Data analysis apparatus, information processing system, and data analysis method

Abstract

A task processing unit (121) having a learning processing unit (1211) that inputs a data set having a plurality of learning tasks each having a plurality of data and outputs data indicating inference model parameters or metadata, and inputs the data set and data indicating the output result by the task processing unit (121), and outputs data indicating the contribution degree of the learning task included in the data set to the output result, or data indicating a learning task characteristic in the contribution degree, is provided.

Description

Technical Field

[0001] The present disclosure relates to a data analysis apparatus, an information processing system, and a data analysis method for analyzing data.

Background Art

[0002] Patent Document 1 discloses a method of analyzing factors for prediction by a learned machine learning model, reconstructing a learning dataset, and re-performing learning. In particular, in Patent Document 1, it is shown that a search unit performs sensitivity analysis of the influence on prediction of changes in learning data, a confirmation unit presents data with a large degree of influence and asks for a user's judgment, and a configuration unit reconstructs data to create data for re-learning.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] On the other hand, when it is desired to improve the reliability of a device using multi-task learning or meta-learning, it is desirable to evaluate the reliability on a task-by-task basis. However, the conventional technique disclosed in Patent Document 1 is not intended to improve reliability, and even if used for that purpose, it cannot be evaluated on a task-by-task basis, and it is necessary to judge by looking at the contribution degree from individual data.

[0005] The present disclosure has been made to solve the above-described problems, and an object thereof is to provide a data analysis apparatus capable of analyzing data on a task-by-task basis.

Means for Solving the Problems

[0006] The data analysis device according to the present disclosure includes a task processing unit having a learning processing unit that inputs a data set having a plurality of learning tasks each having a plurality of data and outputs data indicating inference model parameters or metadata, and inputs data indicating the data set and the output result by the task processing unit, and outputs data indicating the contribution degree of the learning task included in the data set to the output result, or data indicating a learning task characteristic in terms of the contribution degree. It is characterized by including a task search unit.

Effect of the Invention

[0007] According to the present disclosure, since it is configured as described above, data can be analyzed in units of tasks.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

[0009] Hereinafter, embodiments will be described in detail with reference to the drawings. Embodiment 1. FIG. 1 is a block diagram showing a configuration example of an information processing system 1 including a data analysis apparatus 12 according to Embodiment 1. As shown in FIG. 1, for example, the information processing system 1 includes a dataset acquisition unit 11, a data analysis device 12, and an information processing unit 13. Examples of such an information processing system 1 include a character image classification system that classifies character images.

[0010] The dataset acquisition unit 11 acquires a dataset. As shown in FIG. 2, for example, the dataset has a plurality of learning tasks. Each of the plurality of learning tasks has a plurality of data.

[0011] The data analysis device 12 inputs the dataset acquired by the dataset acquisition unit 11 and analyzes the data included in the dataset in task units. A configuration example of this data analysis device 12 will be described later.

[0012] Based on the analysis results by the data analysis device 12, the information processing unit 13 performs information processing in task units. For example, when the information processing system 1 is a character image classification system, the information processing unit 13 classifies the character images indicated by the data input to the information processing system 1 in task units.

[0013] Next, a configuration example of the data analysis device 12 according to Embodiment 1 will be described with reference to FIG. 2. As shown in FIG. 2, for example, the data analysis device 12 includes a task processing unit 121 and a task search unit 122.

[0014] As shown in FIG. 2, for example, the task processing unit 121 has a learning processing unit 1211 and an inference processing unit 1212. The task processing unit 121 shown in FIG. 2 performs multi-task learning.

[0015] The learning processing unit 1211 inputs a dataset, performs learning processing based on the dataset, and outputs data indicating inference model parameters. In the example of FIG. 2, the learning processing unit 1211 inputs a dataset having learning tasks 1 to M. Also, in the example of FIG. 2, learning task 1 has data 1-1 to 1-3, and learning task M has data M-1 to data M-3.

[0016] The inference processing unit 1212 inputs data indicating the output result by the learning processing unit 1211 and data indicating an inference task, performs inference processing based on the data indicating the output result and the inference data, and outputs data indicating an inference processing result for the inference task or a performance evaluation value of the inference processing result. The inference task is a test task or a target task.

[0017] Note that in FIG. 2, the task processing unit 121 shows a case where learning processing and inference processing are performed. However, it is not limited to this. For example, the task processing unit 121 may perform additional adjustment processing and then perform inference processing after performing learning processing. That is, in this case, for example, as shown in FIG. 3, the task processing unit 121 has an adjustment processing unit 1213 in addition to the learning processing unit 1211 and the inference processing unit 1212. The task processing unit 121 shown in FIG. 3 performs multi-task learning or meta-learning.

[0018] In this case, the learning processing unit 1211 inputs a dataset, performs learning processing based on the dataset, and outputs data indicating inference model parameters or data indicating meta-parameters. That is, the learning processing unit 1211 outputs data indicating inference model parameters when performing multi-task learning, and outputs data indicating meta-parameters when performing meta-learning.

[0019] The adjustment processing unit 1213 inputs data indicating the output result by the learning processing unit 1211 and data indicating the inference task, performs additional adjustment processing based on the data indicating the output result and the inference task, and outputs data indicating the inference model parameters. Here, when the output result by the learning processing unit 1211 is data indicating the inference model parameters, the adjustment processing unit 1213 adjusts the inference model parameters in the additional adjustment processing and outputs data indicating the adjusted inference model parameters. Also, when the output result by the learning processing unit 1211 is the meta-parameters, the adjustment processing unit 1213 determines the inference model parameters in the additional adjustment processing and outputs data indicating the inference model parameters.

[0020] Also, the inference processing unit 1212 inputs data indicating the output result by the adjustment processing unit 1213 and data indicating the inference task, performs inference processing based on the data indicating the output result and the inference task, and outputs data indicating the inference processing result for the inference task or the performance evaluation value of the inference processing result.

[0021] Also, for example, the task processing unit 121 may perform only the learning processing without performing the inference processing. That is, in this case, for example, as shown in FIG. 4, the task processing unit 121 includes the learning processing unit 1211. In the task processing unit 121 shown in FIG. 4, multi-task learning or meta-learning is performed.

[0022] In this case, the learning processing unit 1211 inputs the data set, performs learning processing based on the data set, and outputs data indicating the inference model parameters or data indicating the meta-parameters. That is, when the learning processing unit 1211 performs multi-task learning, it outputs data indicating the inference model parameters, and when it performs meta-learning, it outputs data indicating the meta-parameters.

[0023] Further, for example, the task processing unit 121 may perform only learning processing and additional adjustment processing without performing inference processing. That is, in this case, for example, as shown in FIG. 5, the task processing unit 121 includes a learning processing unit 1211 and an adjustment processing unit 1213. The task processing unit 121 shown in FIG. 5 performs multi-task learning or meta-learning.

[0024] In this case, the learning processing unit 1211 inputs a data set, performs learning processing based on the data set, and outputs data indicating inference model parameters or data indicating meta-parameters. That is, when performing multi-task learning, the learning processing unit 1211 outputs data indicating inference model parameters, and when performing meta-learning, outputs data indicating meta-parameters.

[0025] The task search unit 122 performs a calculation process of calculating the contribution degree of the learning task included in the data set to the output result based on the data set and the output result by the task processing unit 121. Alternatively, the task search unit 122 performs the above calculation process and a selection process of selecting a learning task with a characteristic contribution degree based on the calculation result. At this time, in the selection process, the task search unit 122 selects at least one of the learning tasks with a large contribution degree to the output result or the learning tasks with a small contribution degree to the output result. The data indicating the contribution degree calculated by the task search unit 122 or the data indicating the learning task (identification information) selected by the task search unit 122 is output to the outside. As a result, the user can grasp the contribution degree or the learning task with a characteristic contribution degree.

[0026] Note that the "contribution degree" means, for example, the amount of change in the calculation result of the calculation process of a quantity such as a performance evaluation value or a prediction value when the data or its processing method used in the calculation is changed based on some rule, or an approximation thereof. Note that the amount of change may be changed to a change rate when the data is parameterized data.

[0027] For example, the variation amount may be the variation amount of the performance index when learning is performed excluding specific data from the data set. Also, for example, the variation amount may be the variation amount of the performance index when learning is performed by replacing specific data with dummy data in the data set. Also, for example, the variation amount may be the differential coefficient with respect to the weight parameter of the performance index when learning is performed by taking a weighted average of specific data and the data values of dummy data in the data set.

[0028] Also, the "contribution degree for each task" means, for example, the variation amount of the calculation processing result when some coordinated change is made to the entire data included in each task or its processing method, or an approximation thereof. Note that when the above data is parameterized data, the variation amount may be changed to a variation rate.

[0029] For example, the variation amount may be the variation amount of the performance index when learning is performed excluding a specific task from the data set. Also, for example, the variation amount may be the variation amount of the performance index when learning is performed by replacing a specific task with a dummy task in the data set. Also, for example, the variation amount may be the differential coefficient with respect to the weight parameter of the performance index when learning is performed by taking a weighted average of all the data included in a specific task and the data values of dummy data using a common weight parameter in the data set.

[0030] Note that when the task processing unit 121 performs a plurality of processes, for example, the task search unit 122 calculates the contribution degree between the input and output for each process, and synthesizes the contribution degrees to calculate the contribution degree of the learning task to the final output result in the task processing unit 121.

[0031] For example, consider a case where the task processing unit 121 performs learning processing and inference processing, and the task search unit 122 calculates the contribution degree of the learning task to the output result by the inference processing unit 1212. In this case, first, the task search unit 122 calculates the contribution degree of the learning task to the output result by the learning processing unit 1211. This contribution degree is regarded as the first contribution degree. Also, the task search unit 122 calculates the contribution degree of the output result by the learning processing unit 1211 to the output result by the inference processing unit 1212. This contribution degree is regarded as the second contribution degree. Then, the task search unit 122 calculates the contribution degree of the learning task to the output result by the inference processing unit 1212 by synthesizing the first contribution degree and the second contribution degree.

[0032] Also, for example, consider a case where the task processing unit 121 performs learning processing, additional adjustment processing, and inference processing, and the task search unit 122 calculates the contribution degree of the learning task to the output result by the inference processing unit 1212. In this case, first, the task search unit 122 calculates the contribution degree of the learning task to the output result by the learning processing unit 1211. This contribution degree is regarded as the first contribution degree. Also, the task search unit 122 calculates the contribution degree of the output result by the learning processing unit 1211 to the output result by the adjustment processing unit 1213. This contribution degree is regarded as the third contribution degree. Also, the task search unit 122 calculates the contribution degree of the output result by the adjustment processing unit 1213 to the output result by the inference processing unit 1212. This contribution degree is regarded as the fourth contribution degree. Then, the task search unit 122 calculates the contribution degree of the learning task to the output result by the inference processing unit 1212 by synthesizing the first contribution degree, the third contribution degree, and the fourth contribution degree.

[0033] Also, for example, consider a case where the task processing unit 121 performs learning processing and adjustment processing, and the task search unit 122 calculates the contribution degree of the learning task to the output result by the adjustment processing unit 1213. In this case, first, the task search unit 122 calculates the contribution degree of the learning task to the output result by the learning processing unit 1211. This contribution degree is defined as the first contribution degree. Also, the task search unit 122 calculates the contribution degree of the output result by the learning processing unit 1211 to the output result by the adjustment processing unit 1213. This contribution degree is defined as the third contribution degree. Then, the task search unit 122 calculates the contribution degree of the learning task to the output result by the adjustment processing unit 1213 by synthesizing the first contribution degree and the third contribution degree.

[0034] Next, an operation example of the data analysis apparatus 12 according to the first embodiment will be described with reference to FIG. 6. In the following, an operation example in the case where the task processing unit 121 has the configuration shown in FIG. 2 will be described. However, the same applies to the operation example in the case where the task processing unit 121 has the configurations shown in FIGS. 3 to 5. Further, in the following, a case where the task search unit 122 selects a learning task with a characteristic contribution degree is shown.

[0035] In the operation example of the data analysis apparatus 12 according to the first embodiment, for example, as shown in FIG. 6, the task processing unit 121 performs learning processing and inference processing (step ST101). That is, the learning processing unit 1211 inputs a data set having a plurality of learning tasks each having a plurality of data, performs learning processing based on the data set, and outputs data indicating inference model parameters. Then, the inference processing unit 1212 inputs the data indicating the output result by the learning processing unit 1211 and the data indicating the inference task, performs inference processing based on the data indicating the output result and the inference task, and outputs data indicating the inference processing result for the inference task or the performance evaluation value of the inference processing result.

[0036] Next, the task search unit 122 performs a calculation process of calculating the contribution degree of the learning task included in the data set to the output result based on the data set and the output result by the task processing unit 121 (step ST102).

[0037] At this time, first, the task search unit 122 calculates the contribution degree of the learning task to the output result by the learning processing unit 1211. This contribution degree is set as the first contribution degree. In addition, the task search unit 122 calculates the contribution degree of the output result by the learning processing unit 1211 to the output result by the inference processing unit 1212. This contribution degree is set as the second contribution degree. Then, the task search unit 122 calculates the contribution degree of the learning task to the output result by the inference processing unit 1212 by synthesizing the first contribution degree and the second contribution degree.

[0038] Next, the task search unit 122 performs a selection process of selecting a learning task with a characteristic contribution degree based on the above calculation result (step ST103). At this time, in the selection process, the task search unit 122 selects at least one of the learning tasks with a large contribution degree to the output result or the learning tasks with a small contribution degree to the output result. The data indicating the learning task (identification information) selected by this task search unit 122 is output to the outside. As a result, the user can grasp the contribution degree or the learning task with a characteristic contribution degree.

[0039] FIG. 7 is a diagram showing an example of a data set and an inference task (test task) used in the data analysis apparatus 12 according to the first embodiment. In FIG. 7, an example of a data set for classifying a plurality of types of character images is shown as the data set. In FIG. 7, when the data set is a data set for multi-task learning, the learning tasks included in the data set include only the learning data on the left side of FIG. 7. Also, in FIG. 7, when the data set is a data set for meta-learning, the learning tasks included in the data set include both the learning data on the left side of FIG. 7 and the test data on the right side. That is, in the case of meta-learning, since it is necessary to perform learning of meta-parameters for improving the test performance of the learning results of each task, the learning tasks include test data in addition to the learning data. In addition, for each learning task, images for each character type are classified respectively. Also, as shown in FIG. 7, generally, the tasks used for learning and the target task or the test task are different in type.

[0040] FIG. 8 is a diagram for explaining an operation example in the case of multi-task learning of the task processing unit 121 in the first embodiment. Note that the additional adjustment process is generally often called "additional learning" or "fine-tuning". The processing flow shown in FIG. 8 is in the order from left to right.

[0041] In FIG. 8, the first stage shows the case where the task processing unit 121 performs only the learning process, for example, as shown in FIG. 4. In this case, the learning processing unit 1211 inputs a data set, determines inference model parameters based on the learning data included in the data set, and outputs data indicating the inference model parameters.

[0042] The second stage shows the case where the task processing unit 121 performs the learning process and the inference process, for example, as shown in FIG. 2. In this case, the learning processing unit 1211 inputs a data set, determines inference model parameters based on the learning data included in the data set, and outputs data indicating the inference model parameters. In addition, the inference processing unit 1212 inputs the data indicating the inference model parameters output by the learning processing unit 1211 and the data indicating the inference task, performs an inference process or a performance evaluation of the inference process result based on the inference model parameters and the inference task, and outputs data indicating a performance evaluation value of the inference process or the inference process result.

[0043] The third stage shows the case where the task processing unit 121 performs the learning process and the additional adjustment process, for example, as shown in FIG. 5. In this case, the learning processing unit 1211 inputs a dataset, determines inference model parameters based on the learning data included in the dataset, and outputs data indicating the inference model parameters. Also, the adjustment processing unit 1213 inputs data indicating the inference model parameters output by the learning processing unit 1211 and data indicating the inference task, improves the inference model parameters based on the inference task, and outputs data indicating the inference model parameters.

[0044] The fourth stage shows a case where the task processing unit 121 performs learning processing, additional adjustment processing, and inference processing, as shown in FIG. 3, for example. In this case, the learning processing unit 1211 inputs a dataset, determines inference model parameters based on the learning data included in the dataset, and outputs data indicating the inference model parameters. Also, the adjustment processing unit 1213 inputs data indicating the inference model parameters output by the learning processing unit 1211 and data indicating the inference task, improves the inference model parameters, and outputs data indicating the inference model parameters. Also, the inference processing unit 1212 inputs data indicating the inference model parameters output by the adjustment processing unit 1213 and data indicating the inference task, performs inference processing or performance evaluation of the inference processing result based on the inference model parameters and the inference task, and outputs data indicating the performance evaluation value of the inference processing or the inference processing result.

[0045] FIG. 9 is a diagram for explaining an operation example in the case of meta-learning of the task processing unit 121 in the first embodiment. Note that, generally, learning processing is often called "meta-learning", and additional adjustment processing is often called "adaptation" or "learning". The processing flow shown in FIG. 9 is in the order from left to right. Also, in the case of meta-learning, when performing inference processing, additional adjustment processing is essential in the previous stage, so there is nothing corresponding to the second stage in the case of multi-task learning shown in FIG. 8.

[0046] In FIG. 8, the first stage shows a case where the task processing unit 121 performs only learning processing, as shown in FIG. 4 for example. In this case, the learning processing unit 1211 inputs a data set, determines inference model parameters based on the learning data included in the data set, then determines meta-parameters such that the inference performance using test data is optimal, and outputs data indicating the meta-parameters.

[0047] The third stage shows a case where the task processing unit 121 performs learning processing and additional adjustment processing, as shown in FIG. 5 for example. In this case, the learning processing unit 1211 inputs a data set, determines inference model parameters based on the learning data included in the data set, then determines meta-parameters such that the inference performance using test data is optimal, and outputs data indicating the meta-parameters. Also, the adjustment processing unit 1213 inputs data indicating the meta-parameters output by the learning processing unit 1211 and data indicating the inference task, determines inference model parameters based on the meta-parameters and the inference task, and outputs data indicating the inference model parameters.

[0048] The fourth stage shows a case where the task processing unit 121 performs learning processing, additional adjustment processing, and inference processing, as shown in FIG. 3 for example. In this case, the learning processing unit 1211 inputs a data set, determines inference model parameters based on the learning data included in the data set, then determines meta-parameters such that the inference performance using test data is optimal, and outputs data indicating the meta-parameters. Also, the adjustment processing unit 1213 inputs data indicating the meta-parameters output by the learning processing unit 1211 and data indicating the inference task, determines inference model parameters based on the meta-parameters and the inference task, and outputs data indicating the inference model parameters. In addition, the inference processing unit 1212 inputs data indicating the inference model parameters output by the adjustment processing unit 1213 and data indicating the inference task, performs inference processing or performance evaluation of the inference processing results based on the inference model parameters and the inference task, and outputs data indicating the performance evaluation value of the inference processing or the inference processing results.

[0049] FIG. 10 is a diagram for explaining an operation example of calculating the contribution degree of a learning task to the learning processing result in the task search unit 122 in the first embodiment. As shown in FIG. 10, for example, when calculating the contribution degree of the learning task to the learning processing result, the task search unit 122 calculates a sensitivity matrix A that is the differential coefficient for the perturbation parameter. At this time, the sensitivity matrix A, which is the contribution degree calculated by the task search unit 122, has a size of <number of inference model parameters> × <number of learning tasks>.

[0050] That is, for example, the task processing unit 121 determines the inference model parameters based on the loss function for the entire dataset. The loss function for the entire dataset is obtained by synthesizing the loss functions for each learning task included in the dataset for each such learning task. And in this case, the task search unit 122 calculates the differential coefficient for the perturbation parameter based on the loss function for the entire dataset described above and the loss function obtained by deforming the loss function of any one learning task with a perturbation parameter common to all the data within the learning task, and sets the differential coefficient as the contribution degree of the learning task.

[0051] Here, considering the perturbation of the loss function when the loss function (C) for all data (x) is expressed as in the following formula (1) using the inference model parameters (θ), a dataset (D1,..., D M ) having M learning tasks, and the loss function (L) for each learning task. TIFF0007710641000001.tif11166

[0052] Then, for this formula (1), for any one learning task (j) among the learning tasks included in the dataset, consider the loss function (C’) when perturbation by a perturbation parameter (ε) as in the following formula (2) is given. TIFF0007710641000002.tif12166

[0053] Here, the value of θ that minimizes C’ depends on ε. Therefore, the task search unit 122 uses the value of the differential coefficient at ε = 0 as the degree of contribution to the learning processing result of the learning task (j). Then, the task search unit 122 calculates the degree of contribution of each learning task by performing the above processing for all the learning tasks included in the dataset.

[0054] Note that the output of the learning process is represented, for example, as an implicit function of the perturbation parameter by minimizing the loss function. Therefore, the differential coefficient between the variables giving the solution to the minimization problem can be calculated using implicit function differentiation as in the prior art. That is, for example, the task processing unit 121 determines model parameters that are values that achieve the extreme value of the loss function for the entire dataset or approximate values of such values. And in this case, the task search unit 122 calculates the degree of contribution of the learning task to the inference model parameters determined by the task processing unit 121 by implicit function differentiation.

[0055] Here, in the prior art, as in the following formula (3), the situation where data is distinguished by learning tasks is not considered, and perturbation of the loss function is considered for each individual data (x). In contrast, in the data analysis apparatus 12 according to Embodiment 1, as in formula (2), perturbation of the loss function is considered for each learning task (D). That is, in the data analysis apparatus 12 according to Embodiment 1, a plurality of data are grouped into one learning task, and cooperative perturbation of all the data therein is considered. TIFF0007710641000003.tif10166

[0056] Note that FIG. 10 shows an example in which the task search unit 122 calculates the derivative coefficient by numerical differentiation. However, the task search unit 122 is not limited to this, and it is also possible to calculate an exact derivative coefficient using various mathematical facts.

[0057] Also, when the task processing unit 121 performs a plurality of processes, for example, the task search unit 122 calculates the derivative coefficient as the contribution degree between the input and output in each process for each process performed by the task processing unit 121, constructs a matrix, and calculates the matrix product of the matrices for each of the plurality of processes to synthesize the contribution degrees.

[0058] For example, as shown in FIG. 11, when the task processing unit 121 performs a learning process and an inference process, the task search unit 122 calculates a sensitivity matrix B, which is a derivative coefficient, as the contribution degree of the output result by the learning processing unit 1211 to the output result by the inference processing unit 1212 in addition to the sensitivity matrix A. At this time, the sensitivity matrix B, which is the contribution degree calculated by the task search unit 122, has a size of <number of performance evaluation values> × <number of inference model parameters>.

[0059] Then, for example, as shown in FIG. 11, the task search unit 122 calculates the contribution degree of each learning task to the output result by the inference processing unit 1212 by calculating the matrix product (B·A) of the sensitivity matrix A and the sensitivity matrix B. This contribution degree is a matrix with a size of <number of performance evaluation values> × <number of learning tasks>.

[0060] Also, for example, as shown in FIG. 12, when the task processing unit 121 performs a learning process and an additional adjustment process, the task search unit 122 calculates a sensitivity matrix C, which is a derivative coefficient, as the contribution degree of the output result by the learning processing unit 1211 to the output result by the adjustment processing unit 1213 in addition to the sensitivity matrix A. At this time, the sensitivity matrix C, which is the contribution degree calculated by the task search unit 122, has a size of <number of inference model parameters> × <number of inference model parameters>.

[0061] Then, as shown in FIG. 12 for example, the task search unit 122 calculates the matrix product (C·A) of the sensitivity matrix A and the sensitivity matrix C, thereby calculating the contribution degree of each learning task to the output result by the adjustment processing unit 1213. This contribution degree is a matrix with a size of <the number of inference model parameters>×<the number of learning tasks>.

[0062] Also, as shown in FIG. 13 for example, when the task processing unit 121 performs learning processing, additional adjustment processing, and inference processing, the task search unit 122 calculates, in addition to the sensitivity matrix A and the sensitivity matrix C, the sensitivity matrix B' which is a differential coefficient, as the contribution degree of the output result by the adjustment processing unit 1213 to the output result by the inference processing unit 1212. At this time, the sensitivity matrix B' which is the contribution degree calculated by the task search unit 122 has a size of <the number of performance evaluation values>×<the number of inference model parameters>.

[0063] Then, as shown in FIG. 13 for example, the task search unit 122 calculates the matrix product (B'·C·A) of the sensitivity matrix A, the sensitivity matrix C, and the sensitivity matrix B', thereby calculating the contribution degree of each learning task to the output result by the inference processing unit 1212. This contribution degree is a matrix with a size of <the number of performance evaluation values>×<the number of learning tasks>.

[0064] Note that in FIGS. 11 to 13, the case where the task processing unit 121 performs multi-task learning has been described as an example, but the same applies to the case where the task processing unit 121 performs meta-learning.

[0065] Here, as shown in FIG. 14A, in the prior art, the contribution degree is calculated for each piece of data. In contrast, as shown in FIG. 14B, in the data analysis device 12 according to Embodiment 1, the contribution degree is calculated for each learning task having a plurality of pieces of data. As a result, in the data analysis device 12 according to the first embodiment, it is possible to calculate the contribution degree of tasks that summarize them based on a certain standard rather than individual data, present an explanation that is more intuitive and easier for the user to understand, and determine the reliability by whether a task similar to the task to be executed on the device is presented as a task with a high contribution degree.

[0066] In the above description, the case where the information processing system 1 is a character image classification system has been described as an example. However, the information processing system 1 to which the data analysis device 12 is applied is not limited to this. The information processing system 1 may be, for example, a four-wheel torque control device, a power consumption prediction device, or a production line monitoring system.

[0067] The four-wheel torque control device is a device that inputs data indicating sensor values from the outside, recognizes the road surface condition from the sensor values in the information processing unit 13, and adjusts the torque output of each wheel of the vehicle. In the data analysis of this four-wheel torque control device, learning tasks are defined for each road surface condition such as a normal road surface, a snow road, and a gravel road, and the four-wheel torque control device learns efficient torque distribution for each. Then, when switching the control pattern, the four-wheel torque control device displays information on similar road surface conditions experienced during learning on the dashboard.

[0068] Here, in the prior art, only the sensor values used for learning can be displayed. On the other hand, in the four-wheel torque control device to which the data analysis device 12 according to the first embodiment is applied, it is possible to display in units of tasks by an abstracted character string or icon such as "snow road" or "gravel road".

[0069] The power consumption prediction device is applied to the operating environment of an air conditioning system or a production line, etc., inputs data indicating sensor values from the outside, and predicts the power consumption in the information processing unit 13. In the data analysis of this power consumption prediction device, learning tasks are defined for each condition such as the usage environment, date, and season, and the power consumption prediction device learns future power consumption from the time-series data of the power usage state in each case. Then, when the power consumption prediction device starts to be used, it notifies the user whether it has experienced conditions similar to those during learning.

[0070] Here, in the prior art, only the sensor values used for learning can be displayed. On the other hand, in the power consumption prediction device to which the data analysis device 12 according to Embodiment 1 is applied, it is possible to present experience contents based on categories such as seasons or types of devices used, in units of tasks.

[0071] In addition, the production line monitoring system is applied to an environment where people and robots work in close proximity, etc., inputs data indicating sensor values from the outside, and in the information processing unit 13, predicts human behavior patterns and stops the operation of the robot when there is a risk of contact. In the data analysis of this production line monitoring system, learning tasks are defined for each production line environment, and the production line monitoring system learns, from camera images or Lidar data, etc., in each case, the area where a person may intrude or the probability of contact with a robot in the future several seconds. Then, when the production line monitoring system is introduced, it notifies the user whether it has experienced data of a production environment similar to that during learning.

[0072] Here, in the prior art, only the sensor values used for learning can be displayed. On the other hand, in the production line monitoring system to which the data analysis device 12 according to Embodiment 1 is applied, it is possible to present experience contents based on categories such as density or type of production line, in units of tasks.

[0073] In the above description, the case where the task search unit 122 calculates the contribution degree of the learning task to the output result of the final process in the task processing unit 121 and uses it as output data has been described as an example. However, the present invention is not limited to this, and the task search unit 122 may calculate the contribution degree of the learning task to the output result of the intermediate process in the task processing unit 121 and use it as output data. For example, when the task processing unit 121 performs learning processing, additional adjustment processing, and inference processing, the task search unit 122 may calculate the contribution degree of the learning task to the learning processing and use it as output data, or may calculate the contribution degree of the learning task to the additional adjustment processing and use it as output data. As a result, the data analysis device 12 can confirm the state of the information processing system 1 in more detail.

[0074] As described above, according to the first embodiment, the data analysis device 12 includes a task processing unit 121 having a learning processing unit 1211 that inputs a data set having a plurality of learning tasks each having a plurality of data and outputs data indicating inference model parameters or metadata, and inputs the data set and the data indicating the output result by the task processing unit 121, and outputs data indicating the contribution degree of the learning task included in the data set to the output result, or data indicating a learning task characteristic of the contribution degree. Thereby, the data analysis device 12 according to the first embodiment can analyze data in units of tasks. That is, in the data analysis device 12 according to the first embodiment, by presenting explanatory information in units of tasks, it is possible to more appropriately perform reliability evaluation.

[0075] Further, according to the first embodiment, the learning processing unit 1211 outputs data indicating inference model parameters, and the task processing unit 121 has an inference processing unit 1212 that inputs the data indicating the output result by the learning processing unit 1211 and the data indicating the inference task, and outputs data indicating the inference processing result for the inference task or the performance evaluation value of the inference processing result. Also, according to Embodiment 1, the task processing unit 121 includes an adjustment processing unit 1213 that inputs data indicating the output result by the learning processing unit 1211 and data indicating the inference task, and outputs data indicating the inference model parameters. Also, according to Embodiment 1, the task processing unit 121 includes an adjustment processing unit 1213 that inputs data indicating the output result by the learning processing unit 1211 and data indicating the inference task, and outputs data indicating the inference model parameters, and an inference processing unit 1212 that inputs data indicating the output result by the adjustment processing unit 1213 and data indicating the inference task, and outputs data indicating the inference processing result for the inference task or the performance evaluation value of the inference processing result. Thereby, the data analysis apparatus 12 according to Embodiment 1 can analyze data in units of tasks.

[0076] Also, according to Embodiment 1, the task search unit 122 calculates a first contribution degree that is the contribution degree of the learning task included in the data set with respect to the output result by the learning processing unit 1211, and a second contribution degree that is the contribution degree of the output result by the learning processing unit 1211 with respect to the output result by the inference processing unit 1212, and calculates the contribution degree of the learning task with respect to the output result by the inference processing unit 1212 by synthesizing the first contribution degree and the second contribution degree. Also, according to Embodiment 1, the task search unit 122 calculates a first contribution degree that is the contribution degree of the learning task included in the data set with respect to the output result by the learning processing unit 1211, and a third contribution degree that is the contribution degree of the output result by the learning processing unit 1211 with respect to the output result by the adjustment processing unit 1213, and calculates the contribution degree of the learning task with respect to the output result by the adjustment processing unit 1213 by synthesizing the first contribution degree and the third contribution degree. Further, according to Embodiment 1, the task search unit 122 calculates a first contribution degree, which is the contribution degree of the learning task included in the dataset with respect to the output result by the learning processing unit 1211, a third contribution degree, which is the contribution degree of the output result by the learning processing unit 1211 with respect to the output result by the adjustment processing unit 1213, and a fourth contribution degree, which is the contribution degree of the output result by the adjustment processing unit 1213 with respect to the output result by the inference processing unit 1212, and calculates the contribution degree of the learning task with respect to the output result by the inference processing unit 1212 by synthesizing the first contribution degree, the third contribution degree, and the fourth contribution degree. Accordingly, the data analysis apparatus 12 according to Embodiment 1 can analyze data in units of tasks.

[0077] Further, according to Embodiment 1, the task processing unit 121 determines inference model parameters based on the loss function for the entire dataset obtained by synthesizing the loss functions for each learning task included in the dataset for each learning task, and the task search unit 122 calculates a differential coefficient for the perturbation parameter based on a loss function obtained by deforming the loss function of any one learning task with respect to the loss function for the entire dataset by a perturbation parameter common to all the data within the learning task, and sets the differential coefficient as the contribution degree of the learning task. Further, according to Embodiment 1, the task search unit 122 calculates a differential coefficient as the contribution degree between the input and output in each process performed by the task processing unit 121 to form a matrix, and synthesizes the contribution degrees by calculating the matrix product of the matrices for each of the plurality of processes. Further, according to Embodiment 1, the task processing unit 121 determines model parameters that are values that realize the extreme value of the loss function for the entire dataset or approximate values thereof, and the task search unit 122 calculates the contribution degree of the learning task with respect to the inference model parameters determined by the task processing unit 121 by implicit function differentiation. Accordingly, the data analysis apparatus 12 according to Embodiment 1 can analyze data in units of tasks.

[0078] Further, according to Embodiment 1, the information processing system 1 includes a dataset acquisition unit 11 that acquires a dataset having a plurality of learning tasks each having a plurality of data, a data analysis device 12, and an information processing unit 13 that performs information processing in task units based on the output result of the data analysis device 12. The data analysis device 12 inputs the dataset acquired by the dataset acquisition unit 11. Thereby, the information processing system 1 according to Embodiment 1 can perform information processing in task units.

[0079] Further, according to Embodiment 1, in the data analysis method, a task processing unit 121 inputs a dataset having a plurality of learning tasks each having a plurality of data and outputs data indicating inference model parameters or metadata, and a task search unit 122 inputs the dataset and the data indicating the output result of the task processing unit 121, and outputs data indicating the contribution degree of the learning task included in the dataset to the output result, or data indicating the learning task characteristic in the contribution degree. Thereby, the data analysis method according to Embodiment 1 can analyze data in task units.

[0080] Embodiment 2. The data analysis device 12 according to Embodiment 2 shows a case where a dataset is processed based on a learning task characteristic in contribution degree.

[0081] FIG. 15 is a diagram showing a configuration example of the data analysis device 12 according to Embodiment 2. In the data analysis device 12 according to Embodiment 2 shown in this FIG. 15, a data processing unit 123 and a second task processing unit 124 are added to the data analysis device 12 according to Embodiment 1 shown in FIG. 2. Regarding other configuration examples in the data analysis device 12 according to Embodiment 2 shown in FIG. 15, they are the same as the configuration example of the data analysis device 12 according to Embodiment 1, and the same reference numerals are given and their descriptions are made.

[0082] Note that the task search unit 122 outputs data indicating a learning task with a characteristic contribution degree to the data processing unit 123. The data processing unit 123 processes the data included in the data set based on the output result by the task search unit 122.

[0083] At this time, for example, the data processing unit 123 may exclude a learning task with a low contribution degree (e.g., around 0) to the learning processing result among the learning tasks included in the data set. That is, since such a learning task is considered not to affect the inference performance, it is excluded from the data set in order to reduce the computational cost of calculating the explanatory information.

[0084] Also, for example, the data processing unit 123 may add a copy of a learning task with a large positive contribution degree to the inference processing result or the performance evaluation value among the learning tasks included in the data set. That is, since such a learning task is considered to improve the inference performance, a copy of the same learning task is added to the data set.

[0085] Also, for example, the data processing unit 123 may exclude a learning task with a large negative contribution degree to the inference processing result or the performance evaluation value among the learning tasks included in the data set. That is, since such a learning task is considered to deteriorate the inference performance, it is excluded from the data set.

[0086] The second task processing unit 124 has, for example, as shown in FIG. 15, a learning processing unit 1241 and an inference processing unit 1242 that performs inference processing. In the second task processing unit 124 shown in FIG. 15, multi-task learning is performed.

[0087] The learning processing unit 1241 inputs the data set processed by the data processing unit 123, performs learning processing based on the data set, and outputs data indicating inference model parameters.

[0088] The inference processing unit 1242 inputs data indicating the output result by the learning processing unit 1241 and data indicating the inference task, performs inference processing based on the data indicating the output result and the inference data, and outputs data indicating the inference processing result for the inference task or the performance evaluation value of the inference processing result. That is, the function of the second task processing unit 124 is the same as the function of the task processing unit 121.

[0089] Also, in FIG. 15, the case where the task processing unit 121 and the second task processing unit 124 perform learning processing and inference processing is shown. However, it is not limited to this. Similar to FIGS. 3 to 5, the task processing unit 121 and the second task processing unit 124 may perform learning processing, additional adjustment processing, and inference processing, or may perform only learning processing, or may perform learning processing and additional adjustment processing. In this case, the second task processing unit 124 performs multi-task learning or meta-learning.

[0090] Note that in the above, the case where the task search unit 122, the data processing unit 123, and the second task processing unit 124 each perform processing once is shown. However, it is not limited to this. The output result by the second task processing unit 124 may be input again to the task search unit 122 to form a loop among the task search unit 122, the data processing unit 123, and the second task processing unit 124, and the task search unit 122, the data processing unit 123, and the second task processing unit 124 may be configured to perform processing a plurality of times respectively.

[0091] Also, the data analysis apparatus 12 according to Embodiment 2 may present data indicating the screening result in the middle of the learning task in the task search unit 122 to the user, and ask the user to determine whether to perform data processing and reprocessing.

[0092] Note that in the data analysis apparatus 12 according to Embodiment 2, the case where reprocessing is performed using the processed data set after the data in the data set is processed in the data processing unit 123 is shown. However, not limited to this, by outputting the improved data set, which is the output of the data processing unit 123, to the outside, it is also possible to configure it as a data cleansing device. That is, in this case, the second task processing unit 124 is unnecessary.

[0093] Regarding the utilization of the data analysis device 12 according to the second embodiment, for example, the following can be considered. For example, there may be cases where it is not possible to assume in advance which learning tasks will have a learning effect. Therefore, a redundant learning task can be included in the first data set input to the data analysis device 12, and the data set can be processed so as to narrow down the data having a learning effect in the data processing unit 123 according to the output result of the task search unit 122. Such a usage method can be considered. Also, for example, in the case of a classification problem, for a specific learning task included in the first data set, a new learning task with the number of classes or the number of data for each class in the learning task reduced can be created and added to the data set, and the data set after the addition may be input to the data analysis device 12. Also, for example, in addition to the learning task using raw sensor values, a new learning task consisting of data with filter processing added can be created and added to the first data set, and the data set after the addition may be input to the data analysis device 12.

[0094] Also, in FIG. 15, the case where the second task processing unit 124 is provided separately from the task processing unit 121 in the data analysis device 12 is shown. However, not limited to this, for example, as shown in FIG. 16, the second task processing unit 124 may not be provided, and the task processing unit 121 may be configured to include the functions of the second task processing unit 124. That is, in this case, the learning processing unit 1211 included in the task processing unit 121 inputs the data set processed by the data processing unit 123 and performs the processing again.

[0095] As described above, according to the second embodiment, the task search unit 122 outputs data indicating a learning task with characteristic contribution degree, and includes a data processing unit 123 that processes the data included in the data set based on the output result of the task search unit 122. Further, according to the second embodiment, the data analysis device 12 includes a second task processing unit 124 having a learning processing unit 1241 that inputs the data set processed by the data processing unit 123 and outputs data indicating inference model parameters or metadata. Further, according to the second embodiment, the learning processing unit 1211 included in the task processing unit 121 inputs the data set processed by the data processing unit 123 and performs the processing again. As a result, the data analysis device 12 according to the second embodiment can further improve the reliability compared to the data analysis device 12 according to the first embodiment. That is, in the data analysis device 12 according to the second embodiment, since data is handled in units of learning tasks, it is possible to improve the efficiency of the data reconstruction work. Also, even when it is not possible to assume which learning task has a learning effect, it is possible to perform the first task processing using a data set with a redundant configuration and then narrow it down to the necessary learning tasks.

[0096] Embodiment 3. In the data analysis device 12 according to the third embodiment, attribute information is attached to the learning task, and a case of extracting the attribute information attached to the learning task with characteristic contribution degree is shown.

[0097] FIG. 17 is a diagram showing a configuration example of the data analysis device 12 according to the third embodiment. In the data analysis device 12 according to the third embodiment shown in this FIG. 17, a storage unit 125, a storage processing unit 126, an attribute information extraction unit 127, and an additional information reception unit 128 are added to the data analysis device 12 according to the first embodiment shown in FIG. 2. Regarding other configuration examples in the data analysis device 12 according to the third embodiment shown in FIG. 17, they are the same as the configuration example of the data analysis device 12 according to the first embodiment, and the same reference numerals are given and the description thereof is made.

[0098] Note that, for each learning task, the data set input to the data analysis device 12 is attached with attribute information indicating the attributes of the learning task. Examples of the attribute information attached to the learning task include, for example, the description of the learning task, the acquisition date and time of the data possessed by the learning task, or the acquisition conditions of the data. In addition, the task search unit 122 outputs data indicating a learning task with characteristic contribution degree to the attribute information extraction unit 127.

[0099] The storage unit 125 stores the attribute information attached to the learning task together with the information indicating the learning task according to the processing by the storage processing unit 126. Examples of this storage unit 125 include non-volatile or volatile semiconductor memories such as RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable ROM), and EEPROM (Electrically EPROM), magnetic disks, flexible disks, optical disks, compact disks, mini disks, or DVDs (Digital Versatile Discs).

[0100] In addition, FIG. 17 shows the case where the storage unit 125 is provided inside the data analysis device 12. However, the present invention is not limited to this, and the storage unit 125 may be provided outside the data analysis device 12.

[0101] The storage processing unit 126 associates the information indicating the learning task included in the data set and the attribute information attached to the learning task based on the data set and stores them in the storage unit 125. In the storage unit 125 shown in FIG. 17, as attribute information for the learning task 1, information indicating the description of the learning task 1, "Plants in XX region. Photographed in 19XX", and the acquisition date and time of the data possessed by the learning task 1 is stored. Further, in the storage unit 125 shown in FIG. 17, as attribute information for the learning task 2, information indicating the acquisition date and time and acquisition conditions of the data possessed by the learning task 2, "Photographed in XX in 19XX. Equipment failure", is stored. Further, in the storage unit 125 shown in FIG. 17, as attribute information for the learning task M, information indicating the acquisition date and time and acquisition conditions of the data possessed by the learning task M, "Photographed in XX in 20XX. Monochrome", is stored.

[0102] Based on the learning task indicated by the data output by the task search unit 122, the attribute information extraction unit 127 extracts the attribute information associated with the learning task from the storage unit 125. The data indicating the attribute information extracted by this attribute information extraction unit 127 is output to the outside. Thereby, since the data analysis device 12 can output to the outside the attribute information attached to the learning task together with the data indicating the learning task output by the task search unit 122, it is possible to present more detailed information to the user.

[0103] The additional information reception unit 128 receives information indicating the learning task to be added, which is input by the user, and additional information regarding the attribute information attached to the learning task to be added. Examples of the additional information include comments left by the user when using the data analysis device 12. Then, based on the information received by the additional information reception unit 128, the storage processing unit 126 adds and stores additional information to the attribute information attached to the learning task to be added in the storage unit 125.

[0104] Note that FIG. 17 shows the case where the additional information reception unit 128 is provided in the data analysis device 12. However, the additional information reception unit 128 is not an essential configuration of the data analysis device 12, and the additional information reception unit 128 may not be provided in the data analysis device 12.

[0105] Also, in FIG. 17, a case is shown where a storage unit 125, a storage processing unit 126, an attribute information extraction unit 127, and an additional information reception unit 128 are added to the data analysis apparatus 12 according to Embodiment 1. However, not limited to this, a storage unit 125, a storage processing unit 126, an attribute information extraction unit 127, and an additional information reception unit 128 may be added to the data analysis apparatus 12 according to Embodiment 2, and the same effects as described above can be obtained.

[0106] As described above, according to this Embodiment 3, in the data set, attribute information indicating the attribute of the learning task is attached to each learning task, and the task search unit 122 outputs data indicating a learning task with a characteristic contribution degree. Based on the data set, a storage processing unit 126 that associates and stores in the storage unit 125 information indicating the learning tasks included in the data set and the attribute information attached to the learning tasks, and based on the learning task indicated by the data output by the task search unit 122, an attribute information extraction unit 127 that extracts the attribute information associated with the learning task from the storage unit 125. Thereby, the data analysis apparatus 12 according to Embodiment 3 can present more detailed information to the user.

[0107] Also, according to this Embodiment 3, the data analysis apparatus 12 includes an additional information reception unit 128 that receives additional information regarding information indicating a learning task to be added and the attribute information attached to the learning task, which are input by the user. The storage processing unit 126 adds and stores additional information to the attribute information attached to the learning task to be added in the storage unit 125 based on the information received by the additional information reception unit 128. Thereby, the data analysis apparatus 12 according to Embodiment 3 can add the information input by the user as attribute information.

[0108] Finally, with reference to FIG. 18, a hardware configuration example of the data analysis apparatus 12 according to Embodiments 1 to 3 will be described. Here, a hardware configuration example of the data analysis apparatus 12 according to Embodiment 1 will be shown, but the same applies to the hardware configuration examples of the data analysis apparatus 12 according to Embodiments 2 and 3. Each function of the task processing unit 121 and the task search unit 122 in the data analysis apparatus 12 is realized by the processing circuit 51. As shown in FIG. 18A, the processing circuit 51 may be dedicated hardware, or as shown in FIG. 18B, it may be a CPU (Central Processing Unit, also referred to as a central processing unit, processing unit, arithmetic unit, microprocessor, microcomputer, processor, or DSP (Digital Signal Processor)) 52 that executes a program stored in the memory 53.

[0109] When the processing circuit 51 is dedicated hardware, the processing circuit 51 corresponds to, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination thereof. Each function of each of the task processing unit 121 and the task search unit 122 may be realized by the processing circuit 51, or the functions of each part may be collectively realized by the processing circuit 51.

[0110] When the processing circuit 51 is a CPU 52, the functions of the task processing unit 121 and the task search unit 122 are realized by software, firmware, or a combination of software and firmware. The software and firmware are described as programs and stored in the memory 53. The processing circuit 51 realizes the functions of each part by reading and executing the programs stored in the memory 53. That is, the data analysis device 12 includes a memory for storing a program that, when executed by the processing circuit 51, causes each step shown in FIG. 4 to be executed as a result. Also, these programs can be said to cause a computer to execute the procedures and methods of the task processing unit 121 and the task search unit 122. Here, examples of the memory 53 include non-volatile or volatile semiconductor memories such as RAM, ROM, flash memory, EPROM, and EEPROM, magnetic disks, flexible disks, optical disks, compact disks, mini disks, or DVDs.

[0111] Note that, regarding the functions of the task processing unit 121 and the task search unit 122, part of them may be realized by dedicated hardware and part by software or firmware. For example, for the task processing unit 121, its function can be realized by the processing circuit 51 as dedicated hardware, and for the task search unit 122, its function can be realized by the processing circuit 51 reading and executing the program stored in the memory 53.

[0112] As described above, the processing circuit 51 can realize the above-described functions by hardware, software, firmware, or a combination thereof.

[0113] Note that free combinations of the embodiments, or modifications of any components of the embodiments, or omissions of any components in the embodiments are possible.

Industrial Applicability

[0114] The data analysis device 12 according to the present disclosure can analyze data in task units and is suitable for use in a data analysis device 12 that analyzes data and the like.

Description of Reference Numerals

[0115] 1 Information processing system, 11 Data set acquisition unit, 12 Data analysis device, 13 Information processing unit, 51 Processing circuit, 52 CPU, 53 Memory, 121 Task processing unit, 122 Task search unit, 123 Data processing unit, 124 Second task processing unit, 125 Storage unit, 126 Storage processing unit, 127 Attribute information extraction unit, 128 Additional information reception unit, 1211 Learning processing unit, 1212 Inference processing unit, 1213 Adjustment processing unit, 1241 Learning processing unit, 1242 Inference processing unit.

Claims

1. A task processing unit having a learning processing unit that inputs a dataset having a plurality of learning tasks each having a plurality of data and outputs data indicating inference model parameters or metadata, and A task search unit that inputs the dataset and data indicating the output result by the task processing unit, and outputs data indicating the contribution degree of the learning task included in the dataset to the output result, or data indicating a learning task characteristic in the contribution degree A data analysis apparatus comprising:

2. The learning processing unit outputs data indicating inference model parameters, The task processing unit, Inputs data indicating the output result by the learning processing unit and data indicating an inference task, and has an inference processing unit that outputs data indicating an inference processing result for the inference task or a performance evaluation value of the inference processing result The data analysis apparatus according to claim 1, characterized in that

3. The task processing unit, Inputs data indicating the output result by the learning processing unit and data indicating an inference task, and has an adjustment processing unit that outputs data indicating inference model parameters The data analysis apparatus according to claim 1, characterized in that

4. The task processing unit, Inputs data indicating the output result by the learning processing unit and data indicating an inference task, and has an adjustment processing unit that outputs data indicating inference model parameters, and Inputs data indicating the output result by the adjustment processing unit and data indicating an inference task, and has an inference processing unit that outputs data indicating an inference processing result for the inference task or a performance evaluation value of the inference processing result The data analysis apparatus according to claim 1, characterized in that

5. The task search unit calculates a first contribution degree that is the contribution degree of the learning task included in the dataset to the output result by the learning processing unit, and a second contribution degree that is the contribution degree of the output result by the learning processing unit to the output result by the inference processing unit, and calculates the contribution degree of the learning task to the output result by the inference processing unit by synthesizing the first contribution degree and the second contribution degree The data analysis apparatus according to claim 2, characterized in that

6. The task search unit calculates a first contribution degree, which is the contribution degree of the learning task included in the dataset with respect to the output result by the learning processing unit, and a third contribution degree, which is the contribution degree of the output result by the learning processing unit with respect to the output result by the adjustment processing unit, and calculates the contribution degree of the learning task with respect to the output result by the adjustment processing unit by synthesizing the first contribution degree and the third contribution degree. The data analysis device according to claim 3, characterized in that.

7. The task search unit calculates a first contribution degree, which is the contribution degree of the learning task included in the dataset with respect to the output result by the learning processing unit, a third contribution degree, which is the contribution degree of the output result by the learning processing unit with respect to the output result by the adjustment processing unit, and a fourth contribution degree, which is the contribution degree of the output result by the adjustment processing unit with respect to the output result by the inference processing unit, and calculates the contribution degree of the learning task with respect to the output result by the inference processing unit by synthesizing the first contribution degree, the third contribution degree, and the fourth contribution degree. The data analysis device according to claim 4, characterized in that.

8. The task processing unit determines inference model parameters based on the loss function for the entire dataset, which is obtained by synthesizing the loss functions for each learning task included in the dataset for each learning task. The task search unit calculates a differential coefficient with respect to the perturbation parameter based on a loss function obtained by deforming the loss function of any one learning task with a perturbation parameter common to all the data within the learning task, with respect to the loss function for the entire dataset, and uses the differential coefficient as the contribution degree of the learning task. The data analysis device according to any one of claims 1 to 4, characterized in that.

9. The task processing unit determines inference model parameters based on the loss function for the entire dataset, which is obtained by synthesizing the loss functions for each learning task included in the dataset for each learning task. The task search unit calculates a differential coefficient with respect to the perturbation parameter based on a loss function obtained by deforming the loss function of any one learning task with a perturbation parameter common to all the data within the learning task, with respect to the loss function for the entire dataset, and uses the differential coefficient as the contribution degree of the learning task. The data analysis device according to any one of claims 5 to 7, characterized in that...

10. For each process performed by the task processing unit, the task search unit calculates a differential coefficient as the contribution degree between the input and output in the process to form a matrix, and calculates the matrix product of the matrices for the plurality of processes to synthesize the contribution degrees. The data analysis device according to claim 9, characterized in that...

11. The task processing unit determines model parameters that are values realizing the extreme value of the loss function for the entire data set or approximate values of such values. The task search unit calculates the contribution degree of the learning task with respect to the inference model parameters determined by the task processing unit by implicit function differentiation. The data analysis device according to claim 8, characterized in that...

12. The task processing unit determines model parameters that are values realizing the extreme value of the loss function for the entire data set or approximate values of such values. The task search unit calculates the contribution degree of the learning task with respect to the inference model parameters determined by the task processing unit by implicit function differentiation. The data analysis device according to claim 9, characterized in that...

13. The task search unit outputs data indicating a learning task with characteristic contribution degree. Based on the output result by the task search unit, it is provided with a data processing unit that processes the data included in the data set. The data analysis device according to any one of claims 1 to 7, characterized in that...

14. It is provided with a second task processing unit having a learning processing unit that inputs the data set after being processed by the data processing unit and outputs data indicating inference model parameters or metadata. The data analysis device according to claim 13, characterized in that...

15. The learning processing unit included in the task processing unit inputs the data set after being processed by the data processing unit and performs the process again. The data analysis device according to claim 13, characterized in that...

16. In the data set, for each learning task, attribute information indicating the attribute of the learning task is attached. The task search unit outputs data indicating a learning task with characteristic contribution degree. Based on the data set, a storage processing unit that associates and stores in a storage unit information indicating the learning tasks included in the data set and the attribute information attached to the learning tasks. An attribute information extraction unit that extracts, from the storage unit, attribute information associated with the learning task based on the learning task indicated by the data output by the task search unit The data analysis device according to any one of claims 1 to 7, characterized in that

17. An additional information reception unit that receives information indicating a learning task to be added, which is input by a user, and additional information for the attribute information attached to the learning task Based on the information received by the additional information reception unit, the storage processing unit adds and stores additional information to the attribute information attached to the learning task to be added in the storage unit The data analysis device according to claim 16, characterized in that

18. A data set acquisition unit that acquires a data set having a plurality of learning tasks each having a plurality of data, The data analysis device according to any one of claims 1 to 7, An information processing unit that performs information processing in units of tasks based on the output result by the data analysis device, and The data analysis device inputs the data set acquired by the data set acquisition unit An information processing system, characterized in that

19. A step in which a task processing unit inputs a data set having a plurality of learning tasks each having a plurality of data and outputs data indicating inference model parameters or metadata, A step in which a task search unit inputs the data set and the data indicating the output result by the task processing unit, and outputs data indicating the contribution degree of the learning task included in the data set to the output result, or data indicating the learning task having a characteristic contribution degree A data analysis method having

Images