Learning apparatus, generation method, moving object system, program, and storage medium

Abstract

To provide a technique capable of efficiently training a machine learning model that classifies an object as any of a plurality of classifications.SOLUTION: A learning apparatus generates a predetermined machine learning model capable of classifying an object included in data to be processed as any of a plurality of classifications. The learning apparatus includes model generation means for generating the trained predetermined machine learning model using data to be processed to which a label indicating a classification of the object is not assigned. The model generation means classifies the data to be processed using a trained first machine learning model capable of classifying regarding a first classification among the plurality of classifications, classifies the data to be processed using a trained second machine learning model capable of classifying regarding a second classification among the plurality of classifications, and trains the predetermined machine learning model on the basis of classification results of the predetermined machine learning model, the first machine learning model, and the second machine learning model for the data to be processed.SELECTED DRAWING: Figure 1

Description

[Technical field]

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

[0002] In recent years, semi-supervised learning has been known as a method for appropriately learning a model even when there is a small amount of data with correct labels set (Patent Document 1).

[0003] In the technology of Patent Document 1, a trained model A is generated by supervised learning based on data with a correct answer label, and a correct answer label for unlabeled data is inferred using the model A, and the inferred correct answer label is set for data whose reliability at the time of inference is equal to or higher than a threshold. After that, a trained model B is generated by performing supervised learning (i.e., by semi-supervised learning) based on the data with the correct answer label and the data without the correct answer label to which the inferred correct answer label has been set. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent Publication No. 2022-43923 Summary of the Invention [Problem to be solved by the invention]

[0005] Incidentally, training data in which correct labels (e.g., targets in an image) are assigned to data (e.g., image data) are being provided by various entities such as companies and organizations. However, the classification set as the correct label varies depending on the training data. For example, in training data for learning a classifier, there may be a dataset in which the assigned classifications are dogs and cats, and a dataset in which the assigned classifications are dogs, elephants, and monkeys. When generating a machine learning model that classifies subjects in an image (e.g., into one of dogs, birds, and rabbits), a dataset to which a necessary classification is not assigned (e.g., only the classifications of dogs and cats, or birds and cats, are assigned) cannot be used as training data. In addition, Patent Document 1 did not consider the case of handling a model that classifies a target (object) into one of multiple classifications.

[0006] The present invention has been made in consideration of the above-mentioned problems, and has an object to provide a technology for efficiently training a machine learning model that classifies an object into one of a number of categories. [Means for solving the problem]

[0007] According to the present invention, A learning device that generates a predetermined machine learning model capable of classifying an object included in processing target data into one of a plurality of categories, A model generation means for generating the predetermined machine learning model that has been trained using processing target data to which a label indicating a classification of the object has not been assigned, The model generating means classifying the processing target data using a trained first machine learning model capable of classifying into a first classification among the plurality of classifications; classifying the processing target data using a trained second machine learning model capable of classifying a second classification among the plurality of classifications; A learning device is provided, characterized in that the specified machine learning model is trained based on the classification results of the specified machine learning model, the first machine learning model, and the second machine learning model for the data to be processed. Effect of the Invention

[0008] According to the present invention, it is possible to provide a technology for efficiently training a machine learning model that classifies an object into one of a number of categories. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a block diagram showing an example of a functional configuration of a server as an example of a learning device according to an embodiment; [Diagram 2] FIG. 1 is a diagram for explaining a learning method of a target model according to an embodiment; [Diagram 3] FIG. 1 is a diagram for explaining an example of correct answer data according to an embodiment; [Figure 4] FIG. 1 is a diagram for explaining an example of an output of a machine learning model according to an embodiment; [Diagram 5] 1 is a flowchart showing a series of operations for learning a target model according to an embodiment. [Figure 6] 1 is a flowchart showing a series of operations for performing inference using a target model according to an embodiment. [Figure 7] FIG. 1 is a block diagram showing an example of a functional configuration of a moving object according to an embodiment; [Figure 8] FIG. 1 is a diagram showing a main configuration for controlling travel of a moving object according to an embodiment; DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Hereinafter, the embodiments will be described in detail with reference to the accompanying drawings. Note that the following embodiments do not limit the invention according to the claims, and not all combinations of features described in the embodiments are essential to the invention. Two or more features among the multiple features described in the embodiments may be arbitrarily combined. In addition, the same reference numbers are used for the same or similar configurations, and duplicated descriptions are omitted.

[0011] <Server configuration> Next, an example of the functional configuration of the server will be described with reference to Fig. 1. Each of the functional blocks described with reference to the following figures may be integrated or separated, and the functions described may be realized by a different block. Also, what is described as hardware may be realized by software, and vice versa.

[0012] 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 causes the functions of each unit constituting the control unit 104 to be fulfilled by the CPU 110 expanding and executing a computer program stored in the ROM 112 or the storage unit 103 into the RAM 111. In addition to the CPU 110, the control unit 104 may further include a GPU or dedicated hardware for executing machine learning processing.

[0013] The learning data generation unit 113, as described in detail below, acquires data necessary for learning a model from various training data datasets to generate a target dataset. Each of the various datasets is a training data dataset for generating an existing machine learning model by supervised learning. These datasets include processing target data (e.g., image data) and labels (correct answer labels) indicating classifications and positions of objects included in the processing target data. Each dataset is generated for a specific purpose when it is generated. For this reason, the types of objects included in the datasets are different. One training data dataset is a dataset for classifying objects into people and vehicles, and the labels indicating the classification include people and vehicles. However, this training data dataset does not include a label indicating a bicycle. Another training data dataset is a dataset for classifying objects into people and signs, and the labels indicating the classification include people and signs. However, this training data dataset does not include a vehicle label.

[0014] Assume that the machine learning model (also referred to as the target model) to be learned, which will be described below, is a machine learning model that classifies objects in image data into, for example, vehicles and signs. In this case, the learning data generation unit 113 analyzes the above-mentioned various training data datasets and selects training data to which at least one of a label indicating a vehicle and a label indicating a sign is assigned from among these datasets. Then, the learning data generation unit 113 generates a new dataset (i.e., a target dataset) that includes at least the training data to which a label indicating a vehicle is assigned and data to which a label indicating a sign is assigned.

[0015] The model generation unit 114 executes the learning stage process of the machine learning model according to this embodiment. The machine learning model performs calculations of a deep learning algorithm using a deep neural network (DNN), for example, to classify objects (also called targets) included in the image data. Note that the objects may be pedestrians, vehicles, bicycles, signs, roads, animals, buildings, traffic lights, white or yellow lines drawn on the road, and the like included in the image.

[0016] The machine learning model is put into a trained state by performing a learning stage process described below by the model generation unit 114. Then, by inputting unknown data (e.g., new image data) to the trained machine learning model, it is possible to classify objects for the unknown data (inference stage process).

[0017] The model processing unit 115 executes the inference stage processing of the machine learning model. The inference stage processing is executed when the server 100 executes the inference processing using the trained machine model. The server 100 can execute the trained model on the server 100 side and transmit the inference result to an external device such as a moving body (such as a vehicle) or an information processing device. Alternatively, the inference stage processing using the machine learning model may be performed in the moving body. When the inference stage processing using the machine learning model is performed in the moving body, the model providing unit 116 provides information on the trained model to an external device such as the moving body.

[0018] When an inference process using a trained machine learning model is executed in a mobile body, the model providing unit 116 transmits information on the trained model trained in the server 100 to the mobile body. For example, when the mobile body receives information on the trained model from the server 100, the mobile body updates the machine learning model in the vehicle to the latest machine learning model and performs classification processing (inference processing) of objects using the latest machine learning model. The information on the trained machine learning model includes version information of the trained model, and when the machine learning model is a neural network, information such as weight coefficients and hyperparameters of the trained neural network.

[0019] The communication unit 101 is a communication device including, for example, a communication circuit, and communicates with an external device such as a mobile object through a network such as the Internet. The communication unit 101 may receive processing target data (for example, a real image) transmitted from an external device such as a mobile object. The communication unit 101 acquires a part of the above-mentioned training data from an external server or the like through the network. The communication unit 101 may also transmit information on a trained machine learning model to the mobile object at a predetermined timing or cycle. 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 or a semiconductor memory. The storage unit 103 stores training data described below, a program executed by the CPU 110, other data, and the like.

[0020] <Target model learning method> Next, a method for learning a target model according to the embodiment will be described with reference to Fig. 2. Note that in the example shown in Fig. 2, a case will be described in which a machine learning model C serving as a target model is trained to classify objects in data to be processed into vehicles or signs.

[0021] The data set A201, data set B202, and data set N203 shown in FIG. 2 are data sets of existing training data. That is, these data sets are data sets of training data for generating an existing machine learning model by supervised learning. The data set A201 is, for example, a data set of training data for classifying objects into people and vehicles, and the labels indicating the classification include people and vehicles. The target model classifies objects into vehicles or signs, and the classification mode is different from the classification mode of the data set A. For this reason, the target model cannot be trained by using the data set A as it is. In this embodiment, in order to effectively utilize such existing training data sets with different classification modes, data and label information required for training the target model are selectively acquired from these existing data sets to generate a new training data dataset (target data set 210).

[0022] 3 shows an example of a label of the dataset A201. For an image ID included in the dataset A201, information indicating the classification of an object in the image and information indicating the position and size of the object (i.e., the bounding box of the object) are included. For example, the learning data generation unit 113 acquires data classified as a vehicle and its label information from the dataset A201 to generate a target dataset 210 including data 211. The data 211 is used to train a machine learning model A220 that classifies objects as vehicles.

[0023] Similarly, the learning data generation unit 113 acquires data classified as signs and its label information from the data set B202 and the data set N203, and generates data 212. The data 212 is used to train a machine learning model B221 that classifies objects into signs. Note that, as an example, the data set B202 is a data set of training data for classifying objects into people and signs, and the data set N203 is a data set of training data for classifying objects into people, signs, and bicycles.

[0024] At this time, as an example, the learning data generation unit 113 can make the labels of the data 211 and data 212 included in the target dataset 210 include information indicating the classification of the object and information indicating the position and size of the object (i.e., the bounding box of the object), as shown in FIG. 3.

[0025] In this way, the learning data generation unit 113 selects training data to which at least one of a label indicating a vehicle and a label indicating a sign is assigned from among the training data to which labels are assigned, thereby enabling the learning data generation unit 113 to generate a new data set including at least training data to which a label indicating a vehicle is assigned and data to which a label indicating a sign is assigned.

[0026] The learning data generation unit 113 acquires training data for each classification from a plurality of existing training data in a cross-sectional manner, and acquires training data so that the amount of training data for each classification is a predetermined sufficient amount. In this way, it becomes easier to eliminate a shortage of samples for some classification labels and bias in data. In the case where there is a bias in the number of labels, such as people and vehicles, in each existing data set, there is a possibility that the accuracy of some classifications will not be sufficiently improved due to a shortage of samples or bias in data. However, in the present embodiment, in which training data for each classification is acquired cross-sectionally from a plurality of existing training data to acquire sufficient training data, such a problem can be solved.

[0027] Next, as a preliminary step of training a target model, the model generation unit 114 trains individual machine learning models that classify objects into each category using the generated target data set 210. For example, as a first step (S1), the model generation unit 114 trains a machine learning model A220 that classifies vehicles using training data (data 211) to which a label indicating a vehicle is assigned. Also, as the first step (S1), the model generation unit 114 trains a machine learning model B221 that classifies signs using training data (data 212) to which a label indicating a sign is assigned.

[0028] By training machine learning model A220 and machine learning model B221, when unknown data is input, these machine learning models can properly classify objects in the data if the objects are vehicles or signs.

[0029] Next, in a second stage (S2), the model generation unit 114 trains a target model using unknown data 213 and 214 to which no label is added (i.e., generates a trained machine learning model). The model generation unit 114 inputs the unknown data to each of the trained machine learning model A220, the trained machine learning model B221, and the machine learning model C222. For example, the trained machine learning model A220 can output a classification result as shown in FIG. 4(a). For example, the classification result includes the presence or absence of an object, the position and size of the object area, and the probability that the object is a vehicle. Similarly, the trained machine learning model B221 can output a classification result including the presence or absence of an object, the position and size of the object area, and the probability that the object is a sign. In addition, the machine learning model C222 can output a classification result as shown in FIG. 4(b), as an example. The classification result of the machine learning model C222 includes both the probability that the object is a vehicle and the probability that the object is a sign.

[0030] The model generation unit 114 trains the machine learning model C222 based on the classification results of 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 weighting coefficient of the machine learning model C222 so that an error function defined by the classification results of the machine learning model C222 and the machine learning model A220 for unknown data decreases (i.e., the difference between the classification results decreases). In addition, the model generation unit 114 changes the weighting coefficient of the machine learning model C222 so that an error function defined by the classification results of the machine learning model C222 and the machine learning model B221 for unknown data decreases (i.e., the difference between the classification results decreases). The model generation unit 114 optimizes the machine learning model C222 by changing the weighting coefficient in this way using a plurality of unknown data or by repeatedly using a plurality of unknown data. In the above description, an example has been described in which only unknown data is used to train the machine learning model C222, but the target model may be trained using the data 211 and data 212 included in the target data set. That is, the model generation unit 114 may further train the machine learning model C222 by supervised learning. That is, the weighting coefficient of the machine learning model C222 may be changed so that the difference between the classification result of the machine learning model C222 and the label is reduced.

[0031] In this way, in the present embodiment, even if the classification manner of the existing training data is different from the classification manner of the target model, it is possible to effectively utilize the existing training data. In other words, it is possible to efficiently train a machine learning model that classifies an object into one of a plurality of classifications.

[0032] Furthermore, according to the above-described embodiment, training data corresponding to each classification can be collected as necessary without relying on other classifications. In each existing data set, if there is a bias in the number of labels, such as people and vehicles, the accuracy of some classifications may not be sufficiently improved due to a lack of the number of samples or bias in the data. In this embodiment, training data for each classification can be acquired across multiple data sets, making it easier to resolve a lack of the number of samples and bias in the data for some classification labels.

[0033] Furthermore, according to the above-described embodiment, by using multiple individual machine learning models corresponding to individual classifications, it is possible to utilize individual correct answer data from existing training data that does not match the classification pattern to train a new machine learning model with a different classification pattern.

[0034] Next, a series of operations in the learning stage of the machine learning model according to this embodiment will be described with reference to FIG. 5. The series of operations described here are performed by the control unit 104 of the server 100 by developing and executing a computer program stored in the ROM 112 or the storage unit 103 in the RAM 111, thereby exerting the functions of each unit constituting the control unit 104. The following processing will be described taking as an example a case in which the target model is trained to classify objects into a first classification and a second classification (for example, vehicles and signs). The designation of these classifications may be determined in advance or may be input by the administrator of the server (the generator of the machine learning model).

[0035] In S501, the training data generation unit 113 analyzes each data set and generates a data set including data of the first and second categories as described above. Then, in S502, the model generation unit 114 generates a trained first machine learning model (e.g., machine learning model A220) using data to which a label of the first category (e.g., vehicle) is assigned as described above. Also, in S503, the model generation unit 114 generates a trained second machine learning model (e.g., machine learning model B221) using data to which a label of the second category (e.g., sign) is assigned as described above.

[0036] In S504, the model generation unit 114 infers the unknown data with the first machine learning model as described above, and in S505, infers the unknown data with the second machine learning model as described above. The model generation unit 114 acquires the classification results as described above from each machine learning model. In addition, in S506, the model generation unit 114 generates an output of a target model (e.g., machine learning model C222) for the unknown data as described above. Then, in S507, the target model is trained so as to reduce the difference between the classification result by the target model and the classification result by the first machine learning model, and the difference between the classification result by the target model and the classification result by the second machine learning model. When the target model is sufficiently optimized, the control unit 104 ends a series of operations related to this process.

[0037] Next, a series of operations of the inference stage processing in the model processing unit 115 of the server 100 according to this embodiment will be described with reference to Fig. 6. Note that the series of operations described here are performed by the control unit 104 expanding a computer program stored in the ROM 112 or the storage unit 103 into the RAM 111 and executing it, thereby causing the functions of each unit constituting the control unit 104 to be fulfilled.

[0038] In S601, the model processing unit 115 acquires unknown processing target data from an external device such as a mobile body via a network, and inputs the acquired processing target data to a target model (for example, the machine learning model C222).

[0039] In S602, the model processing unit 115 executes the process of the inference stage by the target model, and outputs, for example, the classification result shown in Fig. 4(b). When the inference process ends, the control unit 104 ends the series of operations related to this process.

[0040] In the above example, the server 100 executes the inference process using the trained machine learning model. However, the inference process of the generated machine learning model can be executed in various devices. In the following, a case will be described in which the trained machine learning model generated by the server 100 is distributed to another device via a network, and the device executes the classification process using the distributed machine learning model. As an example, a case will be described in which the other device is, for example, a mobile body. The mobile body may include a vehicle that can run on an internal combustion engine or electric power, a two-wheeled vehicle, or an autonomously running robot.

[0041] <Mobile configuration> An example of the functional configuration of a moving body 700 according to this embodiment will be described with reference to Fig. 7. Each of the functional blocks described with reference to the following figures may be integrated or separated, and the functions described may be realized by another block. Also, what is described as hardware may be realized by software, and vice versa.

[0042] The sensor unit 701 includes a camera (imaging means) that outputs a captured image of the area in front of the moving body (or further backward or surroundings). The sensor unit 701 may further include a Lidar (Light Detection and Ranging) that outputs a distance image obtained by measuring the distance in front of the moving body (or further backward or surroundings). The captured image is used, for example, in inference processing for object classification in the model processing unit 714. The sensor unit 701 may also include various sensors that output the acceleration, position information, steering angle, etc. of the moving body 700.

[0043] The communication unit 702 is a communication device including, for example, a communication circuit, and communicates with the server 100 and surrounding traffic systems via mobile communication standardized, for example, LTE, LTE-Advanced, or so-called 5G. In addition, the communication unit 702 receives part or all of the map data, traffic information, and the like from other servers and surrounding traffic systems.

[0044] The operation unit 703 includes operation members such as buttons and a touch panel attached inside the moving body 700, as well as members that receive inputs for driving the moving body 700, such as a steering wheel and a brake pedal. The power supply unit 704 includes a battery such as a lithium ion battery, and supplies power to each unit inside the moving body 700. The power unit 705 includes, for example, an engine or a motor that generates power for running the moving body.

[0045] The driving control unit 706 controls the driving of the mobile object 700 so that the mobile object 700 drives by, for example, recognizing a sign or following a vehicle ahead based on the result of the inference process (for example, the classification result of an object) output from the model processing unit 714. In this embodiment, this driving control can be performed using a known method. In the description of this embodiment, the driving control unit 706 is illustrated as having a different configuration from the control unit 708, but may be included in the control unit 708.

[0046] The storage unit 707 includes a non-volatile large-capacity storage device such as a semiconductor memory. It temporarily stores the actual image output from the sensor unit 701 and various other sensor data output from the sensor unit 701. It also stores information on the trained machine learning model received by a model data acquisition unit 713 (described later) from, for example, the external server 100 via the communication unit 702. The information on the trained machine learning model includes version information of the trained model, and when the machine learning model is a neural network, information such as weight coefficients and hyperparameters of the trained neural network.

[0047] 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 moving object 700. The control unit 708 also acquires image data from the sensor unit 701 and executes inference processing including object classification processing and the like. The control unit 708 causes each unit of the control unit 708, such as a model processing unit 714, to exhibit its function by the CPU 710 loading a computer program stored in the ROM 712 into the RAM 711 and executing it.

[0048] The CPU 710 includes one or more processors. The RAM 711 is formed of a volatile storage medium such as a DRAM, and functions as a work memory for the CPU 710. The ROM 712 is formed of a non-volatile storage medium, and stores computer programs executed by the CPU 710 and setting values ​​for operating the control unit 708. Note that, in the following embodiment, an example will be described in which the CPU 710 executes the processing of the model processing unit 714, but the processing of the model processing unit 714 may be executed by one or more other processors (e.g., GPUs) not shown.

[0049] As described above, the model data acquisition unit 713 acquires information on the trained machine learning model from the server 100. The model processing unit 714 executes the inference stage processing of the machine learning model acquired by the model data acquisition unit 713. The inference stage processing executed by the model processing unit 714 can be performed in the same manner as the inference stage processing described above in FIG.

[0050] <Main components for controlling the movement of moving objects> Next, a main configuration for controlling the traveling of the moving body 700 will be described with reference to Fig. 8. First, the model data acquisition unit 713 acquires information on the machine learning model transmitted 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.

[0051] The sensor unit 701 captures, for example, an image ahead of the moving body 700, and outputs the captured image data at a predetermined number of frames per second. The image data output from the sensor unit 701 is input to a model processing unit 714 of the control unit 708. The image data input to the model processing unit 714 is used, for example, in object classification processing (processing at the estimation stage) for controlling the current traveling of the moving body.

[0052] The model processing unit 714 inputs image data output from the sensor unit 701, executes object classification processing, and outputs the classification result to the driving control unit 706. The classification result may be similar to the output shown in FIG. 4(b).

[0053] The traveling control unit 706 outputs a control signal to the power unit 705, for example, 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, to perform moving body control of the moving body 700. As described above, the moving body control performed by the traveling control unit 706 can be performed using a known method, so details will be omitted in this embodiment. The power unit 705 controls the generation of power in response to the control signal from the traveling control unit 706.

[0054] As described above, according to the above-described embodiment, even if the classification mode of the data set of the existing training data is different from the classification mode of the target model, the existing training data can be effectively utilized. In other words, a machine learning model that classifies a target into one of a plurality of categories can be efficiently trained.

[0055] <Summary of the embodiment> The above-described embodiments include aspects of a learning apparatus, a generating method, and a mobile system.

[0056] (Item 1) A learning device that generates a predetermined machine learning model capable of classifying an object included in processing target data into one of a plurality of categories, A model generation means for generating the predetermined machine learning model that has been trained using processing target data to which a label indicating a classification of the object has not been assigned, The model generating means classifying the processing target data using a trained first machine learning model capable of classifying into a first classification among the plurality of classifications; classifying the processing target data using a trained second machine learning model capable of classifying a second classification among the plurality of classifications; The present invention also includes a learning device that trains the specified machine learning model based on the classification results of the specified machine learning model, the first machine learning model, and the second machine learning model for the data to be processed.

[0057] According to this embodiment, a machine learning model that classifies an object into one of a plurality of categories can be efficiently trained. In addition, a given machine learning model that can classify an object into one of a plurality of categories can be trained using data that is not labeled with an object's category. This significantly increases the amount of data available for training.

[0058] (Item 2) The model generation means, before training the predetermined machine learning model, training data labeled with a label indicating the first classification, the first machine learning model is trained to classify the first classification; 2. The learning device according to item 1, characterized in that the second machine learning model that classifies the second classification is trained using training data to which a label indicating the second classification is assigned.

[0059] According to this embodiment, by using multiple individual machine learning models corresponding to the classification, it is possible to utilize individual correct answer data from existing training data that does not match the classification pattern to train a new machine learning model with a different classification pattern.

[0060] (Item 3) 3. The learning device according to item 2, further comprising a data generating means for generating a new data set including at least training data assigned with a label indicating the first classification and data assigned with a label indicating the second classification by selecting training data assigned with at least one of a label indicating the first classification and a label indicating the second classification from training data assigned with labels.

[0061] According to this embodiment, usable data from existing training data with mismatched classification aspects can be obtained to generate a data set required for learning individual classifications.

[0062] (Item 4) The learning device according to any one of items 1 to 3, characterized in that the model generation means trains the predetermined machine learning model so that a difference between each classification result between the predetermined machine learning model and the first machine learning model for the processing target data and a difference between each classification result between the predetermined machine learning model and the second machine learning model for the processing target data are reduced.

[0063] According to this embodiment, the output from the first machine learning model and the output from the second machine learning model can be used as training data to learn classification for each classification.

[0064] (Item 5) A method for generating a predetermined machine learning model capable of classifying an object included in processing target data into one of a plurality of categories, comprising: generating the predetermined machine learning model that has been trained using processing target data that is not assigned a label indicating a classification of the object; In the generating step, classifying the processing target data using a trained first machine learning model capable of classifying into a first classification among the plurality of classifications; classifying the processing target data using a trained second machine learning model capable of classifying a second classification among the plurality of classifications; A generation method comprising: training the specified machine learning model based on the classification results of the specified machine learning model, the first machine learning model, and the second machine learning model for the data to be processed.

[0065] According to this embodiment, a machine learning model that classifies an object into one of a plurality of categories can be efficiently generated. In addition, a given machine learning model that can classify an object into one of a plurality of categories can be generated using data that is not labeled with an object category.

[0066] (Item 6) A mobile system including a server and a mobile object, The server is a server that generates a predetermined machine learning model capable of classifying an object included in processing target data into one of a plurality of classifications, and includes a model generation means that generates the predetermined machine learning model that has been trained using processing target data to which a label indicating the classification of the object is not assigned; The mobile body includes a classification means for classifying objects included in processing target data acquired by the mobile body into one of a plurality of classifications by executing the generated predetermined machine learning model; The model generating means classifying the processing target data using a trained first machine learning model capable of classifying into a first classification among the plurality of classifications; classifying the processing target data using a trained second machine learning model capable of classifying a second classification among the plurality of classifications; A mobile system characterized by training the specified machine learning model based on the classification results of the specified machine learning model, the first machine learning model, and the second machine learning model for the data to be processed.

[0067] According to this embodiment, a machine learning model that classifies an object into one of a plurality of categories can be efficiently trained. In addition, a given machine learning model that can classify an object into one of a plurality of categories can be trained using data that is not labeled with an object category.

[0068] The invention is not limited to the above-described embodiment, and various modifications and variations are possible within the scope of the gist of the invention. [Explanation of symbols]

[0069] 100: server; 113: learning data generation unit; 114: model generation unit; 116: model provision unit

Claims

1. A learning device that generates a predetermined machine learning model capable of classifying an object included in processing target data into one of a plurality of categories, A model generation means for generating the predetermined machine learning model that has been trained using processing target data to which a label indicating a classification of the object has not been assigned, The model generating means classifying the processing target data using a trained first machine learning model capable of classifying into a first classification among the plurality of classifications; classifying the processing target data using a trained second machine learning model capable of classifying a second classification among the plurality of classifications; A learning device characterized by training the specified machine learning model based on each classification result of the specified machine learning model, the first machine learning model, and the second machine learning model for the processing target data.

2. The model generation means, before training the predetermined machine learning model, training data labeled with a label indicating the first classification, the first machine learning model is trained to classify the first classification; The learning device according to claim 1 , further comprising: a second machine learning model that classifies data according to the second classification, the second machine learning model being trained using training data to which a label indicating the second classification has been assigned.

3. 3. The learning device according to claim 2, further comprising a data generation means for generating a new data set including at least training data assigned with a label indicating the first classification and training data assigned with a label indicating the second classification by selecting training data assigned with at least one of a label indicating the first classification and a label indicating the second classification from training data assigned with labels.

4. The learning device according to claim 1, characterized in that the model generation means trains the specified machine learning model so that a difference between each classification result between the specified machine learning model and the first machine learning model for the processing target data, and a difference between each classification result between the specified machine learning model and the second machine learning model for the processing target data are reduced, respectively.

5. A method for generating a predetermined machine learning model capable of classifying an object included in processing target data into one of a plurality of categories, comprising: generating the predetermined machine learning model that has been trained using processing target data that is not assigned a label indicating a classification of the object; In the generating step, classifying the processing target data using a trained first machine learning model capable of classifying into a first classification among the plurality of classifications; classifying the processing target data using a trained second machine learning model capable of classifying a second classification among the plurality of classifications; A generation method characterized by training the specified machine learning model based on the classification results of the specified machine learning model, the first machine learning model, and the second machine learning model for the data to be processed.

6. A mobile system including a server and a mobile object, The server is a server that generates a predetermined machine learning model capable of classifying an object included in processing target data into one of a plurality of classifications, and includes a model generation means that generates the predetermined machine learning model that has been trained using processing target data to which a label indicating the classification of the object is not assigned; The mobile body includes a classification means for classifying objects included in processing target data acquired by the mobile body into one of a plurality of classifications by executing the generated predetermined machine learning model; The model generating means classifying the processing target data using a trained first machine learning model capable of classifying into a first classification among the plurality of classifications; classifying the processing target data using a trained second machine learning model capable of classifying a second classification among the plurality of classifications; A mobile system characterized in that the specified machine learning model is trained based on the classification results of the specified machine learning model, the first machine learning model, and the second machine learning model for the data to be processed.

7. A program for causing a computer to function as each of the means of the learning device according to any one of claims 1 to 4.

8. A storage medium for storing a program for causing a computer to function as each of the means of the learning device according to any one of claims 1 to 4.

Images