Learning device, generation method, mobile body system, program product, and storage medium

Abstract

The invention discloses a learning device, a generation method, a moving body system, a program product, and a storage medium, the related learning device generates a predetermined machine learning model capable of classifying an object included in processing object data into any one of a plurality of classifications, and the learning device includes a model generation unit. The model generation unit generates a predetermined machine learning model that has been learned using data to be processed to which a label indicating a classification of a target object has not been assigned, classifies the data to be processed using a first machine learning model that has been learned and can classify a first classification among a plurality of classifications, and generates a second machine learning model that has been learned using a second machine learning model that has been learned and can classify a second classification among the plurality of classifications. The processing target data is classified using a learned second machine learning model capable of classifying a second classification among the plurality of classifications, and the processing target data is classified on the basis of each classification result of a predetermined machine learning model, the first machine learning model, and the second machine learning model for the processing target data. A predetermined machine learning model is learned.

Description

Technical Field

[0001] The present invention relates to a learning device, a generation method, a mobile system, a program product and a storage medium. Background Art

[0002] In recent years, semi-supervised learning has been known as a method for appropriately learning a model even when there is little data to which correct answer labels are set (Patent Document 1).

[0003] In the technology of Patent Document 1, a learned model A is generated by supervised learning based on data with correct answer labels, and the correct answer labels for data without correct answer labels are inferred using the model A, and the inference is set to the correct answer labels of data with a reliability at the time of the inference being greater than a threshold. Then, supervised learning is performed based on the data with correct answer labels and the data with the inferred correct answer labels set in the data without correct answer labels, thereby generating a learned model B (i.e., through semi-supervised learning).

[0004] Prior art literature Patent Literature Patent Document 1: Japanese Patent Application Publication No. 2022-43923 Summary of the invention Problems to be solved by the invention In addition, training data in which correct labels (such as object marks in the image) are assigned to data (such as image data) is provided from various entities such as companies and groups. However, the classification set as the correct label varies depending on the training data. For example, in the training data used to learn a classifier, the assigned classifications can include data sets of dogs and cats, and data sets of dogs, elephants, and monkeys. In the case of generating a machine learning model that classifies the subject in the image (for example, any one of a dog, a bird, and a rabbit), a data set that is not assigned the necessary classification (for example, only the classification of dogs and cats, or birds and cats) cannot be used as training data. In addition, in Patent Document 1, there is no consideration of processing a model that classifies an object (target object) into any one of multiple classifications.

[0005] The present invention has been made in view of the above-mentioned problems, and an object of the present invention is to provide a technology for efficiently learning a machine learning model that classifies an object into any one of a plurality of classifications.

[0006] Means used to solve problems According to the present invention, there is provided a learning device that generates a predetermined machine learning model capable of classifying a target object included in processing target data into any one of a plurality of classifications, characterized in that: The learning device includes a model generation unit that generates the predetermined machine learning model that has been learned using processing object data to which a label indicating the classification of the target object is not assigned, The model generation unit classifies the processing target data using a first machine learning model that has been learned and is capable of classifying a first category among the plurality of categories. The model generation unit classifies the processing target data using a second machine learning model that has been learned and is capable of classifying a second category among the plurality of categories. The model generation unit causes the predetermined machine learning model to learn based on the classification results of the predetermined machine learning model, the first machine learning model, and the second machine learning model for the processing object data.

[0007] In addition, according to the present invention, there is provided a generation method for generating a predetermined machine learning model capable of classifying a target object included in processing target data into any one of a plurality of classifications, characterized in that: The generation method includes a generation step in which the predetermined machine learning model that has been learned is generated using processing object data to which a label indicating the classification of the target object is not assigned, In the generating step, classifying the processing target data using a first machine learning model that has been learned and is capable of classifying a first category among the plurality of categories, classifying the processing target data using a second machine learning model that has been learned and is capable of classifying a second category among the plurality of categories, Based on the classification results of the predetermined machine learning model, the first machine learning model and the second machine learning model for the processing object data, the predetermined machine learning model is learned.

[0008] Furthermore, according to the present invention, there is provided a mobile system, which includes a server and a mobile body, and is characterized in that: The server is a server for generating a predetermined machine learning model capable of classifying a target object contained in processing object data into any one of a plurality of classifications, and the server includes a model generation unit, which generates the predetermined machine learning model after learning using processing object data that is not given a label indicating the classification of the target object. The mobile body includes a classification unit that classifies a target object included in the processing target data acquired by the mobile body into any one of a plurality of classifications by executing the generated predetermined machine learning model. The model generation unit classifies the processing target data using a first machine learning model that has been learned and is capable of classifying a first category among the plurality of categories. The model generation unit classifies the processing target data using a second machine learning model that has been learned and is capable of classifying a second category among the plurality of categories. The model generation unit causes the predetermined machine learning model to learn based on the classification results of the predetermined machine learning model, the first machine learning model, and the second machine learning model for the processing object data.

[0009] According to the present invention, there is provided a program product including a program for causing a computer to function as each unit of a learning device, wherein: The learning device for generating a predetermined machine learning model capable of classifying a target object contained in processing object data into any one of a plurality of classifications comprises a model generation unit, which generates the predetermined machine learning model after learning using processing object data to which no label indicating the classification of the target object is assigned, The model generation unit classifies the processing target data using a first machine learning model that has been learned and is capable of classifying a first category among the plurality of categories. The model generation unit classifies the processing target data using a second machine learning model that has been learned and is capable of classifying a second category among the plurality of categories. The model generation unit causes the predetermined machine learning model to learn based on the classification results of the predetermined machine learning model, the first machine learning model, and the second machine learning model for the processing object data.

[0010] According to the present invention, there is provided a storage medium storing a program for causing a computer to function as each unit of a learning device, wherein: The learning device for generating a predetermined machine learning model capable of classifying a target object contained in processing object data into any one of a plurality of classifications comprises a model generation unit, which generates the predetermined machine learning model after learning using processing object data to which no label indicating the classification of the target object is assigned, The model generation unit classifies the processing target data using a first machine learning model that has been learned and is capable of classifying a first category among the plurality of categories. The model generation unit classifies the processing target data using a second machine learning model that has been learned and is capable of classifying a second category among the plurality of categories. The model generation unit causes the predetermined machine learning model to learn based on the classification results of the predetermined machine learning model, the first machine learning model, and the second machine learning model for the processing object data.

[0011] Effects of the Invention According to the present invention, it is possible to provide a technology for efficiently learning a machine learning model that classifies an object into any one of a plurality of categories. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 This is a block diagram showing a functional configuration example of a server as an example of a learning device according to an embodiment.

[0013] Figure 2 This is a diagram for explaining a method for learning a target model according to an embodiment.

[0014] Figure 3 This is a diagram for explaining an example of correct answer data according to the embodiment.

[0015] Figure 4A-4B This is a diagram illustrating an example of an output of a machine learning model according to an embodiment of the present invention.

[0016] Figure 5 This is a flowchart showing a series of actions for learning a target model according to an embodiment of the present invention.

[0017] Figure 6 This is a flowchart showing a series of operations of inference using a target model according to an embodiment.

[0018] Figure 7 This is a block diagram showing an example of a functional configuration of a moving object according to an embodiment.

[0019] Figure 8 It is a diagram showing a main configuration for use in travel control of a mobile object according to an embodiment.

[0020] Description of Reference Numerals 100: server; 113: learning data generating unit; 114: model generating unit; 116: model providing unit. DETAILED DESCRIPTION

[0021] Hereinafter, the embodiments are described in detail with reference to the accompanying drawings. In addition, the following embodiments do not limit the invention involved in the technical solution. In addition, all combinations of features described in the embodiments are not limited to the combinations required for the invention. Two or more of the multiple features described in the embodiments can be combined arbitrarily. In addition, the same or similar structures are marked with the same figure marks, and repeated descriptions are omitted.

[0022] <Server Structure> Next, refer to Figure 1 The functional structure example of the server is described. In addition, the various functional blocks described with reference to the following figures can be combined or separated, and the functions described in addition can also be implemented by other functional blocks. In addition, the components described as hardware can be implemented by software, and vice versa.

[0023] The control unit 104 includes, for example, a CPU 110, a RAM 111, and a ROM 112, and controls the operation of each unit of the server 100. The CPU 110 includes one or more processors. The control unit 104 loads and executes a computer program stored in the ROM 112 or the storage unit 103 in the RAM 111 through the CPU 110, thereby exerting the functions of each unit constituting the control unit 104. In addition to the CPU 110, the control unit 104 may also include a GPU or dedicated hardware for executing machine learning processing.

[0024] As described later, the learning data generation unit 113 obtains data required for model learning from various training data sets to generate a target data set. The various data sets are respectively data sets of training data for generating existing machine learning models through supervised learning. These data sets include processing object data (such as image data) and labels (correct labels) indicating the classification, position, etc. of the target object contained in the processing object data. Each data set is generated for an inherent purpose at the time of generation. Therefore, the categories of the target objects contained in the data set are different. A data set of a certain training data is a data set for classifying the target objects into people and vehicles, and the labels indicating the classification include people and vehicles. However, the data set of this training data does not include a label indicating a bicycle. A data set of other training data is a data set for classifying the target objects into people and logos, and the labels indicating the classification include people and logos. However, the data set of this training data does not include a label for a vehicle.

[0025] The machine learning model (also referred to as the target model) to be learned in the following description is a machine learning model that classifies the target objects of the image data into, for example, vehicles and logos. In this case, the learning data generation unit 113 analyzes the data sets of the various training data described above, and selects training data to which at least one of a label indicating a vehicle and a label indicating a logo is assigned from these data sets. Then, the learning data generation unit 113 generates a new data set (i.e., a target data set) that includes at least the training data to which a label indicating a vehicle is assigned and the data to which a label indicating a logo is assigned.

[0026] The model generation unit 114 performs the processing of the learning phase of the machine learning model involved in this embodiment. The machine learning model performs the operation of the deep learning algorithm using the deep neural network (DNN) to classify the target objects (also called landmarks) contained in the image data. In addition, the target object may also be a pedestrian, a vehicle, a bicycle, a billboard, a sign, a road, an animal, a building, a traffic light, a line drawn in white or yellow on the road, etc. contained in the image.

[0027] The machine learning model is in a state of being learned by the model generation unit 114 performing the learning phase processing described later. Then, by inputting unknown data (such as new image data) into the learned machine learning model, it is possible to classify the target object with respect to the unknown data (processing in the inference phase).

[0028] The model processing unit 115 performs processing at the inference stage of the machine learning model. When the inference processing using the learned machine learning model is performed in the server 100, the processing at the inference stage is performed. In addition, the server 100 can send the inference result of the learned model after learning is performed on the server 100 side to an external device such as a mobile body (such as a vehicle), an information processing device, etc. Alternatively, the processing at the inference stage based on the machine learning model can also be performed in the mobile body. When the processing at the inference stage based on the machine learning model is performed in the mobile body, the model providing unit 116 provides information on the learned model to an external device such as the mobile body.

[0029] When the inference processing using the learned machine learning model is executed by the mobile body, the model providing unit 116 sends the information of the learned model learned in the server 100 to the mobile body. For example, when the mobile body receives the information of the learned model from the server 100, it updates the machine learning model in the vehicle to the latest machine learning model and uses the latest machine learning model to perform classification processing (inference processing) of the target object. The information of the learned machine learning model includes the version information of the learned model and the weight coefficients, hyperparameters and other information of the learned neural network when the machine learning model is a neural network.

[0030] The communication unit 101 is, for example, a communication device including a communication circuit, and communicates with an external device such as a mobile body through a network such as the Internet. The communication unit 101 can also receive processing object data (such as an actual image) sent from an external device such as a mobile body. The communication unit 101 obtains a part of the above-mentioned training data from an external server or the like through the network. In addition, the communication unit 101 can also send information about the machine learning model that has been learned at a predetermined time or cycle to the mobile body. The power supply unit 102 supplies power to each unit in the server 100. The storage unit 103 is a non-volatile memory such as a hard disk and a semiconductor memory. The storage unit 103 stores the training data described later, the program executed by the CPU 110, other data, etc.

[0031] <Target model learning method> Next, refer to Figure 2 The learning method of the target model involved in the implementation mode is described. Figure 2 In the example shown, as an example, the machine learning model C serving as the target model is trained to classify the target object of the processing target data as a vehicle or a sign.

[0032] Figure 2 The datasets A201, B202, and N203 shown are datasets of existing training data. That is, these datasets are datasets of training data for generating existing machine learning models through supervised learning. For example, dataset A201 is a dataset of training data for classifying target objects into people and vehicles, and the labels indicating the classification include people and vehicles. The target model classifies the target object as a vehicle or a logo, and its classification method is different from the classification method of dataset A. Therefore, dataset A cannot be used directly to learn the target model. In this embodiment, in order to effectively utilize such existing training datasets with different classification methods, the data and label information required for learning the target model are selectively obtained from these existing datasets to generate a new training data set (target dataset 210).

[0033] Figure 3 An example of a label representing the dataset A201. For the image ID included in the dataset A201, information indicating the classification of the target object in the image and information indicating the position and size of the target object (i.e., the bounding box of the target object) are included. For example, the learning data generation unit 113 obtains data classified as a vehicle and its label information from the dataset A201, and generates a target dataset 210 including data 211. The data 211 is used to enable the machine learning model A220 to classify the target object as a vehicle to learn.

[0034] Similarly, the learning data generation unit 113 obtains data classified as identifiers and their label information from the data set B202 and the data set N203 to generate data 212. The data 212 is used to enable the machine learning model B221 that classifies the target object as the identifier to learn. In addition, as an example, the data set B202 is a data set of training data for classifying the target object into a person and an identifier, and the data set N203 is a data set of training data for classifying the target object into a person, an identifier, and a bicycle.

[0035] At this point, as an example, Figure 3 As shown, the learning data generating unit 113 can make the labels of the data 211 and the data 212 included in the target data set 210 include information indicating the classification of the object and information indicating the position and size of the object (ie, the bounding box of the object).

[0036] In this way, the learning data generation unit 113 selects training data assigned with at least one of a label indicating a vehicle and a label indicating a logo from the labeled training data. Thus, the learning data generation unit 113 can generate a new data set including at least training data assigned with a label indicating a vehicle and data assigned with a label indicating a logo.

[0037] The learning data generation unit 113 obtains training data for each classification from multiple existing training data in a traversal manner, and obtains training data so that the amount of training data for each classification becomes a predetermined sufficient amount. In this way, it is easy to eliminate the shortage of sample numbers and data deviations relative to some classification labels. In the case where there is a deviation in the number of labels such as people and vehicles in various existing data sets, the accuracy of some classifications may not be sufficiently improved due to the shortage of sample numbers, data deviations, etc. However, in the present embodiment of traversally obtaining training data for each classification from multiple existing training data and obtaining sufficient training data, such problems can be eliminated.

[0038] Next, as a pre-stage of learning the target model, the model generation unit 114 uses the generated target data set 210 to learn individual machine learning models that classify the target object into each category. For example, as a first stage (S1), the model generation unit 114 uses the training data (data 211) to which a label indicating a vehicle is assigned to learn the machine learning model A220 that classifies vehicles. Also, as a first stage (S1), the model generation unit 114 uses the training data (data 212) to which a label indicating an identifier is assigned to learn the machine learning model B221 that classifies the identifier.

[0039] By allowing the machine learning model A220 and the machine learning model B221 to learn, when unknown data is input, if the target object in the data is a vehicle or a sign, these machine learning models can appropriately classify the target object.

[0040] Next, as the second stage (S2), the model generation unit 114 uses the unknown data 213 and 214 without labels to learn the target model (i.e., generate a learned machine learning model). The model generation unit 114 inputs the unknown data into the learned machine learning model A 220, the learned machine learning model B 221, and the learned machine learning model C 222. For example, the learned machine learning model A 220 can output Figure 4A For example, the classification result includes the presence or absence of the target object, the location and size of the target object area, and the probability that the target object is a vehicle. Similarly, the machine learning model B221 that has been learned can output the classification result including the presence or absence of the target object, the location and size of the target object area, and the probability that the target object is a logo. In addition, as an example, the machine learning model C222 can output the following: Figure 4B The classification result of the machine learning model C222 includes both the probability that the target object is a vehicle and the probability that the target object is a logo.

[0041] The model generation unit 114 learns the machine learning model C222 based on the classification results obtained by the machine learning model A220, the machine learning model B221, and the machine learning model C222 for unknown data. For example, the model generation unit 114 changes the weight coefficient of the machine learning model C222 so that the error function defined by the classification results obtained by the machine learning model C222 and the machine learning model A220 for unknown data is reduced (that is, the difference between the classification results is reduced). In addition, the model generation unit 114 changes the weight coefficient of the machine learning model C222 so that the error function defined by the classification results of the machine learning model C222 and the machine learning model B221 for unknown data is reduced (that is, the difference between the classification results is reduced). The model generation unit 114 uses multiple unknown data or repeatedly uses multiple unknown data to optimize the machine learning model C222. In addition, in the above description, the case where only unknown data is used to learn the machine learning model C222 is used as an example, but the target model can also be learned using the data 211 and data 212 included in the target data set. That is, the model generation unit 114 can also make the machine learning model C222 learn through supervised learning. That is, the weight coefficient of the machine learning model C222 can also be changed to reduce the difference between the classification result of the machine learning model C222 and the label.

[0042] Thus, in this embodiment, even when the classification method of the existing training data is different from the classification method of the target model, the existing training data can be effectively used. In other words, the machine learning model that classifies the object into any of a plurality of classifications can be efficiently learned.

[0043] In addition, according to the above-mentioned embodiment, it is possible to collect training data corresponding to each classification as needed without relying on other classifications. In the case where there is a deviation in the number of labels such as people and vehicles in each existing data set, it may not be possible to fully improve the accuracy for a part of the classification due to insufficient number of samples, data deviation, etc. In this embodiment, it is possible to obtain training data for each classification from multiple data sets in a cross-cutting manner, so it is easy to eliminate the insufficient number of samples and data deviation for some classification labels.

[0044] Furthermore, according to the above-mentioned embodiment, by using multiple individual machine learning models corresponding to each classification, individual correct answer data in existing training data with inconsistent classification methods can be utilized to enable new machine learning models with different classification methods to be learned.

[0045] Next, refer to Figure 5 , a series of actions of processing the learning stage of the machine learning model involved in this embodiment are described. In addition, the series of actions described here are performed by the control unit 104 of the server 100 loading and executing the computer program stored in the ROM 112 or the storage unit 103 in the RAM 111, thereby exerting the functions of each part constituting the control unit 104. In addition, the following processing is described by taking the case where the target model is learned to classify the target object into a first category and a second category (for example, a vehicle and a sign) as an example. The designation of these categories can be predetermined or input by the administrator of the server (the generator of the machine learning model).

[0046] The learning data generation unit 113 analyzes each data set in S501 and generates a data set including data of the first category and the second category as described above. Then, the model generation unit 114 generates a learned first machine learning model (e.g., machine learning model A220) by using data assigned a label of the first category (e.g., vehicle) as described above in S502. In addition, the model generation unit 114 generates a learned second machine learning model (e.g., machine learning model B221) by using data assigned a label of the second category (e.g., identifier) ​​as described above in S503.

[0047] In S504, the model generation unit 114 infers unknown data through the first machine learning model as described above, and in S505, infers unknown data through the second machine learning model as described above. The model generation unit 114 obtains the classification results as described above from each machine learning model. In addition, the model generation unit 114 generates the output of the target model (e.g., machine learning model C222) for unknown data as described above in S506. Then, in S507, the target model is learned to reduce the difference between the classification result of the target model and the classification result of the first machine learning model, and the difference between the classification result of the target model and the classification result of the second machine learning model. If the target model is fully optimized, the control unit 104 ends the series of actions involved in this processing.

[0048] Next, we will refer to Figure 6 A series of operations of the inference stage of the model processing unit 115 of the server 100 according to the present embodiment are described. In addition, the series of actions described here are performed by the control unit 104 loading and executing the computer program stored in the ROM 112 or the storage unit 103 in the RAM 111, thereby exerting the functions of the various units constituting the control unit 104.

[0049] In S601 , the model processing unit 115 acquires unknown processing target data from an external device such as a mobile object via a network, and inputs the acquired processing target data into a target model (eg, a machine learning model C222 ).

[0050] In S602, the model processing unit 115 performs processing at the inference stage based on the target model, such as outputting Figure 4B When the inference process is completed, the control unit 104 ends a series of operations related to this process.

[0051] In addition, in the above example, the case where the server 100 uses the learned machine learning model to perform inference processing is described as an example. However, the inference processing of the generated machine learning model can be performed in various devices. Below, the case where the learned machine learning model generated by the server 100 is published to other devices via the network, and the classification processing is performed using the machine learning model published by the device is described. As an example, the case where other devices are, for example, mobile objects is described as an example. The mobile object may include a vehicle that can be driven by an internal combustion engine or electric power, a two-wheeled vehicle, or a robot that can drive autonomously.

[0052] <Structure of moving body> Reference Figure 7, an example of the functional structure of the mobile body 700 involved in this embodiment is described. In addition, the various functional blocks described with reference to the following figures can be combined or separated, and the functions described in addition can also be implemented by other functional blocks. In addition, the components described as hardware can be implemented by software, and vice versa.

[0053] The sensor unit 701 includes a camera (photographing unit) that outputs a photographed image obtained by photographing the front (or rearward direction, surroundings) of the moving body. The sensor unit 701 may also include a laser radar (Light Detection and Ranging) that outputs a distance image obtained by measuring the distance in front (or rearward direction, surroundings) of the moving body. The photographed image is used, for example, for inference processing of target object classification in the model processing unit 714. In addition, various sensors that output acceleration, position information, steering angle, etc. of the moving body 700 may also be included.

[0054] The communication unit 702 is, for example, a communication device including a communication circuit, and communicates with the server 100, a surrounding transportation system, etc., via a mobile communication standardized as LTE, LTE-Advanced, etc. or the so-called 5G. In addition, the communication unit 702 receives part or all of the map data, traffic information, etc. from other servers and surrounding transportation systems.

[0055] The operation unit 703 includes, in addition to operation components such as buttons and a touch panel installed in the mobile body 700, components such as a steering gear and a brake pedal that receive input for driving the mobile body 700. The power supply unit 704 includes, for example, a battery composed of a lithium-ion battery, etc., and supplies power to various components in the mobile body 700. The power unit 705 includes, for example, an engine or a motor that generates power for driving the mobile body.

[0056] The driving control unit 706 controls the driving of the mobile body 700 based on the result of the inference processing (e.g., the classification result of the target object) output from the model processing unit 714, so as to, for example, recognize a sign or follow the vehicle in front. In addition, in the present embodiment, the driving control can be performed using a known method. In addition, in the description of the present embodiment, the driving control unit 706 is illustrated as a structure different from the control unit 708, but it can also be included in the control unit 708.

[0057] The storage unit 707 includes a non-volatile large-capacity storage device such as a semiconductor memory. The actual image output from the sensor unit 701 and other various sensor data output from the sensor unit 701 are temporarily stored. In addition, the model data acquisition unit 713 described later stores, for example, information on the learned machine learning model received from the external server 100 via the communication unit 702. The information on the learned machine learning model includes the version information of the learning model and, when the machine learning model is a neural network, information such as weight coefficients and hyperparameters of the learned neural network.

[0058] The control unit 708 includes, for example, a CPU 710, a RAM 711, and a ROM 712, and controls the operation of each unit of the mobile body 700. In addition, the control unit 708 acquires image data from the sensor unit 701, and performs inference processing including classification processing of target objects. The control unit 708 loads and executes the computer program stored in the ROM 712 in the RAM 711 through the CPU 710, thereby exerting the functions of each unit such as the model processing unit 714 of the control unit 708.

[0059] The CPU 710 includes one or more processors. The RAM 711 is composed of a volatile storage medium such as a DRAM, and functions as a working memory of the CPU 710. The ROM 712 is composed of a non-volatile storage medium, and stores computer programs executed by the CPU 710, setting values ​​for operating the control unit 708, and the like. In addition, in the following embodiments, the case where the CPU 710 executes the processing of the model processing unit 714 is described as an example, but the processing of the model processing unit 714 may also be executed by one or more other processors (such as a GPU) not shown in the figure.

[0060] The model data acquisition unit 713 acquires information about the machine learning model that has been learned from the server 100 as described above. The model processing unit 714 performs inference processing of the machine learning model acquired by the model data acquisition unit 713. The inference processing performed by the model processing unit 714 can be performed in the same manner as described above. Figure 6 The processing of the inference stage mentioned above is carried out in the same way.

[0061] <Main structure for driving control of mobile body> Next, refer to Figure 8 , the main structure for driving control of the mobile body 700 is described. First, the model data acquisition unit 713 acquires information of the machine learning model sent from the server 100. The model data acquisition unit 713 updates the machine learning model executed by the model processing unit 714 based on the acquired information.

[0062] The sensor unit 701 captures the front of the moving body 700, for example, and outputs captured image data at a predetermined number of frames per second. The image data output from the sensor unit 701 is input to the model processing unit 714 of the control unit 708. The image data input to the model processing unit 714 is used, for example, for classification processing (processing in the estimation stage) of target objects for controlling the travel of the moving body at the current moment.

[0063] The model processing unit 714 inputs the image data output from the sensor unit 701 to perform object classification processing, and outputs the classification result to the driving control unit 706. The classification result can be Figure 4B The output shown is the same.

[0064] The driving control unit 706 outputs a control signal to the power unit 705, for example, to control the moving body 700 based on the classification result and various sensor information such as the acceleration and steering angle of the moving body obtained from the sensor unit 701. As described above, the moving body control performed by the driving control unit 706 can be performed using a known method, so detailed description is omitted in this embodiment. The power unit 705 controls the generation of power according to the control signal of the driving control unit 706.

[0065] As described above, according to the above-mentioned embodiment, even if the classification method of the data set of existing training data is different from the classification method of the target model, the existing training data can be effectively used. In other words, the machine learning model that classifies the object into any one of multiple categories can be efficiently learned.

[0066] <Summary of implementation methods> The above-mentioned embodiments include aspects of a learning device, a generation method, and a mobile system.

[0067] (Item 1) A learning device that generates a predetermined machine learning model capable of classifying a target object included in processing target data into any one of a plurality of classifications, characterized in that: The learning device includes a model generation unit that generates the predetermined machine learning model that has been learned using processing object data to which a label indicating the classification of the target object is not assigned, The model generation unit classifies the processing target data using a first machine learning model that has been learned and is capable of classifying a first category among the plurality of categories. The model generation unit classifies the processing target data using a second machine learning model that has been learned and is capable of classifying a second category among the plurality of categories. The model generation unit causes the predetermined machine learning model to learn based on the classification results of the predetermined machine learning model, the first machine learning model, and the second machine learning model for the processing object data.

[0068] According to this embodiment, a machine learning model that classifies an object into any one of a plurality of categories can be efficiently learned. In addition, for a predetermined machine learning model that can classify a target object into any one of a plurality of categories, it can be learned using data that is not given a label indicating the classification of the target object. Therefore, the data that can be used for learning is greatly increased.

[0069] (Item 2) The learning device according to item 1 is characterized in that Before the model generation unit causes the predetermined machine learning model to learn, Using training data to which a label representing the first category is assigned, the first machine learning model for classifying the first category is trained, The second machine learning model for classifying the second category is learned using training data that has been assigned a label representing the second category.

[0070] According to this embodiment, by using multiple individual machine learning models corresponding to the classification, it is possible to use individual correct answer data in existing training data with inconsistent classification methods to learn new machine learning models with different classification methods.

[0071] (Item 3) The learning device according to Item 2 is characterized in that the learning device further includes a data generating unit, which generates a new data set including at least training data with a label representing the first category and data with a label representing the second category by selecting training data with a label representing at least one of the label representing the first category and the label representing the second category from the labeled training data.

[0072] According to this embodiment, it is possible to obtain usable data from existing training data with different classification methods, and generate a data set required for learning each classification.

[0073] (Item 4) A learning device according to any one of items 1 to 3, characterized in that the model generation unit causes the predetermined machine learning model to learn so that the differences between the classification results of the predetermined machine learning model and the first machine learning model for the processing object data, and the differences between the classification results of the predetermined machine learning model and the second machine learning model for the processing object data are reduced respectively.

[0074] According to this embodiment, the output based on the first machine learning model and the output based on the second machine learning model can be used as training data respectively to enable learning of classification for each category.

[0075] (Item 5) A generation method for generating a predetermined machine learning model capable of classifying a target object included in processing target data into any one of a plurality of classifications, characterized in that: The generation method includes a generation step in which the predetermined machine learning model that has been learned is generated using processing object data to which a label indicating the classification of the target object is not assigned, In the generating step, classifying the processing target data using a first machine learning model that has been learned and is capable of classifying a first category among the plurality of categories, classifying the processing target data using a second machine learning model that has been learned and is capable of classifying a second category among the plurality of categories, Based on the classification results of the predetermined machine learning model, the first machine learning model and the second machine learning model for the processing object data, the predetermined machine learning model is learned.

[0076] According to this embodiment, a machine learning model that classifies an object into any one of a plurality of categories can be efficiently generated. In addition, a predetermined machine learning model that can classify a target object into any one of a plurality of categories is generated using data that is not given a label indicating the classification of the target object.

[0077] (Item 6) A mobile system, comprising a server and a mobile body, characterized in that: The server is a server for generating a predetermined machine learning model capable of classifying a target object contained in processing object data into any one of a plurality of classifications, and the server includes a model generation unit, which generates the predetermined machine learning model after learning using processing object data that is not given a label indicating the classification of the target object. The mobile body includes a classification unit that classifies a target object included in the processing target data acquired by the mobile body into any one of a plurality of classifications by executing the generated predetermined machine learning model. The model generation unit classifies the processing target data using a first machine learning model that has been learned and is capable of classifying a first category among the plurality of categories. The model generation unit classifies the processing target data using a second machine learning model that has been learned and is capable of classifying a second category among the plurality of categories. The model generation unit causes the predetermined machine learning model to learn based on the classification results of the predetermined machine learning model, the first machine learning model, and the second machine learning model for the processing object data.

[0078] According to this embodiment, a machine learning model that classifies an object into any one of a plurality of categories can be efficiently learned. In addition, a predetermined machine learning model that can classify a target object into any one of a plurality of categories is learned using data that is not given a label indicating the classification of the target object.

[0079] The present invention is not limited to the above-described embodiments, and various modifications and changes can be made within the scope of the gist of the present invention.

Claims

1. A learning device that generates a predetermined machine learning model capable of classifying a target object included in processing target data into any one of a plurality of classifications, characterized in that: The learning device includes a model generation unit that generates the predetermined machine learning model that has been learned using processing object data to which a label indicating the classification of the target object is not assigned, The model generation unit classifies the processing target data using a first machine learning model that has been learned and is capable of classifying a first category among the plurality of categories. The model generation unit classifies the processing target data using a second machine learning model that has been learned and is capable of classifying a second category among the plurality of categories. The model generation unit causes the predetermined machine learning model to learn based on the classification results of the predetermined machine learning model, the first machine learning model, and the second machine learning model for the processing object data.

2. The learning device according to claim 1, characterized in that Before the model generation unit causes the predetermined machine learning model to learn, Using training data to which a label representing the first category is assigned, the first machine learning model for classifying the first category is trained, The second machine learning model for classifying the second category is learned using training data that has been assigned a label representing the second category.

3. The learning device according to claim 2, characterized in that: The learning device also includes a data generating unit, which generates a new data set including at least training data assigned a label representing the first category and training data assigned a label representing the second category by selecting training data assigned a label representing at least any one of the first category and the second category from the labeled training data.

4. The learning device according to claim 1, characterized in that: The model generation unit causes the predetermined machine learning model to learn so that the differences between the classification results of the predetermined machine learning model and the first machine learning model for the processing object data, and the differences between the classification results of the predetermined machine learning model and the second machine learning model for the processing object data are reduced respectively.

5. A method for generating a predetermined machine learning model capable of classifying a target object included in processing target data into any one of a plurality of classifications, characterized in that: The generation method includes a generation step in which the predetermined machine learning model that has been learned is generated using processing object data to which a label indicating the classification of the target object is not assigned, In the generating step, classifying the processing target data using a first machine learning model that has been learned and is capable of classifying a first category among the plurality of categories, classifying the processing target data using a second machine learning model that has been learned and is capable of classifying a second category among the plurality of categories, Based on the classification results of the predetermined machine learning model, the first machine learning model and the second machine learning model for the processing object data, the predetermined machine learning model is learned.

6. A mobile system, comprising a server and a mobile body, characterized in that: The server is a server for generating a predetermined machine learning model capable of classifying a target object contained in processing object data into any one of a plurality of classifications, and the server includes a model generation unit, which generates the predetermined machine learning model after learning using processing object data that is not given a label indicating the classification of the target object. The mobile body includes a classification unit that classifies a target object included in the processing target data acquired by the mobile body into any one of a plurality of classifications by executing the generated predetermined machine learning model. The model generation unit classifies the processing target data using a first machine learning model that has been learned and is capable of classifying a first category among the plurality of categories. The model generation unit classifies the processing target data using a second machine learning model that has been learned and is capable of classifying a second category among the plurality of categories. The model generation unit causes the predetermined machine learning model to learn based on the classification results of the predetermined machine learning model, the first machine learning model, and the second machine learning model for the processing object data.

7. A program product comprising a program for causing a computer to function as each unit of a learning device, characterized in that: The learning device for generating a predetermined machine learning model capable of classifying a target object contained in processing object data into any one of a plurality of classifications comprises a model generation unit, which generates the predetermined machine learning model after learning using processing object data to which no label indicating the classification of the target object is assigned, The model generation unit classifies the processing target data using a first machine learning model that has been learned and is capable of classifying a first category among the plurality of categories. The model generation unit classifies the processing target data using a second machine learning model that has been learned and is capable of classifying a second category among the plurality of categories. The model generation unit causes the predetermined machine learning model to learn based on the classification results of the predetermined machine learning model, the first machine learning model, and the second machine learning model for the processing object data.

8. A storage medium storing a program for causing a computer to function as each unit of a learning device, characterized in that: The learning device for generating a predetermined machine learning model capable of classifying a target object contained in processing object data into any one of a plurality of classifications comprises a model generation unit, which generates the predetermined machine learning model after learning using processing object data to which no label indicating the classification of the target object is assigned, The model generation unit classifies the processing target data using a first machine learning model that has been learned and is capable of classifying a first category among the plurality of categories. The model generation unit classifies the processing target data using a second machine learning model that has been learned and is capable of classifying a second category among the plurality of categories. The model generation unit causes the predetermined machine learning model to learn based on the classification results of the predetermined machine learning model, the first machine learning model, and the second machine learning model for the processing object data.

Images