JP2019091177A - Learning support system, learning support method and learning support processing program - Google Patents

Abstract

To facilitate learning relating to data analysis.SOLUTION: A learning support system in an embodiment includes: a data analysis unit which outputs an analysis result for input data; a simulated data generation unit which acquires source data of the input data to generate simulated data which has features different from features of the source data and simulates the input data; a data selection unit which selects data for verification of analysis performance of the data analysis unit and data for learning of an algorithm of analysis processing in the data analysis unit, from the simulated data; a learning unit which learns the algorithm on the basis of the selected data for learning; and a verification unit which acquires an analysis result with the data for verification as input data by using the learned algorithm and verifies the analysis performance on the basis of this analysis result.SELECTED DRAWING: Figure 1

Description

  Embodiments of the present invention relate to a learning support system, a learning support method, and a learning support processing program.

There is known a system equipped with artificial intelligence (AI) that analyzes input data by an algorithm to obtain output data.
As an example of services that can be provided by this system, there is an interactive question-and-answer service by artificial intelligence. This service is provided by an expert system (a program that behaves like an expert in the field by inputting data about each specialized field), which is an analysis result for a question prepared in advance by the developer. , Is a service to acquire new knowledge and information from the dialogue with the user.

JP 2005-196752 A JP 2002-157571 A JP 2001-75808 A

In the practical application of the existing artificial intelligence, it is necessary to design the flow of processing for using the artificial intelligence for each task to be solved by the artificial intelligence. For this reason, a large amount of input data is required for learning artificial intelligence, which leads to an increase in cost until artificial intelligence is introduced to a real machine. In addition, when trying to apply a new algorithm to artificial intelligence, the existing algorithm can not be diverted, and in many cases it is necessary to redesign the whole, so the extensibility is poor.
As described above, in the existing technology, it was not possible to smoothly perform learning related to data analysis.

  The problem to be solved by the present invention is to provide a learning support system, a learning support method, and a learning support processing program capable of smoothly performing learning related to data analysis.

  The learning support system according to the embodiment acquires a data analysis unit that outputs an analysis result on input data, and acquires original data of the input data, and simulates the input data with characteristics different from the characteristics of the original data. A simulated data generation unit that generates data; and a data selection unit that selects, from the simulation data, data for verification of analysis performance of the data analysis unit, and learning data for an algorithm of analysis processing by the data analysis unit. A learning unit for learning the algorithm based on the selected learning data; and the analysis using the learned algorithm for the input data when the verification data is the input data. And a verification unit that acquires a result and verifies the analysis performance based on the analysis result.

BRIEF DESCRIPTION OF THE DRAWINGS The block diagram which shows the function structural example of the learning assistance system in embodiment. The flowchart which shows an example of the operation | movement procedure by the learning assistance system in embodiment. The figure which shows an example of the form of the simulation data production | generation by the simulation data production | generation part of the learning assistance system in embodiment. The figure explaining an example of the data which the simulation data generation part of the learning support system in an embodiment acquires. The figure explaining an example of the data which the simulation data generation part of the learning support system in an embodiment acquires. The figure explaining an example of the data which the simulation data generation part of the learning support system in an embodiment acquires. The figure explaining an example of the data which the simulation data generation part of the learning support system in an embodiment acquires. The figure which shows an example of the simulation data produced | generated by the simulation data production | generation part of the learning assistance system in embodiment. The figure which shows an example of the form of the data selection by the data selection part of the learning assistance system in embodiment. The figure which shows an example of a production | generation form of the rare case data contained in the data set for verification stored in the data storage part for verification of the learning assistance system in embodiment. The figure which shows an example of learning of artificial intelligence by the algorithm learning part of the learning assistance system in embodiment. The figure which shows an example of drafting of the command by the artificial intelligence part of the learning assistance system in embodiment. The figure explaining the verification by the artificial intelligence verification part of the learning assistance system in an embodiment.

Hereinafter, embodiments will be described using the drawings.
FIG. 1 is a block diagram showing an example of a functional configuration of a learning support system in the embodiment.
In this embodiment, an example is described in which an artificial intelligence learning support system is applied to support learning of an artificial intelligence algorithm that makes a decision in a simulation game (for example, determination of an instruction related to the action of the own team). This learning support system can also be applied to support learning of an artificial intelligence algorithm that performs other processing (eg, character recognition, image recognition, kana-kanji conversion, translation, etc.).

  As shown in FIG. 1, the learning support system in the embodiment includes a data creation unit 11, a data simulation unit 12, a data management storage unit 20, an artificial intelligence learning unit 30, and an artificial intelligence verification unit 40.

  The above data creation unit 11, data simulation unit 12, data management storage unit 20, artificial intelligence learning unit 30, and artificial intelligence verification unit 40 have, for example, hardware configurations (eg, input / output device, CPU, memory, storage device, etc.) The computer can be implemented by a combination of a computer, such as a personal computer and software, which can be a program that is pre-installed on a computer from a network or storage medium and that can be executed by the computer.

  The data creation unit 11 receives an input operation from the system administrator for data serving as a source of learning of artificial intelligence with respect to an input device (not shown), and creates new data to be sent to the data simulation unit 12.

The data simulation unit 12 includes a simulation data generation unit 13 and a data selection unit 14.
The simulation data generation unit 13 is operated on a trial basis with a new data generated by the data generation unit 11 and a real machine on which artificial intelligence is installed, for example, a computer for devising an instruction in a simulation game, that is, Before formal operation, field data obtained on site when a real machine is temporarily operated for learning artificial intelligence is respectively input.
The field data should include actual data (data input in the past) of input data for artificial intelligence and actual data (for example, an instruction to one's own team in the past) which is an analysis result for this input data. Can.

  The simulated data generation unit 13 artificially simulates the new data described above, a feature in which at least a part of the feature included in the field data is changed, and the original new data and replication (proliferation) data based on the field data. Input data to intelligence is generated as data to simulate, and a data set including this simulation data, the above new data, and field data is generated.

  As described above, data including features obtained by changing at least a part of features included in the new data can be used as simulation data. In addition, data including features obtained by changing at least a part of the features included in the actual data of the input data in the field data can be used as simulation data.

The new data and field data in the above simulation game have various conditions as data characteristics, for example, terrain data (including information on roads) in the simulation game, weather data, land use method data (for example, movement of other team vehicles) Route, time data, information of other teams, information of one's team can be included. In addition, field data in the simulation game can include, as a feature of the data, an instruction to the own team, for example, an instruction of an action to be taken next by the own team based on information of other teams and information of the own team.
When the new data and the field data are data in the character recognition and the image recognition, the new data and the field data may not particularly include the various conditions.

  From the data set generated by the simulation data generation unit 13, the data selection unit 14 uses a fixed number of verification data used to verify the analysis performance of artificial intelligence in a fixed number or a data set generated by the simulation data generation unit 13. Randomly and sequentially sort by ratio, and sort other data sequentially as a training data set used for learning artificial intelligence algorithm. The training data set may be simply referred to as training data.

The data management storage unit 20 includes a learning data storage unit 21, an interface conversion unit 22, and a verification data storage unit 23.
The learning data storage unit 21 includes a storage device such as a non-volatile memory, and stores the learning data sequentially sorted by the data sorting unit 14 in the storage device.
The interface conversion unit 22 converts the format of the learning data stored in the storage device of the learning data storage unit 21 into a format suitable for processing by the artificial intelligence learning unit 30.
The verification data storage unit 23 includes a storage device such as a non-volatile memory, and stores a verification data set including verification data sequentially sorted by the data sorting unit 14 in the storage device.

The artificial intelligence learning unit 30 includes an algorithm learning unit 31 and an artificial intelligence unit (data analysis unit) 32.
The algorithm learning unit 31 performs learning processing of an algorithm that configures artificial intelligence based on learning data from the data management storage unit 20, that is, an analysis processing algorithm that analyzes input data and obtains draft data (for example, Perform supervised learning.

  An example of learning is machine learning by deep learning. Deep learning is a method of learning parameters between layers in a multi-layered neural network. As a neural network, CNN (Convolutional Neural Network), RNN (Recurrent Neural Network), etc. are used. The algorithm learning unit 31 may perform machine learning by another method such as SVM (Support Vector Machine), logistics regression, Bayesian filter, and the like.

  The artificial intelligence unit 32 is a main part of artificial intelligence. The artificial intelligence unit 32 obtains draft data which is an analysis result for the input data according to the algorithm subjected to the learning process by the algorithm learning unit 31, and outputs the draft data together with the input data.

  The artificial intelligence verification unit 40 has an artificial intelligence verification processing unit 41. The artificial intelligence verification processing unit 41 uses the data set for verification stored in the storage device of the verification data storage unit 23 as input data for artificial intelligence, and data proposed by the algorithm of the artificial intelligence unit 32 (analysis result The artificial intelligence analysis performance is verified by verifying whether or not it is appropriate data (appropriate analysis result for input data).

Next, an operation procedure by the learning support system in the embodiment will be described.
FIG. 2 is a flowchart showing an example of an operation procedure by the learning support system in the embodiment. Here, it is assumed that the artificial intelligence unit 32 is mounted on a real machine and operated on a trial basis.

  First, the data creation unit 11 receives an input operation of new data to be a source of learning by the learning support system, creates new data to be sent to the data simulation unit 12, and sends the created data to the data simulation unit 12 S11).

  The simulated data generation unit 13 of the data simulation unit 12 receives the new data created by the data creation unit 11 and the field data obtained at the operation site by the actual machine on which the artificial intelligence unit 32 is installed, and these data The simulation data is generated based on the above, and a data set including the simulation data is generated as a data set to the data selection unit 14 (S12).

FIG. 3 is a diagram showing an example of a form of simulated data generation by the simulated data generation unit of the learning support system in the embodiment.
In the example shown in FIG. 3, the simulation data generation unit 13 selects at least a part of the data obtained by combining the new data and the field data as data of the template, and data including conditions different from the conditions included in the data of the template A set (condition creation data) is generated to be a data set including simulated data. This template data is the original data of the data to be analyzed by the artificial intelligence unit 32.

  For example, when the data of the template has the condition a, the condition b, and the condition c, based on the data of the template, (1) data of the template itself having the condition a, the condition b, and the condition c, (2) the condition a ′ (the condition a has been changed), condition b, data having the condition c, (3) condition a, condition b ′ (the condition b has been changed), data having the condition c, (4) the condition a, Condition b, condition c '(a change of condition c), (5) data having condition a', condition b ', condition c, (6) data having condition a', condition b, condition c ', 7) A data set including a plurality of simulated data such as data having condition a, condition b 'and condition c', (8) data having condition a ', condition b' and condition c ' The generation of a data set of 14 is taken as an example.

  Although FIG. 3 shows an example in which a data set including a plurality of simulation data is generated based on one template data, the present invention is not limited thereto, and a plurality of template data is selected and conditions included in these data A data set may be generated based on a combination of

  For example, if the first data in the form of the wedge has conditions a, b and c and the data in the second form has conditions d, e and f, then (1) condition a, condition b, first wedge data itself having condition c, (2) data having condition a ′, condition b, condition c based on first wedge data, (3) based on first wedge data , Condition a, condition b ', data having condition c, (4) data having condition a, condition b, condition c' based on data of the first template, (5) condition d, condition e, condition f Data of the second template having itself, (6) based on the data of the second template, the condition d ′ (the variation of the condition d), data having the condition e, and the condition f, (7) the second template The condition d, the condition e ′ (the condition e is changed), the data having the condition f, and (8) the second A data set including a plurality of simulated data, such as data having condition d, condition e, and condition f '(a change of condition f) based on the data of the form is generated as a data set to the data selection unit 14 Being taken as an example.

  In addition, when it is possible to generate data combining simulation data of the first template and data of the second template as simulation data by the simulation data generation unit 13, for example, at least a part of the conditions a to f described above Data combining the first template data and the second template data having changed conditions such as condition a ′, condition b, condition c, condition d, condition e, and condition f is generated as simulation data It is also good.

Next, an example of new data and field data that are the basis of simulated data will be described.
FIG.4, FIG.5, FIG.6 and FIG. 7 are figures explaining an example of the data which the simulation data production | generation part of the learning assistance system in embodiment acquires.
In the above simulation game, the new data and the field data may include terrain data, weather data, land usage method data, time data, information of other teams, information of one's own team, and an action instruction to another team. This instruction is an instruction as teacher data.

In the example shown in FIG. 4, the information of the other team is the type of the other team (for example, vehicle, walking member), the position (coordinates) of the other team, the state of the other team (for example, moving or waiting), the other team Contains information to which the action policy (for example, waiting, active) is associated.
Also, in the example shown in FIG. 5, the information of the own team is the name of the member of the own team (for example, the own team A), the type of tool of the own team (for example, tool A), the position of the own team, The information (for example, completion of preparation for action) is included.

  In FIG. 6, the command to the other team in the new data created by the data creation unit 11 is shown. This instruction is issued to the team based on new topography data, new weather data, new land use data, new time data, new information of other teams, and new information of the own team. . This command is issued to own team A to advance (move) to a predetermined first goal, and issued to own team B as a member of own team to a predetermined second goal. including.

  FIG. 7 shows the result of the instruction to the own team in the field data. This command is a track record of commands based on past topography data, past weather data, past land usage method data, past time data, past information of other teams, and past information of one's own team. A command to move back to a predetermined first place issued to A, and a command to move back to a predetermined second place issued to own team B are included.

Next, an example of simulated data of a plurality of patterns in which the conditions in the above-mentioned template data are changed will be described.
FIG. 8 is a diagram showing an example of simulation data generated by a simulation data generation unit of the learning support system in the embodiment.
In the example shown in FIG. 8, the terrain data and land use data included in the data of the template which is the source of generation of the simulation data by the simulation data generation unit 13 are members of the own team (your own team A, own team B and own team C) are arranged in a predetermined area, indicating that another team is moving on a road near this area.
Also, the condition A included in the data of this template is the weather, the time zone, and the action policy of the other team, and the weather of this condition A is "fine weather" and the time zone is "3: 00-4: 00 (24-hour notation, the same applies below), and the action policies of other teams are "standby". In the present embodiment, three simulated data are generated based on this template data.

The weather of condition A1 included in the first simulated data is "rainy weather", the time zone is "5:00 to 6:00", and the action policy of the other team is "standby". Further, in this simulated data, the moving speed of the other team vehicle is reduced because the weather is rainy, which is different from the data of the template.
That is, this simulated data is data in which the weather, the time zone, and the moving speeds of the other team vehicles are different from each other for the data of the template.

The weather of condition A2 given to the second simulated data is "fine weather", the time zone is "6:00 to 12:00", and the action policy of the other team is "standby". Also, in this simulated data, the route of another team vehicle is a route that is difficult for the team to find because the surroundings are bright, and this is different from the data of the template.
That is, this simulated data is data in which the time zone and the route of another team vehicle are made different from each other for the template data.

The weather of condition A3 given to the third simulated data is "rainy weather", the time zone is "18:00 to 24:00", and the action policy of the other team is "active". Further, in this simulated data, the route of the other team vehicle is a route which goes straight to the own team ignoring the road, which is different from the data of the template, as the action policy of the other team is "active".
That is, this simulated data is data in which the time zone, the action policy of another team, and the route of another team vehicle are different from each other for the template data.

  Next, the data sorting unit 14 randomly sorts out the verification data set used for verification of the artificial intelligence from the data set output from the simulation data generation unit 13 at a fixed number or a fixed ratio in the data set. The data is set as a learning data set used for artificial intelligence learning (S13).

The verification data set includes the verification data selected by the data selection unit 14 and data separately set for the verification data.
FIG. 9 is a diagram illustrating an example of a form of data sorting by the data sorting unit of the learning support system in the embodiment.
As shown in FIG. 9, in the present embodiment, the data set separately as described above is called rare case data (singularity data).
The rare case data is a feature different from the features of the above new data, the features of the field data, and the features of the simulated data, and should be data having rare features as compared to these features. Can.

The rare case data may be any data created by the data creation unit 11 according to the input operation by the administrator.
In addition, the data creation unit 11 acquires a data set that is the basis of selection by the data selection unit 14, and the rare feature is compared with these features that are different from the features possessed by each data in this data set. Data formed by the data creation unit 11 automatically including data may be used as rare case data.

  FIG. 10 is a diagram showing an example of a form of creation of rare case data included in the verification data set accumulated in the verification data accumulation unit of the learning support system in the embodiment. Here, although the example which produces rare case data based on the data of a template is demonstrated, not only this but rare case data may be created from the state without data of a template.

  In the example shown in FIG. 10, the data of the template which is the source of the rare case data is the same as the data of the template shown in FIG. The condition A included in the data of this template is the weather, time zone, and the action policy of other teams, and the weather of this condition A is "fine weather" and the time zone is "3:00 to 4:00", and other teams The action policy of is "waiting", and three rare case data are generated based on the data of this template.

The weather of condition B1 included in the first rare case data is "typhoon", the time zone is "3:00 to 4:00", and the action policy of the other team is "standby". Also, in this rare case data, the road condition is bad because the weather is a typhoon. For this reason, it is not possible to pass the vehicle on this road, and the other team vehicles are traveling off the road, which is different from the data of the template.
That is, this rare case data is data in which the weather is changed to a weather whose frequency of occurrence is rare with respect to a typhoon, that is, fine weather or rainy weather with respect to the data of the template, and the routes of other team vehicles are different.

The weather of condition B2 included in the second rare case data is "fine weather", the time zone is "6:00 to 11:00", and the action policy of the other team is "unknown". Here, it is assumed that "unknown" is a behavior policy whose frequency of occurrence is rare with respect to "waiting" or "active". Further, in this rare case data, the route of the other team vehicle is unstable and travels on the road or off road because the action policy of the other team is "unknown", which is different from the data of the template.
That is, the rare case data is data in which the time zone is different from the data of the template, the action policy of the other team is changed to "unknown", and the route of the other team vehicle is different.

The weather of condition B3 included in the third rare case data is "fine weather", the time zone is "18:00 to 24:00", and the action policies of the other teams are "negative". Here, it is assumed that "negative" is a behavior policy whose frequency of occurrence is rare with respect to "waiting" or "active". Further, in this rare case data, other teams' vehicles are hardly moved from the predetermined position, which is different from the data of the template, as the action policy of the other team is "negative".
That is, this rare case data is data in which the time zone is different from the template data, and the action policies of other teams are changed to "negative", and the routes of other team vehicles are different.

  The learning data set selected by the data selection unit 14 is stored in the learning data storage unit 21 of the data management storage unit 20. In addition, a verification data set, that is, data obtained by combining the verification data selected by the data selection unit 14 and the rare case data is stored in the verification data storage unit 23 of the data management storage unit 20 (see FIG. S14).

  The learning data accumulated in the learning data accumulation unit 21 is converted by the interface conversion unit 22 and sequentially input to the algorithm learning unit 31 of the artificial intelligence learning unit 30 (S15). The timing of this input is, for example, the timing at which the learning data stored in the learning data storage unit 21 is updated.

FIG. 11 is a diagram illustrating an example of artificial intelligence learning by the algorithm learning unit of the learning support system according to the embodiment.
As shown in FIG. 11, the learning data stored in the learning data storage unit 21 is a simulated data generation unit based on the new data created by the data creation unit 11 when the new data is used as a template data. A simulated data group consisting of a plurality of simulated data generated by 13 can be included.

  The data created by the data creating unit 11 described here is, for example, new terrain data, new weather data, new land use method data, new time data, new team data, as shown in FIG. It is data that includes information, new information from other teams, and instructions to your team.

Further, as shown in FIG. 11, when the field data is used as template data, the above-mentioned stored data for learning is composed of a plurality of simulation data generated by the simulation data generation unit 13 based on this data. It can contain simulated data groups.
The field data described here is, for example, the past landform data, the past weather data, the past land use method data, the past time data, the past information of the own team, and the past information of the other teams shown in FIG. , And data including past orders for the team.

  The algorithm learning unit 31 analyzes the simulated data group, and updates the algorithm of the artificial intelligence unit 32 with a more suitable algorithm, thereby performing learning of the artificial intelligence algorithm (S16).

FIG. 12 is a diagram illustrating an example of the generation of an instruction by the artificial intelligence unit of the learning support system in the embodiment.
The artificial intelligence unit 32 compares the new input data with the learning data stored in the learning data storage unit 21 using the algorithm learned by the algorithm learning unit 31 to obtain the new input. It is possible to obtain draft data which is an analysis result for data, for example, a new command for one's own team.

  The new input data to be described here is data different from the new data that is the basis of the above-mentioned simulation data generation, and for example, new topography data, new weather data, new land use method data, new data It is data including time data, new information of one's team, and new information of other teams.

Next, verification by the artificial intelligence verification unit 40 will be described. Here, an example will be described in which artificial intelligence is verified using data in a verification data set as new input data for verification by the artificial intelligence verification unit 40. FIG. 13 is a diagram for explaining verification by the artificial intelligence verification unit of the learning support system in the embodiment.
In the example shown in FIG. 13 (a), while the terrain data in the new input data for verification by the artificial intelligence verification unit contains other team vehicles in the radar range W1 by its own team A, its own team B Indicates that the other team vehicle is not included in the radar range W2 according to. The radar range W2 partially overlaps with the radar range W1 and is narrower than the radar range W1.
In this case, desirable drafting data (instructions) as an analysis result for the verification data set is, for example, monitoring of the team A from the other team vehicle to a larger number of times.

In the example shown in FIG. 13B, the terrain data in the new input data for verification includes another team vehicle in the radar range W1 of the own team A and the radar range W2 of the own team B, and the own team It shows that you can monitor other team vehicles from A and your own team B.
In this case, the desirable drafting data is, for example, an instruction to alternately perform monitoring by the own team A and monitoring by the own team B in order to effectively grasp the other team vehicles.

In the example illustrated in FIG. 13C, the terrain data in the new input data for verification is temporarily entered by another team vehicle in the radar range W1 of the own team A and the radar range W1 of the own team B. This other team vehicle moving speed is high, and this other team vehicle is approaching the own team A, own team B, and own team C at high speed in a state where it does not immediately move out of the radar range W1 and enter the radar range W2. Indicates that.
In this case, the desirable drafting data is, for example, a reverse command to each member (your team A, your team B, your team C) that compose your team.

  The artificial intelligence verification processing unit 41 of the artificial intelligence verification unit 40 uses the algorithm of the artificial intelligence unit 32, and the data in the verification data set accumulated in the verification data accumulation unit 23, for example, FIG. (B) or data shown in FIG. 13 (c) is input, and drafting data that is an analysis result for the input data is obtained using the above algorithm.

  Then, the artificial intelligence verification processing unit 41 compares this drafting data with the predetermined reference data corresponding to the drafting data, that is, the above-mentioned desirable drafting data, and compares the drafting data with the standard data. It is determined whether the similarity is equal to or greater than a predetermined value, or whether the draft data and the reference data are identical. This reference data is desirable data as an analysis result for the verification data set.

Here, an example of creation of reference data will be described.
Based on the input operation by the administrator, reference data as a desirable analysis result for the rare case data in the verification data set stored as described above in the verification data storage unit 23 is determined by the data creation unit 11 using the rare case data. It can be created along with creation of the data itself, and can be stored in the verification data storage unit 23 in association with the rare case data.

  The reference data to be compared with the verification data set stored in the verification data storage unit 23 after being sorted by the data selection unit 14 in accordance with the input operation by the administrator is a separate data creation unit or data It can be created by the creation unit 11 and stored in the verification data storage unit 23 in association with the verification data set.

  Based on the above determination, the artificial intelligence verification processing unit 41 determines whether the analysis performance of the artificial intelligence unit 32 is a performance suitable for formal operation when the artificial intelligence unit 32 is mounted on a real machine, that is, satisfies the predetermined condition It is verified whether draft data can be obtained by the height accuracy (S17).

  With the above verification, it is possible to verify that the analysis result for the data selected by the data selection unit 14 among the verification data that is the source of the verification is a desirable analysis result, and for the verification that is the source of the verification Among the data, it can be verified that the analysis result for the data created as rare case data is the desired analysis result.

  Also, as another example of the processing of the artificial intelligence verification processing unit 41, the artificial intelligence verification processing unit 41 does not use the algorithm of the artificial intelligence unit 32, and the artificial intelligence unit 32 is an analysis result for the verification data set Outgoing data may be obtained, and the artificial intelligence verification processing unit 41 may perform the above verification by acquiring the draft data and comparing it with predetermined reference data.

  As a result of verification by the artificial intelligence verification processing unit 41, when the analysis performance of the artificial intelligence unit 32 is not the performance suitable for the above-mentioned formal operation (No in S18), the artificial intelligence verification unit 40 generates the simulated data generation unit 13 Is instructed to create the additional new data and the simulation data based on the field data again, and the process after S11 is performed again.

  When the above processing from S11 to S17 is performed, the learning data accumulated in the learning data accumulation unit 21 and the data accumulated in the verification data accumulation unit 23 are held as they are.

  Then, when additional new data is generated by the data generation unit 11 and new simulation data is generated by the simulation data generation unit 13, new learning data selected by the data selection unit 14 from the simulation data is generated. The additional verification data is accumulated in the learning data accumulation unit 21, and the new verification data selected by the data selection unit 14 is accumulated in the verification data accumulation unit 23 as an additional verification data set.

  That is, since data generated by the data generation unit 11 and field data input to the simulation data generation unit 13 are respectively increased as compared with the processing in the previous steps S11 to S17, the learning data accumulation is performed. The amount of learning data accumulated in the unit 21 and data accumulated in the verification data accumulation unit 23 increases.

  When the additional verification data set is stored in the verification data storage unit 23, the existing data stored in the verification data storage unit 23 is stored so that the number of the verification data stored is less than a predetermined number. A portion of the verification data set may be overwritten and updated by the additional verification data set.

  As a result, the learning accuracy of the artificial intelligence algorithm is improved, and the draft data compared by the artificial intelligence verification processing unit 41 matches the reference data, or approaches the reference data, and the analysis performance of the artificial intelligence unit 32 becomes a real machine. Approaches the performance suitable for the formal operation in and will reach this performance.

  As a result of verification by the artificial intelligence verification processing unit 41, when the analysis performance of the artificial intelligence unit 32 is the performance suitable for the above-mentioned formal operation (Yes in S18), the processing by the learning support system is ended.

  As described above, the learning support system in the embodiment generates simulation data based on the new data and the field data, selects the learning data and the verification data from the data set including the simulation data, and performs learning. After learning in the algorithm of the artificial intelligence unit using the data for use, verification of the artificial intelligence unit is performed based on the draft data when the verification data is input data of the artificial intelligence unit.

Therefore, learning of the artificial intelligence unit can be performed using a large amount of data as compared to the number of new data and field data, and the artificial intelligence unit using part of data of the data set including simulated data Can be verified.
That is, the burden of preparing data necessary for learning and the burden of verifying the artificial intelligence unit can be reduced, so that learning concerning analysis of data can be smoothly performed.

In addition, the method by the learning support system described in the above embodiment is a program that can be executed by a computer, such as a magnetic disk (floppy (registered trademark) disk, hard disk, etc.), optical disk (CD-ROM, DVD, etc.), It can also be stored and distributed in a storage medium such as a magneto-optical disk (MO) or a semiconductor memory.
In addition, as the storage medium, any storage format may be used as long as it can store a program and can be read by a computer.

In addition, an operating system (OS) operating on a computer based on instructions of a program installed in the computer from a storage medium, MW (middleware) such as database management software, network software, etc. realize the above embodiment. Part of each process may be executed.
Furthermore, the storage medium in the embodiment is not limited to a medium independent of the computer, but includes a storage medium downloaded and stored or temporarily stored a program transmitted by a LAN, the Internet, or the like.

Further, the storage medium is not limited to one, and the processing in the above embodiment may be executed from a plurality of media and included in the storage medium in the present invention, and the medium configuration may be any configuration. Note that the computer in the embodiment executes each process in the above embodiment based on the program stored in the storage medium, and an apparatus consisting of one such as a personal computer and a plurality of apparatuses are connected to a network. It may be any configuration such as a system.
Further, the computer in the embodiment is not limited to a personal computer, but includes an arithmetic processing unit, a microcomputer and the like included in an information processing device, and collectively refers to devices and devices capable of realizing the functions of the present invention by a program. .

  While certain embodiments of the present invention have been described, these embodiments have been presented by way of example only, and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the scope and the gist of the invention, and are included in the invention described in the claims and the equivalent scope thereof.

  11: data creation unit, 12: data simulation unit, 13: simulation data generation unit, 14: data selection unit, 20: data management storage unit, 21: learning data storage unit, 22: interface conversion unit, 23: for verification Data storage unit, 30: artificial intelligence learning unit, 31: algorithm learning unit, 32: artificial intelligence unit, 40: artificial intelligence verification unit, 41: artificial intelligence verification processing unit.

Claims (5)

A data analysis unit that outputs an analysis result for input data;
A simulation data generation unit that acquires original data of the input data, generates a simulation data that has characteristics different from the characteristics of the original data, and that simulates the input data;
A data sorting unit that sorts, from the simulated data, data for verification of analysis performance of the data analysis unit and data for learning of an algorithm of analysis processing by the data analysis unit;
A learning unit for learning the algorithm based on the selected data for learning;
A verification unit for acquiring the analysis result using the learned algorithm for the input data when the verification data is the input data, and verifying the analysis performance based on the analysis result Learning support system. The original data includes past input data to the data analysis unit and analysis results on the input data,
The simulated data generation unit
The learning support system according to claim 1, wherein simulation data is generated having characteristics different from the characteristics of the past input data and simulating new input data. The verification unit
When arbitrary data separately set for the sorted verification data is used as the input data, the analysis result using the learned algorithm for the input data is acquired, and the analysis result is obtained as the analysis result. The learning support system according to claim 1, wherein the analysis performance is verified based on the information. A computer that outputs an analysis result by analysis processing on input data,
Obtaining original data of the input data, and generating simulation data having characteristics different from the characteristics of the original data and simulating the input data;
The verification data of the performance of the analysis process and the learning data of the algorithm of the analysis process are respectively selected from the simulated data,
The learning of the algorithm is performed based on the selected data for learning.
A learning support method for acquiring the analysis result using the learned algorithm for the input data when the verification data is the input data, and verifying the performance of the analysis process based on the analysis result . Computer,
Data analysis unit that outputs analysis results for input data,
A simulation data generation unit that acquires original data of the input data, and generates simulation data that has characteristics different from the characteristics of the original data and that simulates the input data,
A data sorting unit that sorts, from the simulated data, data for verification of analysis performance of the data analysis unit and data for learning of an algorithm of analysis processing by the data analysis unit;
A learning unit that performs learning of the algorithm based on the selected learning data, and the analysis result using the learned algorithm for the input data when the verification data is the input data The learning support processing program for functioning as a verification part which acquires and verifies said analysis performance based on the said analysis result.