# US 12,118,438 B2 · Learning data generation device, learning data generation method, and program > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,118,438 B2 - Document ID: US12118438B2 - Status: Granted - Kind: B2 - Application number: 17/270,862 - Filed: 2019-09-09 - Published: 2021-07-01 - Granted: 2024-10-15 - Earliest claimed: 2018-09-20 - Assignees: Nippon Telegraph and Telephone Corporation - Applicants: Nippon Telegraph and Telephone Corporation - Inventors: Yuki Kurauchi; Naoto Abe; Hiroshi Konishi; Hitoshi Seshimo - Web page: https://patentgrep.com/patent/US12118438B2 ## CPC - G06N 20/00: Machine learning - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06N 3/09: Supervised learning - G06N 3/045: Combinations of networks - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract A training data generation device (10) according to the present invention includes a training data correction unit (11) configured to correct training data that is time-series data that indicates states of an object and in which a label corresponding to a state of the object is assigned to each section that corresponds to the state indicated by the label, wherein the training data correction unit (11) corrects a break time that divides a first section from a second section adjacent to the first section, based on time-series data in the first section and time-series data in the second section. ## Claims ### Claim 1 A training data generation device comprising: a processor configured to correct training data that is time-series data that indicates states of an object and in which a label corresponding to a state of the object is assigned to each section that corresponds to the state indicated by the label, wherein the processor corrects a break time that divides a first section from a second section adjacent to the first section, based on time-series data in the first section and time-series data in the second section, the processor corrects the break time so as to maximize a sum of a difference between an average of the time-series data in the first section and an average of the time-series data in the second section and a difference between a variance of the time-series data in the first section and a variance of the time-series data in the second section, and the training data, after correction, is used to train a model that estimates and outputs a condition of a road surface corresponding to the training data. ### Claim 2 (depends on 1) The training data generation device according to claim 1, wherein the training data includes road surface data that is detected by a sensor mounted on a moving body moving on the road surface and indicates a situation of the road surface. ### Claim 3 (depends on 2) The training data generation device according to claim 2, wherein the moving body is one of a car, a pedestrian, or a wheelchair. ### Claim 4 (depends on 3) The training data generation device according to claim 3, wherein the sensor is one of an acceleration sensor, a gyroscopic sensor, or a gravitational sensor. ### Claim 5 (depends on 4) The training data generation device according to claim 4, wherein the training data indicates at least one of a step or a slope in the road surface. ### Claim 6 (depends on 5) The training data generation device according to claim 5, wherein the model is a convolutional neural network. ### Claim 7 (depends on 5) The training data generation device according to claim 5, wherein the model is a support vector machine. ### Claim 8 A training data generation method for use in a training data generation device, the method comprising: correcting training data that is time-series data that indicates states of an object and in which a label corresponding to a state of the object is assigned to each section that corresponds to the state indicated by the label, wherein in correcting the training data, a break time that divides a first section from a second section adjacent to the first section is corrected based on time-series data in the first section and time-series data in the second section, the break time is corrected so as to maximize a sum of a difference between an average of the time-series data in the first section and an average of the time-series data in the second section and a difference between a variance of the time-series data in the first section and a variance of the time-series data in the second section, and the training data, after correction, is used to train a model that estimates and outputs a condition of a road surface corresponding to the training data. ### Claim 9 (depends on 8) The training data generation method according to claim 8, wherein the training data includes road surface data that is detected by the sensor mounted on a moving body moving on a road surface and indicates a situation of the road surface. ### Claim 10 (depends on 9) The training data generation method according to claim 9, wherein the moving body is one of a car, a pedestrian, or a wheelchair. ### Claim 11 (depends on 10) The training data generation method according to claim 10, wherein the sensor is one of an acceleration sensor, a gyroscopic sensor, or a gravitational sensor. ### Claim 12 (depends on 11) The training data generation method according to claim 11, wherein the training data indicates at least one of a step or a slope in the road surface. ### Claim 13 (depends on 12) The training data generation method according to claim 12, wherein the model is a convolutional neural network. ### Claim 14 (depends on 13) The training data generation method according to claim 13, wherein the model is a support vector machine. ### Claim 15 A non-transitory computer-readable medium encoded with computer-readable instructions that, when executed by a processor, cause the processor to perform a method comprising: correcting training data that is time-series data that indicates states of an object and in which a label corresponding to a state of the object is assigned to each section that corresponds to the state indicated by the label, wherein in correcting the training data, a break time that divides a first section from a second section adjacent to the first section is corrected based on time-series data in the first section and time-series data in the second section, the break time is corrected so as to maximize a sum of a difference between an average of the time-series data in the first section and an average of the time-series data in the second section and a difference between a variance of the time-series data in the first section and a variance of the time-series data in the second section, and the training data, after correction, is used to train a model that estimates and outputs a condition of a road surface corresponding to the training data. ### Claim 16 (depends on 15) The non-transitory computer-readable medium according to claim 15, wherein the training data includes road surface data that is detected by a sensor mounted on a moving body moving on the road surface and indicates a situation of the road surface, the moving body is one of a car, a pedestrian, or a wheelchair, and the sensor is one of an acceleration sensor, a gyroscopic sensor, or a gravitational sensor. ### Claim 17 (depends on 16) The non-transitory computer-readable medium according to claim 16, wherein the training data indicates at least one of a step or a slope in the road surface. ### Claim 18 (depends on 17) The non-transitory computer-readable medium according to claim 17, wherein the model is a convolutional neural network. ### Claim 19 (depends on 18) The non-transitory computer-readable medium according to claim 18, wherein the model is a support vector machine.