# US 11,113,569 B2 · Information processing device, information processing method, and computer program product > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,113,569 B2 - Document ID: US11113569B2 - Status: Granted - Kind: B2 - Application number: 16/533,853 - Filed: 2019-08-07 - Published: 2020-02-27 - Granted: 2021-09-07 - Earliest claimed: 2018-08-24 - Assignees: Kabushiki Kaisha Toshiba; Toshiba Digital Solutions Corporation - Applicants: Kabushiki Kaisha Toshiba; Toshiba Digital Solutions Corporation - Inventors: Ryohei Tanaka - Web page: https://patentgrep.com/patent/US11113569B2 ## CPC - G06V 30/19147: {Obtaining sets of training patterns; Bootstrap methods, e.g. bagging or boosting} - G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting - G06F 18/217: {Validation; Performance evaluation; Active pattern learning techniques} - G06F 18/2185: {the supervisor being an automated module, e.g. intelligent oracle} - G06F 18/285: {Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system} - G06V 30/10: Character recognition - G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract An information processing device according to an embodiment includes a determination unit and a first training unit. The determination unit determines whether an unlabeled data point whose class label is unknown is a non-targeted data point that is not targeted for pattern recognition. The first training unit trains a first classifier for use in the pattern recognition through semi-supervised learning using a first training dataset including unlabeled data determined not to be the non-targeted data and not including unlabeled data determined to be the non-targeted data. ## Claims ### Claim 1 An information processing device, comprising: processing circuitry configured to determine whether unlabeled data whose class label is unknown is non-targeted data that is not targeted for pattern recognition to be performed by a first classifier, the determination being performed by a second classifier; train the first classifier through semi-supervised learning using a first training dataset including a piece of the unlabeled data determined not to be the non-targeted data and not including a piece of the unlabeled data determined to be the non-targeted data; cut apiece of portion data out of a piece of the unlabeled data; generate a piece of pseudo-non-targeted data by assigning, to the piece of portion data, a non-targeted label indicating the non-targeted data; and train the second classifier by using a second training dataset including the generated pseudo-non-targeted data. ### Claim 2 (depends on 1) The information processing device according to claim 1, wherein the processing circuitry is further configured to train the first classifier through semi-supervised learning using the first training dataset including the unlabeled data determined not to be the non-targeted data and labeled data to which a class label is assigned and not including the unlabeled data determined to be the non-targeted data. ### Claim 3 (depends on 1) The information processing device according to claim 1, wherein the processing circuitry is further configured to train the first classifier through semi-supervised learning using the first training dataset and a pre-trained classifier, the first training dataset including the unlabeled data determined not to be the non-targeted data and not including labeled data to which a class label is assigned and the unlabeled data determined to be the non-targeted data. ### Claim 4 (depends on 1) The information processing device according to claim 1, wherein the processing circuitry is further configured to train the second classifier through supervised learning using the second training dataset including the pseudo-non-targeted data and labeled data to which a class label is assigned. ### Claim 5 (depends on 1) The information processing device according to claim 1, wherein the processing circuitry is further configured to train the second classifier through semi-supervised learning using the second training dataset including the pseudo-non-targeted data and the unlabeled data. ### Claim 6 (depends on 1) The information processing device according to claim 1, wherein the processing circuitry is further configured to train the second classifier through semi-supervised learning using the second training dataset including the pseudo-non-targeted data, labeled data to which a class label is assigned, and the unlabeled data. ### Claim 7 (depends on 1) The information processing device according to claim 1, wherein the processing circuitry is further configured to cut out portion data from input data, the portion data being subjected to the pattern recognition, wherein the unlabeled data are the portion data cut out by the processing circuitry. ### Claim 8 An information processing method, comprising: determining whether unlabeled data whose class label is unknown is non-targeted data that is not targeted for pattern recognition to be performed by a first classifier, the determination being performed by a second classifier; training the first classifier through semi-supervised learning using a training dataset including a piece of the unlabeled data determined not to be the non-targeted data and not including apiece of the unlabeled data determined to be the non-targeted data; cutting a piece of portion data out of apiece of the unlabeled data; generating a piece of pseudo-non-targeted data by assigning, to the piece of portion data, a non-targeted label indicating the non-targeted data; and training the second classifier by using a second training dataset including the pseudo-non-targeted data. ### Claim 9 A computer program product comprising a non-transitory computer-readable recording medium including programmed instructions, wherein the programmed instructions, when executed by a computer, cause the computer to perform a method, comprising: determining whether unlabeled data whose class label is unknown is non-targeted data that is not targeted for pattern recognition to be performed by a first classifier, the determination being performed by a second classifier; training the first classifier through semi-supervised learning using a training dataset including a piece of the unlabeled data determined not to be the non-targeted data and not including a piece of the unlabeled data determined to be the non-targeted data; cutting a piece of portion data out of a piece of the unlabeled data; generating a piece of pseudo-non-targeted data by assigning, to the piece of portion data, a non-targeted label indicating the non-targeted data; and training the second classifier by using a second training dataset including the generated pseudo-non-targeted data.