8 (depends on 7)

The predictor creation device according to claim 7, wherein
a loss function related to the qualitative prediction basis characteristic is a loss function related to constraints of a maximum value and a minimum value as the prediction basis characteristic.

### Claim 9 (depends on 7)

The predictor creation device according to claim 7, wherein
a loss function related to the qualitative prediction basis characteristic is a loss function related to a constraint of monotonicity as the prediction basis characteristic.

### Claim 10 (depends on 7)

The predictor creation device according to claim 7, wherein
in the second acquisition processing, the processor acquires the qualitative expected prediction basis characteristic related to the feature after correction as a result of outputting the qualitative expected prediction basis characteristic related to the feature, and
in the determination processing, the processor calculates the loss function based on the prediction basis characteristic related to the feature and the qualitative expected prediction basis characteristic related to the feature, and determines a combination coefficient of the plurality of predictors based on the loss function.

### Claim 11 (depends on 2)

The predictor creation device according to claim 2, wherein
the training data is unstructured data, and
in the second acquisition processing, the processor acquires the expected prediction basis characteristic related to the feature based on the prediction basis characteristic related to the feature, which is selected from a specific area in the unstructured data, as a result of outputting the unstructured data and the prediction basis characteristic related to the feature calculated by the calculation processing.

### Claim 12 (depends on 1)

The predictor creation device according to claim 1, wherein
the processor
executes setting processing of performing a setting for the calibration target ensemble predictor by adding one or more predictors to the ensemble predictor calibrated by the calibration processing,
determines, in the determination processing, a combination coefficient of each of the plurality of predictors based on a prediction basis characteristic of each of the plurality of predictors in the calibration target ensemble predictor set in the setting processing, and
calibrates, in the calibration processing, the calibration target ensemble predictor based on the combination coefficient of each of the plurality of predictors determined by the determination processing.

### Claim 13 (depends on 12)

The predictor creation device according to claim 12, wherein
in the setting processing, the processor performs a setting for the calibration target ensemble predictor until the calibrated ensemble predictor satisfies a predetermined condition.

### Claim 14

A predictor creation method executed by a predictor creation device including a processor configured to execute a program and a storage device that stores the program,
the predictor creation method comprising:
the processor executing first acquisition processing of acquiring a calibration target ensemble predictor obtained by combining a plurality of predictors based on a training data set which is a combination of training data and ground truth data;
the processor executing calculation processing of calculating a prediction basis characteristic related to a feature of the training data for each of the plurality of predictors;
the processor executing second acquisition processing of acquiring an expected prediction basis characteristic related to the feature based on the prediction basis characteristic related to the feature as a result of outputting the prediction basis characteristic related to the feature calculated by the calculation processing;
the processor executing determination processing of determining a combination coefficient of each of the plurality of predictors based on the prediction basis characteristic related to the feature and the expected prediction basis characteristic related to the feature acquired in the second acquisition processing; and
the processor executing calibration processing of calibrating the calibration target ensemble predictor based on the combination coefficient determined by the determination processing.
