Information processing device and information processing program

Abstract

To provide an information processing apparatus configured to provide information so as not to make an immediate change on each user and the whole of users of a service.SOLUTION: An information processing apparatus 100 includes: an information generation unit 11 which generates multiple pieces of output information based on input information; an information output unit 12 which acquires multiple pieces of output information from the information generation unit 11 and outputs at least one piece of output information; and an information management unit 13 which gradually updates the output information output from the information output unit 12.SELECTED DRAWING: Figure 4

Description

[Technical field]

[0001] The present invention relates to an information processing device and an information processing program. [Background technology]

[0002] Businesses that provide various services (hereinafter, simply referred to as businesses) provide users with information necessary for the services. The information provided may include, for example, software (programs), data, and work procedures used by users, as well as the results of processing using machine learning models created by the businesses.

[0003] This information is updated by the provider as appropriate and provided to users. If the updated information is provided to many users at once, any defects in the information will have a large impact. Therefore, in order to reduce the impact of defects, the updated information is provided to some users first, and if no defects are detected there, or after any detected defects have been corrected, it is provided to other users. This type of two-stage or multi-stage information provision is called a canary release (Patent Document 1, Non-Patent Document 1, etc.). [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2017-33136 A [Non-patent literature]

[0005] [Non-Patent Document 1] [Canary Release in 3 minutes] Hideo Nakayama, Nikkei xTECH, released on August 21, 2019, URL = https: / / xtech.nikkei.com / atcl / nxt / keyword / 18 / 00002 / 081900087 / Summary of the Invention [Problem to be solved by the invention]

[0006] A canary release is a gradual change to the overall user population as the percentage of users affected by the change in information changes. Therefore, a canary release is a gradual change to the overall user population of a service.

[0007] However, for individual users who have been provided with the updated information in advance, this will be a sudden change as they will now be receiving information that is different from what they had before.

[0008] Therefore, an object of the present invention is to provide an information processing device and an information processing program that can provide information without causing abrupt changes to both all users of a service and to individual users. [Means for solving the problem]

[0009] The above object of the present invention can be achieved by the following means.

[0010] (1) an information generating unit that generates a plurality of pieces of output information based on input information; an information output unit that acquires a plurality of pieces of output information from the information generation unit and outputs at least one piece of output information; an information management unit that gradually updates the output information output from the information output unit; An information processing device having the above configuration.

[0011] (2) the information generation unit generates first information and second information for at least one of the input information; The information processing device according to (1) above, wherein the information management unit updates the output information output from the information output unit from the first information to the second information.

[0012] (3) the information generation unit has a machine learning model, The information processing device described in (2) above, wherein when the machine learning model is updated, output data obtained by processing input data input as the input information using the machine learning model before the update is generated as the first information, and output data obtained by processing input data using the machine learning model after the update is generated as the second information when the machine learning model is updated.

[0013] (4) The information processing device according to (2) or (3) above, wherein the information management unit updates the first information to the second information by changing the first information at a predetermined rate.

[0014] (5) The information processing device according to (2) or (3) above, wherein the information management unit calculates a weighted average of the first information and the second information while changing the weighting, and sets the calculation result as the output information.

[0015] (6) generating a plurality of pieces of output information based on input information; (b) outputting at least one piece of output information from among the plurality of pieces of output information generated while gradually updating the output information to be output; An information processing program having the above structure.

[0016] (7) the step (a) generates first information and second information for at least one of the input information; The information processing program according to (6) above, wherein the step (b) updates the output information from the first information to the second information.

[0017] (8) The step (a) includes performing processing using a machine learning model; The information processing program described in (7) above, wherein when the machine learning model is updated, output data obtained by processing input data input as the input information using the machine learning model before the update is generated as the first information, and output data obtained by processing input data using the machine learning model after the update is generated as the second information.

[0018] (9) The information processing program according to (7) or (8) above, wherein in the step (b), the first information is updated to the second information by changing the first information at a predetermined rate.

[0019] (10) The information processing program according to (7) or (8) above, wherein the step (b) calculates a weighted average of the first information and the second information while changing the weighting, and sets the calculation result as the output information. Effect of the Invention

[0020] In the present invention, when providing information to each user, at least one piece of information is output from among the multiple pieces of information generated, and the output information is updated stepwise. This makes it possible to prevent a sudden change in the information provided from occurring to the user receiving the information, even if the information provided changes. [Brief description of the drawings]

[0021] [Figure 1] FIG. 1 is a schematic diagram for explaining an information providing system. [Diagram 2] FIG. 2 is a block diagram showing an example of a hardware configuration of an information processing device. [Diagram 3] FIG. 2 is a block diagram showing an example of hardware of a user terminal. [Figure 4] FIG. 2 is a block diagram illustrating functions of the information processing device. [Diagram 5] 13 is a flowchart showing a procedure for updating information at a predetermined rate. [Figure 6] FIG. 11 is a data diagram for explaining an example of updating information at a predetermined rate. [Figure 7] 13 is a flowchart showing a procedure for updating information by weighted averaging. [Figure 8] FIG. 13 is a data diagram for explaining an example of updating information by weighted averaging. [Figure 9] FIG. 13 is a conceptual diagram for explaining the effect of the update process. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0022] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. In the description of the drawings, the same elements are given the same reference numerals, and duplicated descriptions will be omitted. In addition, the dimensional ratios of the drawings are exaggerated for the convenience of explanation, and may differ from the actual ratios.

[0023] (Information provision system) First, an information providing system that connects a business providing information with a user receiving the information will be described. Fig. 1 is a schematic diagram for explaining the information providing system.

[0024] As shown in FIG. 1, in this system, an information processing device 100 of a business providing information and user terminals 300 (301 to 303) of users receiving the information are connected via a network 200.

[0025] (Information processing device) The information processing device 100 is a computer. The information processing device 100 provides various software (programs), data, work procedures (manuals), and processing results of machine learning models by artificial intelligence (AI) to the user terminal 300 as information.

[0026] FIG. 2 is a block diagram showing an example of a hardware configuration of the information processing device 100. As shown in FIG.

[0027] 2, the information processing device 100 includes a CPU (Central Processing Unit) 110, a ROM (Read Only Memory) 120, a RAM (Random Access Memory) 130, a storage 140, a communication interface 150, and an operation display unit 160. Each component is connected to each other via a bus 190 so as to be able to communicate with each other.

[0028] The CPU 110 executes programs stored in the ROM 120 or the storage 140 to execute the functions of each unit, which will be described later.

[0029] The ROM 120 stores various programs and various data.

[0030] The RAM 130 serves as a working area for temporarily storing programs and data.

[0031] The storage 140 stores various programs including an operating system and various data. In this embodiment, the storage 140 particularly stores an information processing program for updating information that is created according to a procedure described below.

[0032] In the case of the information processing device 100, a large-capacity storage medium such as a hard disk drive (HDD) is mainly used as the storage 140. In addition, a semiconductor storage medium such as an embedded multi media card (eMMC) or a solid state drive (SSD) may be used as the storage 140 together with or instead of the HDD.

[0033] The communication interface 150 transmits and receives (communicates) data to and from the user terminal 300. To communicate with the connected network 200, the communication interface 150 uses, for example, a network interface based on standards such as Ethernet (registered trademark), SATA, PCI Express, USB, or IEEE1394, or a wireless communication interface such as Bluetooth (registered trademark) or IEEE802.11.

[0034] The operation display unit 160 is, for example, a touch panel display that displays various information and accepts various inputs. The operation display unit 160 may be connected to input devices such as a keyboard and a mouse, a monitor, and the like.

[0035] (network) The network 200 connects the information processing device 100 and the user terminal 300 based on any one of the above-mentioned communication standards. Therefore, the network 200 may be wired or wireless. The network 200 may also be a wide area network such as the Internet.

[0036] (User device) The user terminal 300 is, for example, a mobile terminal 301 that the user can carry around, such as a smartphone or a tablet computer, a portable notebook computer 302, a fixed desktop computer 303, or the like.

[0037] The mobile terminal 301 is, for example, carried individually by a user and receives information provided by a carrier.

[0038] The notebook computer 302 is used by an individual user or a group of multiple users to receive information provided by a business operator.

[0039] The desktop computer 303 is used by an individual user or a group of users to receive information provided by a business operator.

[0040] In the description of the embodiment, the mobile terminal 301, the notebook computer 302, and the desktop computer 303 will be referred to as the user terminal 300 unless there is a particular need to distinguish between them.

[0041] FIG. 3 is a block diagram showing an example of hardware of the user terminal 300.

[0042] The user terminal 300 is a computer. As shown in FIG. 3, the user terminal 300 has a CPU 310, a ROM 320, a RAM 330, a storage 140, a communication interface 350, and an operation display unit 360. Each component is connected to each other via a bus 390 so as to be able to communicate with each other. The basic configuration of each of these components is similar to that of the computer of the information processing device 100 described above, and therefore detailed description thereof will be omitted. In this embodiment, the user terminal 300 mainly executes various processes based on information provided by the information processing device 100, and also displays the provided information.

[0043] 1, three user terminals 300 (three types) are shown, but in an actual service, more user terminals 300 are used. Also, the user terminals 300 used may be of the same type or different types.

[0044] (Functions of information processing device) Next, the functions of the information processing device 100 will be described.

[0045] FIG. 4 is a block diagram for explaining the functions of the information processing device 100. As shown in FIG.

[0046] The information processing device 100 has, as its functions, an information generating section 11, an information output section 12, and an information managing section 13.

[0047] The information generating unit 11 generates a plurality of pieces of output information (output data) based on the input information (input data) that is input. The plurality of pieces of information generated by the information generating unit 11 are created, for example, by a machine learning model. When the machine learning model is updated, the results may differ before and after the update even if the input data is the same. In this embodiment, the information created by the machine learning model before the update (old model, the same below) is defined as the first information, and the information created by the machine learning model after the update (new model, the same below) is defined as the second information. Therefore, the information generating unit 11 generates at least the first information of the processing result by the old model and the second information of the processing result by the new model.

[0048] The information generating unit 11 is not limited to a machine learning model. The information generating unit 11 may simply perform some calculation on input data to generate output data. In this case, if the calculation formula is changed, the calculation result before the change becomes the first information, and the calculation result after the calculation formula is changed becomes the second information.

[0049] The information generated by the information generating unit 11 differs depending on the service provided by the business operator.

[0050] The information output unit 12 outputs at least one piece of information (first information and second information) generated by the information generation unit 11. The output destination is at least one user terminal 300 among the multiple user terminals 300. The user terminal 300 that is the output destination is predetermined by the operator. The user terminal 300 that is the output destination is, for example, a part of the multiple user terminals 300, similar to the conventional canary release. The user terminal 300 that is the output destination may also be determined based on a request from the user side. In particular, when the user terminal 300 that is the output destination is determined based on a request from the user, the update (release) of the information may be information customized for that specific user. The customized information is generated in advance by the information generation unit 11 for a specific user.

[0051] The information management unit 13 gradually updates the information to be output by the information output unit 12 from among the multiple pieces of information generated by the information generation unit 11 so that it becomes the second information from the first information. Specifically, for example, the first information is updated to the second information in multiple batches at a predetermined ratio (described in detail below). Also, for example, a weighted average of the first information and the second information is calculated while changing the weighting, and the first information is gradually updated to become the second information (described in detail below).

[0052] (Information update procedure) The procedure for updating (releasing) information will be described. First, an example of updating at a predetermined rate will be described. The information is updated by the CPU 110 executing an information processing program created according to this procedure, thereby implementing the functions of each of the above-mentioned units.

[0053] FIG. 5 is a flowchart showing a procedure for updating information at a predetermined rate.

[0054] First, the CPU 110, as a function of the information generating unit 11, receives input data (input information) (S101).

[0055] Next, the CPU 110, as a function of the information generating unit 11, executes a predetermined process on the input data and outputs the result (S102 to S105). Here, an example will be described in which the machine learning model is updated (changed). In the case of a machine learning model, the CPU 110 executes a process using the updated machine learning model (new model) (S102) and outputs the result as output data (output information (second information)) (S103). The CPU 110 also executes a process using the machine learning model before the update (old model) (S104) and outputs the result as output data (output information (first information)) (S105). In this embodiment, multiple pieces of output information are generated for the input information. The input information may be one or multiple pieces of input information, and differs depending on the processing content of the machine learning model. That is, multiple pieces of output information may be generated from one piece of input information or multiple pieces of input information.

[0056] Next, CPU 110, as a function of information management unit 13, determines whether to output the output data of the new model or the output data of the old model based on a predetermined ratio from the generated output data of the new and old models (S106).

[0057] Here, the predetermined ratio is set as a parameter. The parameter is, for example, the number of days, the rate of decrease, the rate of increase, etc., and is set according to the service to be implemented. The set parameter is stored in storage 140 (storage unit). Specifically, for example, when the number of days is set as a parameter, the number of output data to be updated at one time (update ratio) is also set. CPU 110 reads the parameter from storage 140 (may be read into RAM 130), and when the set number of days has passed, selects output data from the output data of the new and old models such that the number of output data of the new model is equal to the set update ratio. At this time, the output data of the new model to be selected is selected, for example, in the order of the output numbers of each output data (increasing number, decreasing number, every other number, every third number, etc.), the number of times equal to the update ratio.

[0058] The parameter may be, for example, a decrease rate or an increase rate. When a decrease rate or an increase rate is set, the output data in which the output data of the new model reaches a predetermined decrease rate or increase rate relative to the output data of the old model is selected.

[0059] Next, CPU 110, as the function of information output section 12, outputs the selected output data to user terminal 300 (S107). User terminal 300 as the output destination is determined in advance, as already described.

[0060] Next, the CPU 110, as a function of the information management unit 13, determines whether or not all of the output data has been changed to output data of the new model (S108). If all of the output data has not been changed to output data of the new model (S108: NO), the process returns to step S101 and continues. On the other hand, if all of the output data has been changed to output data of the new model (S108: YES), the CPU 110 ends the process.

[0061] FIG. 6 is a data diagram for explaining an example of updating information at a predetermined rate.

[0062] The update example shown in FIG. 6 is information processing when the machine learning model is updated. The machine learning model outputs numerical values ​​from 1 to 4 as output data for one piece of input data. In FIG. 6, the output data by the machine learning model before the update is referred to as the new model value, and the output data by the machine learning model after the update is referred to as the old model value. Furthermore, the machine learning model generates 15 pieces of output data (output numbers 1 to 15) in one processing.

[0063] In this update example, the parameter is updated by 20% each time (for one day). Also, the update is performed every 4th output number. As shown in FIG. 6, on the first day, 20% of the 15 output data are updated to the new model value. Here, the actual outputs of output numbers 4, 8, and 12 become the new model value. On the second day, 40% of the 15 output data (previous update + 20%) are updated to the new model value. Here, since the update is performed every 4th output number, the intermediate numbers are updated. Therefore, the actual outputs of output numbers 2, 6, 10, and 14 become the new model value. Then, on the 5th day, which is the last update day, all of the 15 output data (100%) are updated to the new model value. The update interval may be determined arbitrarily according to the contents of the service, for example, every few days, every few weeks, or every few months.

[0064] Next, an example of using weighted averaging as another information updating procedure will be described below. Fig. 7 is a flow chart showing the procedure for updating information by weighted averaging.

[0065] First, the CPU 110 executes generation of output data in the same manner as steps S101 to S105 shown in FIG.

[0066] Next, the CPU 110 calculates a weighted average between the output data of the old model and the output data of the new model (S116). The formula for calculating the weighted average is (w1x1+w2x2) / (w1+w2), where w1 is the weight (importance) of x1, and w2 is the weight of x2. In this embodiment, the weight when calculating the weighted average is set so as to change from the first information to the second information.

[0067] Fig. 8 is a data diagram for explaining an example of updating information by weighted averaging. The example of Fig. 8 is a case where the machine learning model is updated, similar to the example of Fig. 6. The example of Fig. 8 is a case where the weighting is changed in 20% increments so that the old model value (first information) changes stepwise to the new model value (second information).

[0068] In this example, CPU 110 sets the weighting to old model:new model = w1:w2 = 80:20 in the first update. Therefore, for output number 1, the old model value is x1 = 1 and the new model value is x2 = 1, so (80 x 1 + 20 x 1) / (80 + 20) = 1. For output number 2, the old model value is x1 = 2 and the new model value is x2 = (none), so in this case, no data = 0, so (80 x 1 + 20 x 0) / (80 + 20) = 0.8. CPU 110 performs similar calculations up to output number 15.

[0069] In the second update, CPU 110 assigns weights as old model:new model = w1:w2 = 60:40. Therefore, for output number 1, the old model value is x1 = 1 and the new model value is x2 = 1, so (60 x 1 + 40 x 1) / (60 + 40) = 1. For output number 2, the old model value is x1 = 4 and the new model value is x2 = 1, so (60 x 4 + 40 x 1) / (60 + 40) = 2.8. CPU 110 performs similar calculations up to output number 15.

[0070] Then, in the final, fifth update, CPU 110 sets the weighting as old model:new model = w1:w2 = 0:100. Therefore, for output number 1, the old model value is x1 = 2 and the new model value is x2 = 1, so (0 x 2 + 100 x 1) / (0 + 100) = 1. For output number 2, the old model value is x1 = 2 and the new model value is x2 = 1, so (0 x 2 + 100 x 1) / (0 + 100) = 1. CPU 110 performs similar calculations up to output number 15.

[0071] Here, the update may be performed at specific intervals as in Fig. 6, but may also be performed, for example, every time the information processing device 100 receives an update request from the user terminal 300. In this case, the information processing device 100 may notify the user terminal 300 that update data remains until all updates are completed.

[0072] Next, the CPU 110, as a function of the information output unit 12, outputs the calculated weighted average value to the user terminal 300 (S117).

[0073] Next, the CPU 110, as a function of the information management unit 13, determines whether or not all of the output data has been changed to output data of the new model (S108). If all of the output data has not been changed to output data of the new model (S108: NO), the process returns to step S101 and continues. On the other hand, if all of the output data has been changed to output data of the new model (S108: YES), the CPU 110 ends the process.

[0074] (effect) FIG. 9 is a conceptual diagram for explaining the effect of information updating according to this embodiment.

[0075] As shown in Fig. 9, in the conventional update method of canary release (Patent Document 1 or Non-Patent Document 1), the update rate increases gradually for all users as time (period) passes. However, in the canary release, the update rate increases suddenly for individual users during the release period.

[0076] On the other hand, in this embodiment, the degree of update increases gradually over time (period) for both all users and individual users during the release period. Therefore, this embodiment can prevent sudden changes for both all users and individual users when information is provided (released) by the business operator.

[0077] (Application examples) Next, an application example of this embodiment will be described.

[0078] (Case 1) Example 1 is an application to the healthcare field. For example, an application to the healthcare field is to replace a "disease location identification model based on X-ray images" used by multiple doctors with a new model (Reference URL: https: / / www.siemens-healthineers.com / jp / press-room / press-releases / pr-20200629-ai-rad-chest-ct.html).

[0079] The challenges with traditional canary releases in the healthcare field are as follows:

[0080] (1) In the canary release, if the model itself turns out to be inappropriate, several doctors who use it early on will end up using incorrect information to make decisions.

[0081] (2) In addition, the canary release displays results based on an updated model, which can cause individual users (doctors) to feel uncomfortable with the sudden changes in the model output.

[0082] When this embodiment is applied, the following improvements are achieved.

[0083] (a) In this embodiment, the verification of the new model can be carried out in stages. Therefore, even if there is a problem with the new model, the defect can be discovered at a stage when the scope of its impact is small.

[0084] (b) Furthermore, in this embodiment, the model output changes gradually, so that individual users do not feel uncomfortable.

[0085] (c) This embodiment is also applicable to a disease location identification model that uses medical data such as CT (Computed Tomography), examination results, and MRI (Magnetic Resonance Imaging) examination results.

[0086] (d) Furthermore, this embodiment can be applied not only to diagnostic images, but also to tests using artificial intelligence (AI) and machine learning, such as disease probability prediction models based on structured data such as blood test results and urine test results.

[0087] (Case 2) Case 2 is an application to the behavior monitoring field. For example, applications to the behavior monitoring field include replacing models that detect dangerous human behavior using surveillance cameras installed in multiple locations (Reference URL: https: / / robotstart.info / 2020 / 01 / 31 / mji-sec-ai.html).

[0088] The challenges with traditional canary releases in the behavioral monitoring field are as follows:

[0089] (1) In the field of behavior monitoring, the operating environment (brightness, contrast, distance to people, etc.) differs for each surveillance camera. In such a situation, when a canary release is performed, the surveillance camera to which the new model has been applied first may operate in an environment different from that of a surveillance camera to which the new model has not been applied. If the applied model released as a new model is inappropriate, the risk of undetecting or falsely detecting dangerous behavior increases in the applicable surveillance camera.

[0090] When this embodiment is applied, the following improvements are achieved.

[0091] (a) Because the verification of the new model can be carried out in stages, even if there are surveillance cameras in the behavioral monitoring field described above that are incompatible with the new model, the new model will be applied gradually, making it possible to discover defects at a stage when the scope of impact is small.

[0092] (b) Furthermore, this embodiment can be applied not only to risky behavior detection, but also to object detection, human joint point detection, and segmentation.

[0093] (Case 3) Case 3 is an application to a monitoring system using surveillance cameras, similar to case 2. One application to a monitoring system is to customize the fall detection function of a monitoring sensor for each user in an elderly care facility (Reference URL: https: / / www.konicaminolta.com / jp-ja / care-support / index.html).

[0094] The challenges with conventional canary releases in monitoring systems are as follows:

[0095] (1) In a monitoring system, you want to release customized features to specific users (which can be specific individuals or specific groups) who are being monitored, but a canary release does not allow for releases targeted at such specific users.

[0096] (2) In the case of a monitoring system, the monitoring content may no longer be suitable for users to whom a new model has been released first.

[0097] When this embodiment is applied, the following improvements are achieved.

[0098] (a) This embodiment enables gradual release to specific users who are the subject of monitoring.

[0099] Example 1: When releasing a customized version of the monitoring content for specific users (e.g., daily judgments), the customized parts can be gradually replaced until all functions (daily judgments) are finally replaced with the customized functions.

[0100] Example 2: Output a weighted average of the fall probability calculated by the standard function and the fall probability calculated by the customized function, and gradually increase the weight so that the function is updated to the customized function.

[0101] (b) This embodiment also reduces the risk of applying customized fall detection features through a phased release.

[0102] (c) Furthermore, this embodiment is applicable not only to fall detection, but also to functions for detecting other actions such as falling, walking, sitting down, standing up, etc. Note that these detections may be made by artificial intelligence or a machine learning model.

[0103] (Case 4) Case 4 is application to industrial equipment. For example, when a model is optimized for a specific mold, there are cases where the model for that specific mold needs to be replaced with a new model due to aging (Reference URL: https: / / ai.brainpad.co.jp / case-study / 194 / ).

[0104] The challenges with traditional canary releases in industrial equipment are as follows:

[0105] (1) Canary release cannot be applied to a specific mold.

[0106] When this embodiment is applied, the following improvements are achieved.

[0107] (a) This embodiment allows for staged release for a particular mold.

[0108] (b) Furthermore, this embodiment can reduce the risk of applying a new model to a deteriorated mold by gradual release.

[0109] (c) This embodiment can also be applied when replacing a conventional system that does not include a machine learning model with a new machine learning model. In such a case, in this embodiment, values ​​in a conventional system that does not include a machine learning model can be gradually replaced with values ​​from a machine learning model. In particular, this embodiment can be applied even if the internal configurations of the old and new systems and models are different, since it only compares the output results of the systems and models.

[0110] (d) Inspection of industrial equipment can be applied not only to molds but also to detecting abnormalities in the inspection process of various items produced in factories, such as food.

[0111] Although the embodiments and examples of application of the present invention have been described above, the present invention is not limited to such embodiments and examples.

[0112] The above-mentioned embodiments and examples have been described by taking as an example a case where the results of processing using new and old machine learning models are updated as output information. However, the present invention is not limited to the results of processing using machine learning models, and can also be applied to the provision and updating of various software (programs), data, or work procedures (manuals) provided by businesses.

[0113] In this embodiment, the information generating unit 11 is disposed in the information processing device 100. However, the present invention is not limited to this, and for example, a computer that implements a machine learning model may be separately prepared, and the processing results of the machine learning model by the computer may be received as multiple pieces of output information. In this case, the separately prepared computer becomes the information generating unit 11.

[0114] The information processing program according to the present invention can also be realized by a dedicated hardware circuit. The information processing program can also be provided by a computer-readable recording medium such as a Universal Serial Bus (USB) memory or a Digital Versatile Disc (DVD)-Read Only Memory (ROM), or can be provided online via a network 200 such as the Internet, regardless of the recording medium. When provided online, the information processing program is recorded on a recording medium (storage) such as a magnetic disk in a computer connected to the network 200.

[0115] Furthermore, the present invention allows various modifications based on the configurations described in the claims, and these modifications are also within the scope of the present invention. [Explanation of symbols]

[0116] 11 Information generation section, 12 Information output unit, 13 Information Management Department, 100 Information processing device, 110 CPUs, 150 communication interface, 160 Operation display section, 200 networks, 300 user terminals.

Claims

1. an information generation unit that generates first and second output information using a first machine learning model before the update and a second machine learning model after the update, respectively, based on input information that has been input; an information output unit that acquires the first and second output information from the information generation unit and outputs one of the first and second output information based on a set update rate α; an information management unit that gradually updates the output information output from the information output unit; having The information management unit gradually increases the update rate α, and the information output unit selects and outputs the second output information for a number of ratios obtained by multiplying the number of pieces of input information by the update rate α, and selects and outputs the first output information for a number of ratios obtained by multiplying the number of pieces of input information by 1-α.

2. The information processing device described in Claim 1, wherein the information management unit increases the update ratio α at a predetermined increase rate every predetermined period based on set parameters.

3. An information generation unit that generates first and second output information using a first machine learning model before the update and a second machine learning model after the update, respectively, based on input information that has been input; the first and second output information being numerical information, an information output unit that acquires the first and second output information from the information generation unit, and outputs a weighted average of the first and second output information based on weights w1 and w2 set for the first and second output information, respectively; an information management unit that gradually updates the output information output from the information output unit; having The information management unit gradually increases the weight w1 and decreases the weight w2, and the information output unit outputs the weighted average result calculated using the changed weights w1 and w2 as updated output information.

4. The information processing device of claim 3, wherein the information management unit increases the weight w1 and decreases the weight w2 at a predetermined rate every predetermined period based on set parameters.

5. A step (a) of generating first and second output information using a first machine learning model before the update and a second machine learning model after the update, respectively, based on input information that has been input; (b) outputting one of the first and second output information based on a set update ratio α from the generated first and second output information; (c) gradually updating the output information to be output; having In the step (c), the update rate α is increased stepwise, and in the step (b), the second output information is selected and output for a number of rates obtained by multiplying the number of pieces of input information by the update rate α, and the first output information is selected and output for a number of rates obtained by multiplying the number of pieces of input information by 1-α.

6. An information processing program as described in claim 5, wherein in the step (c), the update ratio α is increased at a predetermined increase rate for each predetermined period based on set parameters.

7. A step (a) of generating first and second output information using a first machine learning model before the update and a second machine learning model after the update, respectively, based on input information that has been input; the first and second output information being numerical information, (b) performing a weighted average of the first and second output information based on weights w1 and w2 set for the first and second output information, respectively; (c) gradually updating the output information outputted in the step (b); having In the step (c), the weight w1 is gradually increased and the weight w2 is gradually decreased, and in the step (b), the weights w1 and w2 after the changes are used to calculate the weighted average, and the result is output as updated output information.

8. An information processing program as described in claim 7, wherein in the step (c), the weight w1 is increased and the weight w2 is decreased at a predetermined rate every predetermined period based on set parameters.

Images