# US 10,482,351 B2 · Feature transformation device, recognition device, feature transformation method and computer readable recording medium > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 10,482,351 B2 - Document ID: US10482351B2 - Status: Granted - Kind: B2 - Application number: 15/545,769 - Filed: 2016-02-05 - Published: 2018-01-18 - Granted: 2019-11-19 - Earliest claimed: 2015-02-06 - Assignees: NEC Corporation - Applicants: NEC Corporation - Inventors: Masato Ishii - Web page: https://patentgrep.com/patent/US10482351B2 ## CPC - G06N 20/10: using kernel methods, e.g. support vector machines [SVM] - G06F 18/2135: based on approximation criteria, e.g. principal component analysis - G06F 18/22: Matching criteria, e.g. proximity measures - G06F 18/243: relating to the number of classes - G06N 20/00: Machine learning - G06V 10/761: {Proximity, similarity or dissimilarity measures} - G06V 10/7715: {Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods} - G06V 20/00: Scenes; Scene-specific elements (control of digital cameras H04N23/60) - G06V 30/196: using sequential comparisons of the image signals with a plurality of references - G06V 30/242: Division of the character sequences into groups prior to recognition; Selection of dictionaries - 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 Provided are a feature transformation device and others enabling feature transformation with high precision. The feature transformation device includes receiving means for receiving training data and test data each including a plurality of samples, optimization means for optimizing weight and feature transformation parameter based on an objective function related to the weight and the feature transformation parameter, the optimization means including weight derivation means for deriving the weight assigned to each element included in the training data and feature transformation parameter derivation means for deriving the feature transformation parameter that transforms each of the samples included in the training data or the test data, objective function derivation means for deriving a value of the objective function, the objective function derivation means including a constraint determination means for determining whether the weight satisfies a prescribed constraint and regularization means for regularizing at least one of the weight or the feature transformation parameter, and transformation means for transforming an element included in at least one of the training data or the test data based on the feature transformation parameter. ## Claims ### Claim 1 A feature transformation device, comprising: at least one processing component configured to: receive learning data and test data each including a plurality of samples; derive weight assigned to each element included in the learning data, derive feature transformation parameter transforming each of the samples included in the learning data and the test data, and optimize the weight and the feature transformation parameter based on an objective function related to the weight and the feature transformation parameter, the transformation parameter is a parameter of a transformation operation; determine whether the weight satisfies a prescribed constraint, regularize at least one of the weight or the feature transformation parameter, and derive a value of the objective function; and transform an element included in the learning data and the test data based on the feature transformation parameter by using the transformation operation, wherein the objective function includes a difference between the training data and data obtained by applying an inverse operation to the transformed learning data and a difference between the test data and data obtained by applying the inverse operation to the transformed test data, the inverse operation being an inverse operation of the transform operation. ### Claim 2 (depends on 1) The feature transformation device according to claim 1, wherein the at least one processing component further configured to: regularize at least one of the weight or the feature transformation parameter in such a way that a difference between the weights values for each of the plurality of samples included in the learning data, is included in a predetermined range. ### Claim 3 (depends on 1) The feature transformation device according to claim 1, wherein the at least one processing component further configured to: determine whether the weight for each of the plurality of samples included in the learning data is a nonnegative value, and a sum of the weight satisfies a predetermined condition. ### Claim 4 (depends on 1) The feature transformation device according to claim 1, wherein the at least one processing component further configured to: receive the learning data in which a correct class that indicates contents represented by the sample is assigned to each of the plurality of samples and the test data in which the correct class is not assigned to each of the plurality of samples. ### Claim 5 (depends on 1) The feature transformation device according to claim 1, wherein the at least one processing component further configured to: receive the learning data in which a correct class that indicates contents represented by the sample is not assigned to each of the plurality of samples, and the test data in which the correct class is assigned to each of the plurality of samples. ### Claim 6 (depends on 1) The feature transformation device according to claim 1, wherein the at least one processing component further configured to: receive learning data in which a correct class that indicates contents represented by the sample is not assigned to each of the plurality of samples, and test data in which the correct class is not assigned to each of the plurality of samples. ### Claim 7 (depends on 1) A recognition device, comprising: the feature transformation device according to claim 1; and at least one processing component configured to: perform learning of a recognition model based on the learning data transformed based on the weight and the feature transformation parameter which are derived by the feature transformation device; and perform recognition on the test data based on the recognition model and the test data transformed based on the feature transformation parameter which is derived by the feature transformation device. ### Claim 8 A feature transformation method, comprising: receiving learning data and test data each including a plurality of samples; determining whether weight assigned to each element included in the learning data satisfies a prescribed constraint, regularizing at least one of the weight or a feature transformation parameter transforming each of the samples included in the learning data and the test data, and deriving a value of an objective function, the transformation parameter is a parameter of a transformation operation; deriving the weight, deriving the feature transformation parameter, and optimizing the weight and the feature transformation parameter based on the objective function related to the weight and the feature transformation parameter; and transforming an element included in the learning data and the test data based on the feature transformation parameter by using the transformation operation, wherein the objective function includes a difference between the training data and data obtained by applying an inverse operation to the transformed learning data and a difference between the test data and data obtained by applying the inverse operation to the transformed test data, the inverse operation being an inverse operation of the transform operation. ### Claim 9 A non-transitory computer readable recording medium which causes a computer to execute: a process of receiving learning data and test data each including a plurality of samples; a process of deriving weight assigned to each element included in the learning data so as to optimize the weight, a process of deriving a feature transformation parameter transforming each of the samples included in the learning data and the test data so as to optimize the feature transformation parameter, and a process of optimizing the weight and the feature transformation parameter based on an objective function related to the weight and the feature transformation parameter, the transformation parameter is a parameter of a transformation operation; a process of determining whether the weight satisfies a prescribed constraint, a process of regularizing at least one of the weight or the feature transformation parameter, and a process of deriving a value of the objective function; and a process of transforming an element included in the learning data and the test data based on the feature transformation parameter by using the transformation operation, wherein the objective function includes a difference between the training data and data obtained by applying an inverse operation to the transformed learning data and a difference between the test data and data obtained by applying the inverse operation to the transformed test data, the inverse operation being an inverse operation of the transform operation.