JP2021174335A - Information processing device and program - Google Patents

Abstract

To provide an information processing device and program capable of executing machine learning by using data acquired in the past in executing the machine learning again after stopping the execution of the machine learning.SOLUTION: An information processing device includes a storage part for storing a learning model, and a control part including an acquisition part, a learning part and an execution control part. The learning model is generated by machine learning. The acquisition part acquires learning data to be used for machine learning. The learning part executes the machine learning to the learning model by using the learning data. The execution control part executes the machine learning by using a first dataset being a dataset including at least a portion of the learning data acquired by the acquisition part during a stop period in the case of executing the machine learning after the stop period when the execution of the learning is stopped.SELECTED DRAWING: Figure 3

Description

The present invention relates to an information processing device and a program.

In recent years, technology using machine learning has become widespread.

Patent Document 1 discloses a technique for reflecting and integrating the results of individual learning in a new model by comparing and retraining the individual model and the new common model in the individual model derived from the common model. Has been done.

JP-A-2018-147261

By the way, when a learning model tuned for a specific organization is used, the accuracy of the learning model may temporarily decrease by continuing machine learning.

In view of the above circumstances, the present invention provides an information processing device and a program capable of executing machine learning using data acquired in the past when executing machine learning again after stopping the execution of machine learning. bottom.

According to one aspect of the present invention, the information processing apparatus includes a learning model, an acquisition unit, a learning unit, and an execution control unit. The learning model is generated by machine learning, and the acquisition unit is the machine. The learning data used for learning can be acquired, the learning unit can execute machine learning on the learning model using the learning data, and the execution control unit can execute the machine learning. When the machine learning is executed after the stopped period, the machine learning is configured to be executable using the first data set, and the first data set is acquired by the acquisition unit during the stopped period. An information processing device is provided, which is a data set containing at least a part of the learned data.

It is the schematic of the system 100 including the information processing apparatus 1 which concerns on one Embodiment of this invention. It is a figure which shows the hardware composition of the information processing apparatus 1. It is a functional block diagram of the control unit 10. It is a conceptual diagram of the learning model M including a plurality of versions. This is an example of training data. This is an example of an image displayed on the user terminal 2. It is an activity diagram which shows an example of the processing by the information processing apparatus 1.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. The various features shown in the embodiments shown below can be combined with each other.

By the way, the program for realizing the software appearing in the present embodiment may be provided as a non-temporary recording medium that can be read by a computer, may be provided as a downloadable form from an external server, or may be provided. It may be provided so that the program is started by an external computer and the function is realized (so-called cloud computing) by the client terminal.

Further, in the present embodiment, the "part" may include, for example, a combination of hardware resources implemented by a circuit in a broad sense and information processing of software that can be concretely realized by these hardware resources. .. In addition, various information is handled in this embodiment, and these information are, for example, physical values of signal values representing voltage and current, and signal values as a bit aggregate of a binary number composed of 0 or 1. It is represented by high-low or quantum superposition (so-called qubit), and communication / calculation can be executed on a circuit in a broad sense.

A circuit in a broad sense is a circuit realized by at least appropriately combining a circuit, a circuit, a processor, a memory, and the like. That is, an integrated circuit for a specific application (Application Special Integrated Circuit: ASIC), a programmable logic device (for example, a simple programmable logic device (Simple Programmable Logical Device: SPLD), a composite programmable logic device (Complex Program)) It includes a programmable gate array (Field Programmable Gate Array: FPGA) and the like.

1. 1. Outline of System 100 As shown in FIG. 1, the system 100 according to an embodiment of the present invention includes an information processing device 1. The function of the information processing device 1 is not particularly limited, but for example, it may provide an accounting service or a household account book service as a cloud type or an application. The information processing device 1 is configured to be able to communicate with the user terminal 2 used by the user and the cooperation system 3. In this embodiment, the system 100 further includes a user terminal 2 and a cooperation system 3. The user terminal 2 is configured to be able to use the service provided by the information processing device 1.

The cooperation system 3 is a system used for the operation of the organization. For example, the cooperation system 3 may be an attendance management system, a personnel management system, an expense settlement system, or a schedule management system.

Then, the information processing device 1, the user terminal 2, and the cooperation system 3 are connected to each other via the network NW. The information processing device 1 is composed of, for example, a server. The user terminal 2 is composed of an arbitrary computer, for example, a smartphone, a tablet terminal, a smart watch, a notebook PC, and a desktop PC.

2. Hardware Configuration of Information Processing Device 1 As shown in FIG. 2, the information processing device 1 includes a control unit 10, a communication unit 20, a storage unit 30, an input unit 40, and a display unit 50.
The control unit 10 is a CPU (Central Processing Unit) or the like, and controls the entire information processing device 1.
The communication unit 20 is a NIC (Network Interface Card) or the like, and is configured to be capable of data communication with other information processing devices or components by wire or wirelessly.
The storage unit 30 stores various programs and data, and is composed of, for example, a memory, an HDD (Hard Disk Drive), a ROM (Read Only Memory), a RAM (Random Access Memory), and the like. In addition, the storage unit 30 stores programs, various data, and the like. Then, various functions are realized by the control unit 10 executing various processes based on the program stored in the storage unit 30.

In the present embodiment, the storage unit 30 may store the learning model M shown in FIG. The learning model M is generated by machine learning. The learning model M may include a first learning model M1 and a second learning model M2. In the example of FIG. 4, the learning model M includes the first learning model M1 to the Nth learning model MN. Here, the second learning model M2 is a learning model generated by machine learning for the first learning model M1. That is, in the present embodiment, when additional machine learning is executed for a certain learning model and a new learning model is generated, the learning models before and after the learning are managed as different versions.

The input unit 40 inputs various information to the information processing device 1, and is composed of a mouse, a keyboard, a pointing device, and the like.
The display unit 50 displays text and images (including still images and moving images), and is composed of an arbitrary display. When the information processing device 1 is a server, the display unit 50 is composed of a display connected to the information processing device 1.

3. 3. Functions of Information Processing Device 1 As shown in FIG. 3, the control unit 10 of the information processing device 1 includes an acquisition unit 11, a learning unit 12, and an execution control unit 13. In the present embodiment, the control unit 10 further includes a designation unit 14, a selection unit 15, a test unit 16, an output unit 17, and a detection unit 18.

(Acquisition unit 11)
The acquisition unit 11 is configured to be able to acquire learning data used for machine learning. Here, regardless of the type of learning data used for machine learning, various data can be used depending on the application destination of the learning model M. Specifically, when the information processing device 1 is applied as an accounting service as a cloud type or an application, in addition to the input amount, item, account, date, department name, etc., the set account, etc. are added. Acquire as training data. In other words, the acquisition unit 11 may acquire input data related to accounting processing or tax processing. Further, the acquisition unit 11 may acquire input data regarding a project, a product, a department, a branch office, a region, a business partner, and the like.

(Learning Department 12)
The learning unit 12 is configured to be able to execute machine learning on the learning model M using the learning data. Here, the machine learning method can be appropriately selected. For example, when a neural network is used, input values based on various parameters are input to the neural network. The neural network can be implemented as software or hardware, and can be implemented, for example, on the firmware of the information processing apparatus 1. Here, the predetermined weight w for each calculation node N is set by machine learning in advance by the information processing device 1 or another information processing device.

(Execution control unit 13)
The execution control unit 13 controls the type and amount of learning data used for machine learning by the learning unit 12, the period of use as learning data, the execution timing or frequency of machine learning, and the like. In the present embodiment, the execution control unit 13 is configured to be able to execute machine learning using the first data set when the machine learning is executed after the stop period in which the execution of the machine learning is stopped. Here, the first data set is a data set including at least a part of the learning data acquired by the acquisition unit 11 during the stop period.

As shown in FIG. 5, for example, after performing machine learning to train the learning model M using the data acquired in the period T0 from before January 1 to January 1, January. If the execution of machine learning is stopped during the period Ts from the 2nd to the 30th of June (present), the period Ts becomes the suspension period. Then, the learning data LD1 to the learning data LDN acquired in the period Ts are stored in the storage unit 30 of the information processing device 1 or another information processing device. Here, the learning data LD1 is the learning data acquired on May 1, and the learning data LD2 is the learning data acquired on May 2. At this time, the first data set is data including at least a part of the learning data LD1 to the learning data LDN acquired during the stop period (period Ts).

Further, the execution control unit 13 may be configured to execute machine learning using the second data set. Here, the second data set is a data set including at least a part of the learning data acquired by the acquisition unit 11 in the acquisition period designated by the designation unit 14. As shown in FIG. 5, machine learning was not executed during the stop period (period Ts), but when machine learning is executed on June 30 (present), the learning data acquired during period T2 Machine learning can be performed using a second dataset that contains at least a portion. At this time, the period T2 is the acquisition period. Further, the data including at least a part of the learning data LD3 to the learning data LDN acquired in the acquisition period (period T2) becomes the second data set. Here, the learning data LD1 and the learning data LD2 are not included in the second data set because they are the data acquired before the acquisition period (period T2). That is, the data acquired in the exclusion period (period T1) excluding the acquisition period (period T2) in the suspension period (period Ts) is not used for machine learning.

Further, the execution control unit 13 may be configured so that the manual mode and the automatic mode can be switched. In the present embodiment, in the manual mode, machine learning, selection of the learning model M to be applied described later, or test processing described later is executed based on an instruction from the user. On the other hand, in the automatic mode, machine learning, selection of the learning model M to be applied, or test processing is automatically executed. That is, according to the automatic mode, for example, machine learning can be executed at a predetermined timing to update the learning model M. The manual mode and the automatic mode can be switched as appropriate. Further, different modes can be applied to a plurality of processes, such as applying a manual mode for machine learning and applying an automatic mode for test processing or selection of a learning model M to be applied.

(Designated part 14)
The designation unit 14 is configured to be able to specify the acquisition period of the learning data used for machine learning. For example, as shown in FIG. 6, the icon I representing the learning model M and the slider on which the acquisition period can be specified may be displayed together on the user terminal 2. Then, the user can specify the start date and the end date of the acquisition period by operating the user terminal 2. Here, the method of specifying the acquisition period is not limited to this, and the start date and end date of the acquisition period may be input in a format such as a text format, a pull-down format, or a drum roll format. Further, the acquisition period may be specified in a predetermined period unit (eg, a period from the present to 1 month, 3 months, 6 months, 12 months ago). Further, the acquisition period may be specified in hours, minutes, and seconds instead of days.

Further, the designation unit 14 may be configured so that the stop time of the execution of machine learning can be specified. In the example of FIG. 5, the start period of the stop period (period Ts) is designated by the designation unit 14.

(Selection unit 15)
The selection unit 15 is configured to be able to select the first learning model M1 or the second learning model M2 as the learning model to be applied. In the present embodiment, as shown in FIG. 4, different versions of the first learning model M1 and the second learning model M2 are stored as the learning model M, and the learning model selected by the selection unit 15 is used. Perform various processes.

(Test unit 16)
The test unit 16 is configured to be able to execute a test process for confirming the accuracy of the first learning model M1 and the second learning model M2. For example, the accuracy of a plurality of learning models M is confirmed by comparing the output signal output from the learning model M by inputting the learning data into the learning model M with a predetermined threshold value. At this time, when the selection unit 15 evaluates that the second learning model M2 has higher accuracy than the first learning model M1 as a result of the test processing by the test unit 16, the second learning is applied as a learning model. The model M2 is configured to be selectable. This is because, among the learning models M managed as a plurality of versions, various processes are executed by using the highly accurate learning model M.

(Output unit 17)
When the information processing device 1 provides the function of accounting processing or tax processing, the output unit 17 may be configured to input input data to the learning model M and output information related to accounting processing or tax processing. Specifically, by inputting input data to the learning model M learned from the learning data, information on accounting processing or tax processing is output. Here, as the input data, the purchase or advance amount, the item, the account, the date, the department name, and the like can be mentioned.

Further, the output unit 17 may be configured to input input data into the learning model M and output at least one of an account item, a sub-item, a description, and a tax classification corresponding to the input data. Specifically, by inputting the purchase or advance amount, item, date, department name as input data into the learning model M, at least one of the account, sub-item, description, and tax classification corresponding to the input data can be obtained. It is output. In addition, if at least one of the account, sub-item, description, and tax classification manually entered by the user is different from the account estimated by the learning model M, it may be indicated to that effect.

Further, the output unit 17 may be configured to input input data into the learning model M and output at least one of tags, debits, and credits corresponding to the input data. Here, the journal data, which is an example of information on accounting or tax processing output by the learning model M, may include projects, products, departments, branches, regions, business partners, etc. estimated by the learning model M. .. At this time, in conjunction with the calendar application, man-hours or process control including the date and time, or budget management may be realized for each project.

(Detection unit 18)
The detection unit 18 is configured to be able to detect changes in the rules regarding accounting or tax processing. The type and content of the rules are not particularly limited. The changes to the rules are, for example:
<Rule change example 1>
Accountant A treated a specific expense as "Account X", but after changing to his successor Accountant B, he started to treat the same expense as "Account Y"<Rules Change example 2>
In the case of business model A, a specific expense was treated as "account X", but after adopting another business model B after changing the business format, the same expense should be treated as "account Y". <Example of rule change 3>
A company that used to treat the account of tea for visitors as "meeting fee" stopped providing tea for visitors at a certain time, and all employees used PET bottled tea as an emergency food in case of disaster prevention. When the amount of tea is purchased separately and stored in the company, and the account item of tea in PET bottles is treated as "consumables cost"<Example 4 of rule change>
When the rules are changed due to changes in the external environment, such as when "Account X" is abolished and "Account Y" is newly used due to changes in accounting standards stipulated by law, etc.

Here, the detection unit 18 may detect the change of the rule based on various data acquired from the cooperation system 3. For example, when the cooperation system 3 is a personnel management system, it is possible to grasp that the accounting manager has changed by acquiring the personnel information registered in the personnel management database. Then, when the person in charge of accounting is changed, it may be determined that the accounting rules have changed.

Further, the execution control unit 13 may be configured to execute machine learning using the third data set when the detection unit 18 detects that the rule has been changed. Here, the third data set is a data set including at least a part of the training data acquired by the acquisition unit 11 in the period from the change of the rule to the present. As a result, the execution stop of machine learning is released, and the learning model M can be trained by using the learning data corresponding to the new rule.

Further, when the detection unit 18 detects that the rule has been changed, it may indicate to the user that the rule may have been changed. Such a presentation method is not particularly limited, and is realized by a display on the user terminal 2, an audio output, or another method.

4. Outline of Processing Next, an outline of processing by the information processing apparatus 1 will be described with reference to FIG. 7.

First, the learning data is acquired by the acquisition unit 11 (A1). Then, in the machine learning stop period for the learning model M, when the stop period is canceled by the designated unit 14, and the acquisition period is specified by the designated unit 14, the second data set (the third data set in the figure). 2DS) is acquired (A2a). Specifically, as shown in FIG. 5, if you notice that the rules were changed on May 3 as of June 30 (current), the suspension period (period Ts) Among them, the second data set including at least a part of the learning data LD3 to the learning data LDN acquired in the period T2 can be used for machine learning. In this case, the second data set and the third data set are the same data set.

On the other hand, in the machine learning stop period for the learning model M, when the stop period is canceled by the designated unit 14 and the acquisition period is not specified by the designated unit 14, the first data set (the first data set in the figure). 1DS) is acquired (A2b). Specifically, as shown in Fig. 5, as of June 30 (current), if you notice that the rules have been changed at any time after January 2, stop. The first data set including at least a part of the learning data LD1 to the learning data LDN acquired in the period (period Ts) can be used for machine learning.

On the other hand, in the machine learning stop period for the learning model M, when the stop period is not canceled by the designated unit 14 and the stop period continues, the process is returned to A1.

Then, in A3, machine learning is executed using the second data set or the first data set.

On the other hand, if it is the learning period of machine learning for the learning model M after acquiring the learning data in A1, the process proceeds to A3, and a data set containing at least a part of the learning data acquired in the learning period is used. , Perform machine learning. Here, when the target of machine learning in A3 is the first learning model M1 included in the learning model M, the second learning model M2 is generated as a result of machine learning.

After that, in A4, the test unit 16 executes test processing on the first learning model M1 (learning model before the execution of machine learning) and the second learning model M2 (learning model after the execution of machine learning). Then, as a result of the test process, a learning model evaluated to have high accuracy is adopted.

When the use of the learning model M is continued, the process is returned to A1. On the other hand, when the use of the learning model M is terminated, the process is terminated.

Here, regarding machine learning, selection of the learning model M to be applied, or execution of the test process, it may be considered whether the mode applied by the execution control unit 13 is the manual mode or the automatic mode.

5. Summary According to the information processing device 1 according to the present embodiment, the learning of the learning model M can be stopped when the learning result is satisfied by the process described above. On the other hand, when the rule changes, the learning of the learning model M can be restarted. In particular, when learning data is acquired in the background during the stop period and learning is restarted, the learning model M can be relearned retroactively by using the learning data acquired during the stop period. It will be possible.

6. Others In the above description, the information processing device 1 is configured to have various functions, but these functions may be distributed to a plurality of information processing devices. For example, when the learning model M is implemented in the service used by the user, the learning model M and the information processing device 1 may be distributed in different housings.

Further, the generation request of the learning model M may be accepted by machine learning using the learning data acquired in an arbitrary period specified by the user.

Further, when there is a large change in the trend and a sufficient amount of learning data has not been acquired, the output unit 17 may stop the output of information related to accounting processing or tax processing. For example, the automatic proposal function of the account corresponding to the input data may be stopped.

Further, in addition to the accounting service or the household account book service as a cloud type or an application, the control of functional learning for the learning model M by the information processing device 1 may be applied. For example, when implementing a learning model M that is trained by strongly reflecting the characteristics of an organization for ERP (Enterprise Resource Planning), RPA (Robotic Process Automation), etc., the processing by the information processing apparatus 1 is suitable.

It may be provided in each of the following aspects.
The information processing apparatus includes a designated unit, the designated unit is configured to be able to specify an acquisition period of learning data used for the machine learning, and the execution control unit performs the machine learning using a second data set. An information processing apparatus that is configured to be executable and includes at least a part of learning data acquired by the acquisition unit during the acquisition period specified by the designation unit.
The information processing device includes a designated unit, and the designated unit is configured to be capable of designating a stop time of execution of the machine learning.
The information processing apparatus includes a selection unit, the learning model includes a first learning model and a second learning model, and the second learning model is a learning model generated by the machine learning with respect to the first learning model. The selection unit is an information processing apparatus configured to be able to select the first learning model or the second learning model as the learning model to be applied.
The information processing apparatus includes a test unit, the test unit is configured to be able to execute a test process for confirming the accuracy of the first learning model and the second learning model, and the selection unit is the test unit. As a result of the test processing, when the second learning model is evaluated to have higher accuracy than the first learning model, the second learning model can be selected as the learning model to be applied. Device.
In the information processing apparatus, the execution control unit is configured to be able to switch between a manual mode and an automatic mode. In the manual mode, the machine learning, the selection of the learning model to be applied, or the test process is performed by the user. An information processing apparatus that executes based on an instruction, and in the automatic mode, automatically executes the machine learning, the selection of the learning model to be applied, or the test process.
The information processing apparatus includes an output unit, the acquisition unit acquires input data related to accounting processing or tax processing, and the output unit inputs the input data to the learning model, and the accounting processing or the above An information processing device that can output information related to tax processing.
In the information processing device, the output unit is configured to input the input data into the learning model and output at least one of an account item, a sub-item, a description, and a tax classification corresponding to the input data. , Information processing equipment.
The information processing device includes a detection unit, and the detection unit is configured to be able to detect that a rule regarding the accounting process or the tax process has been changed.
In the information processing apparatus, when the detection unit detects that the rule has been changed, the execution control unit is configured to be able to execute the machine learning using the third data set, and the third data set. Is an information processing device, which is a data set including at least a part of the training data acquired by the acquisition unit in the period from the change of the rule to the present.
A program that causes a computer to function as each part of the information processing device.
Of course, this is not the case.

Further, the above-mentioned computer may be provided as a computer-readable non-temporary recording medium for storing a program for functioning as each part of the information processing apparatus.

Finally, various embodiments of the present invention have been described, but these are presented as examples and are not intended to limit the scope of the invention. The novel embodiment can be implemented in various other embodiments, and various omissions, replacements, and changes can be made without departing from the gist of the invention. The embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the scope of the invention described in the claims and the equivalent scope thereof.
The above-described embodiment and a plurality of modifications can be executed in any combination.

1: Information processing device 2: User terminal 3: Cooperation system 10: Control unit 11: Acquisition unit 12: Learning unit 13: Execution control unit 14: Designation unit 15: Selection unit 16: Test unit 17: Output unit 18: Detection unit 20: Communication unit 30: Storage unit 40: Input unit 50: Display unit 100: System I: Icon

Claims (11)

It is an information processing device
It has a learning model, an acquisition unit, a learning unit, and an execution control unit.
The learning model is generated by machine learning and
The acquisition unit is configured to be able to acquire learning data used for the machine learning.
The learning unit is configured to be able to execute machine learning on the learning model using the learning data.
When the machine learning is executed after the stop period in which the execution of the machine learning is stopped, the execution control unit is configured to be able to execute the machine learning using the first data set.
The first data set is a data set including at least a part of the training data acquired by the acquisition unit during the stop period.
Information processing device. In the information processing apparatus according to claim 1,
Equipped with a designated part
The designation unit is configured so that the acquisition period of the learning data used for the machine learning can be specified.
The execution control unit is configured to be able to execute the machine learning using the second data set.
The second data set is a data set including at least a part of the training data acquired by the acquisition unit during the acquisition period specified by the designation unit.
Information processing device. In the information processing apparatus according to claim 1 or 2.
Equipped with a designated part
The designation unit is configured so that the stop time of the execution of the machine learning can be specified.
Information processing device. In the information processing apparatus according to any one of claims 1 to 3.
Equipped with a selection
The learning model includes a first learning model and a second learning model.
The second learning model is a learning model generated by the machine learning with respect to the first learning model.
The selection unit is configured to be able to select the first learning model or the second learning model as the learning model to be applied.
Information processing device. In the information processing apparatus according to claim 4,
Equipped with a test section
The test unit is configured to be able to execute a test process for confirming the accuracy of the first learning model and the second learning model.
The selection unit uses the second learning model as a learning model to be applied when the second learning model is evaluated to have higher accuracy than the first learning model as a result of the test processing by the test unit. Configured to be selectable
Information processing device. In the information processing apparatus according to claim 5,
The execution control unit is configured to be able to switch between manual mode and automatic mode.
In the manual mode, the machine learning, the selection of the learning model to be applied, or the test process is executed based on the instruction from the user.
In the automatic mode, the machine learning, the selection of the applicable learning model, or the test process is automatically executed.
Information processing device. The information processing apparatus according to any one of claims 1 to 6.
Equipped with an output section
The acquisition department acquires input data related to accounting processing or tax processing, and
The output unit is configured to be able to input the input data into the learning model and output information related to the accounting process or the tax process.
Information processing device. In the information processing apparatus according to claim 7,
The output unit is configured to input the input data into the learning model and output at least one of an account item, a sub-item, a description, and a tax classification corresponding to the input data.
Information processing device. In the information processing apparatus according to claim 7 or 8.
Equipped with a detector
The detector is configured to be able to detect changes in the rules regarding accounting or tax processing.
Information processing device. In the information processing apparatus according to claim 9,
When the detection unit detects that the rule has been changed, the execution control unit is configured to be able to execute the machine learning using the third data set.
The third data set is a data set containing at least a part of the training data acquired by the acquisition unit in the period from the change of the rule to the present.
Information processing device. It ’s a program
A program that causes a computer to function as each part of the information processing apparatus according to any one of claims 1 to 19.

Images