Information processing device, information processing server, information processing method, and non-transitory computer readable storage medium

Abstract

To achieve both protection of privacy and high inference accuracy.SOLUTION: Provided is an information processing device comprising a learning unit and a communication unit. The learning unit clusters hierarchical data on the basis of a plurality of inference models distributed from an information processing server, and performs learning using the inference model corresponding to each cluster. The communication unit transmits intermediate results generated for each cluster in the learning by the learning unit to the information processing server. The hierarchical data includes information for identifying a main element, and a log collected or generated in association with the main element.SELECTED DRAWING: Figure 5

Description

Technical Field

[0001] The present disclosure relates to an information processing apparatus, an information processing server, an information processing method, and a non-transitory computer-readable storage medium.

Background Art

[0002] In recent years, development of models for making some inferences based on collected data has been carried out. Further, techniques for improving the accuracy of the above inferences have been proposed. For example, Patent Document 1 discloses a technique for clustering data used for inference.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, the technique disclosed in Patent Document 1 may not obtain sufficient effects depending on the structure of the data used for inference.

Means for Solving the Problems

[0005] According to an aspect of the present disclosure, there is provided an information processing apparatus including: a learning unit that clusters hierarchical data based on a plurality of inference models distributed from an information processing server, and performs learning using the corresponding inference model for each cluster; and a communication unit that transmits intermediate results generated for each cluster in the learning by the learning unit to the information processing server, wherein the hierarchical data includes information specifying main elements and logs collected or generated in association with the main elements.

[0006] Also, according to another aspect of the present disclosure, a processor clusters hierarchical data based on a plurality of inference models distributed from an information processing server, performs learning using the corresponding inference model for each cluster, and transmits intermediate results generated for each cluster in the learning to the information processing server. The hierarchical data includes information identifying main elements and logs collected or generated in association with the main elements, and an information processing method is provided.

[0007] Also, according to another aspect of the present disclosure, a computer is caused to function as an information processing apparatus including a learning unit that clusters hierarchical data based on a plurality of inference models distributed from an information processing server and performs learning using the corresponding inference model for each cluster, and a communication unit that transmits intermediate results generated for each cluster in the learning by the learning unit to the information processing server. The hierarchical data includes information identifying main elements and logs collected or generated in association with the main elements, and a non-transitory computer-readable storage medium storing a program for causing the computer to function as such is provided.

[0008] Also, according to another aspect of the present disclosure, an information processing server is provided, including a learning unit that generates a plurality of inference models respectively corresponding to a plurality of clusters, and a communication unit that transmits information related to the plurality of inference models generated by the learning unit to a plurality of information processing apparatuses. The communication unit receives intermediate results generated by learning based on hierarchical data clustered based on the plurality of inference models and the corresponding inference model for each cluster from the plurality of information processing apparatuses. The learning unit updates the plurality of inference models based on the plurality of intermediate results. The hierarchical data includes information identifying main elements and logs collected or generated in association with the main elements.

[0009] According to another aspect of the present disclosure, a processor generates a plurality of inference models respectively corresponding to a plurality of clusters, transmits information related to the plurality of generated inference models to a plurality of information processing apparatuses, receives intermediate results generated by learning based on hierarchical data clustered based on the plurality of inference models and the inference models corresponding to each cluster from the plurality of information processing apparatuses, and updates the plurality of inference models based on the plurality of intermediate results, wherein the hierarchical data includes information identifying main elements and logs collected or generated in association with the main elements, and an information processing method is provided.

[0010] According to another aspect of the present disclosure, a computer includes a learning unit that generates a plurality of inference models respectively corresponding to a plurality of clusters, and a communication unit that transmits information related to the plurality of inference models generated by the learning unit to a plurality of information processing apparatuses, wherein the communication unit receives intermediate results generated by learning based on hierarchical data clustered based on the plurality of inference models and the inference models corresponding to each cluster from the plurality of information processing apparatuses, the learning unit updates the plurality of inference models based on the plurality of intermediate results, and the hierarchical data includes information identifying main elements and logs collected or generated in association with the main elements, and a non-transitory computer-readable storage medium storing a program for causing the computer to function as an information processing server is provided.

Brief Description of the Drawings

[0011]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Mode for Carrying Out the Invention

[0012] Hereinafter, preferred embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. In the present specification and drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant description is omitted.

[0013] Note that the description will be made in the following order. 1. Embodiment 1.1. Outline 1.2. Example of System Configuration 1.3. Example of Configuration of Information Processing Apparatus 10 1.4. Example of Configuration of Information Processing Server 20 1.5. Details of Functions 1.6. Application Examples 2. Hardware Configuration Examples 3. Summary

[0014] <1. Embodiment> <<1.1. Overview>> First, the overview of one embodiment of the present disclosure will be described.

[0015] As described above, in recent years, models (inference models) for making inferences based on collected data have been developed.

[0016] According to the inference model, it is also possible to accurately make various inferences based on unknown data. Therefore, the generation and utilization of inference models are actively carried out in various fields.

[0017] However, depending on the generation of the inference model and the structure of the data used for inference by the inference model, it may be difficult to generate an inference model with high accuracy.

[0018] Examples of such data include, for example, hierarchical data.

[0019] Hierarchical data may be defined, for example, as data including information for identifying a main element and logs collected or generated in association with the main element.

[0020] As an example, assume a case where device 80 communicates with a plurality of other devices and makes some inferences based on the logs collected for each communication partner.

[0021] In this case, the main element may be the device that is the communication partner of device 80.

[0022] Also, in this case, characteristics corresponding to the communication partner, the pair of device 80 and the communication partner, etc. may occur in the generated logs.

[0023] Therefore, if learning using logs is performed without distinguishing communication partners, not only is it difficult to generate an inference model with high accuracy, but there is also a possibility of an increase in the time required for convergence, overfitting, etc.

[0024] On the other hand, for example, as disclosed in Patent Document 1, there is also a method of performing learning for each piece of data belonging to each cluster after clustering the data.

[0025] Thus, a method of classifying data with different characteristic tendencies into a plurality of clusters and performing learning for each piece of data belonging to each cluster seems to be useful also for hierarchical data.

[0026] However, with the method as disclosed in Patent Document 1, since the clustering result cannot be corrected based on the inference result, it is difficult to generate a highly accurate inference model.

[0027] Also, for example, when a server collects data from a plurality of devices and performs learning based on the data, protection of privacy also becomes an issue.

[0028] On the other hand, as a method for protecting the privacy of data when performing learning using data collected from a plurality of devices, there is also a method called Federated Learning.

[0029] However, general federated learning aims to generate a single inference model from data received from a plurality of devices.

[0030] Therefore, when using data having various tendencies such as hierarchical data in general federated learning, there is a possibility of an increase in the time required for convergence, overfitting, etc.

[0031] In order to reduce the above-mentioned possibilities, for example, a method of clustering the data according to the characteristics of the devices that collected the data and performing learning for each piece of data belonging to each cluster can also be considered.

[0032] However, in this case, all the data collected from a certain device will be classified into the same cluster.

[0033] Therefore, in a case where inference is performed using the logs generated for each of the plurality of communication partners described above, it may be difficult to reduce the possibility of an increase in the time required for convergence, overfitting, etc.

[0034] The technical idea according to an embodiment of the present disclosure was conceived by paying attention to the above points, and realizes both the protection of privacy and high inference accuracy.

[0035] For this reason, the information processing apparatus 10 according to an embodiment of the present disclosure includes a learning unit 120 that clusters hierarchical data based on a plurality of inference models distributed from the information processing server 20 and performs learning using the corresponding inference model for each cluster.

[0036] Also, the information processing apparatus 10 according to an embodiment of the present disclosure includes a communication unit 130 that transmits the intermediate results generated for each cluster in the above learning by the learning unit 120 to the information processing server 20.

[0037] That is, the information processing apparatus 10 according to an embodiment of the present disclosure clusters the hierarchical data held by itself using a plurality of inference models distributed from the information processing server 20 and transmits the intermediate results generated by learning for each cluster to the information processing server 20.

[0038] Here, the intermediate result may be information including values calculated from the feature amounts and labels included in the hierarchical data, and may be information for which it is difficult to restore the feature amounts and labels.

[0039] On the other hand, the information processing server 20 according to an embodiment of the present disclosure includes a learning unit 210 that generates a plurality of inference models respectively corresponding to a plurality of clusters, and a communication unit that transmits information related to the plurality of inference models generated by the learning unit 210 to the plurality of information processing apparatuses 10.

[0040] Here, the communication unit 220 is characterized in that it receives, from a plurality of information processing apparatuses 10, intermediate results generated by learning based on hierarchical data clustered based on a plurality of inference models and the inference models corresponding to each cluster.

[0041] Further, the learning unit 210 is characterized in that it updates a plurality of inference models based on a plurality of the intermediate results.

[0042] That is, the information processing server 20 according to an embodiment of the present disclosure updates an inference model corresponding to each cluster based on the intermediate results for each cluster received from a plurality of information processing apparatuses 10, and distributes information related to the update to each of the information processing apparatuses 10.

[0043] As described above, in the system 1 according to an embodiment of the present disclosure, clustering by the information processing apparatus 10, transmission of intermediate results, update of the inference model by the server, and distribution of information related to the update may be repeatedly executed.

[0044] According to the process as described above, it is possible to guarantee the convergence of the inference model and to improve the clustering system and the inference accuracy at the same time.

[0045] Also, according to the process as described above, it is possible to enhance the privacy protection performance by using intermediate results that are difficult to restore to the original data.

[0046] Hereinafter, a system configuration example for realizing the above will be described in detail.

[0047] <<1.2. System configuration example>> FIG. 1 is a block diagram showing a configuration example of a system 1 according to an embodiment of the present disclosure.

[0048] As shown in FIG. 1, the system 1 according to the present embodiment includes a plurality of information processing apparatuses 10 and an information processing server 20.

[0049] Each information processing apparatus 10 and the information processing server 20 are communicably connected to each other via the network 30.

[0050] Note that FIG. 1 illustrates a case where the system 1 includes the information processing apparatuses 10a and 10b, but the number of information processing apparatuses 10 according to the present embodiment is not limited to such an example.

[0051] (Information processing apparatus 10) The information processing apparatus 10 according to the present embodiment performs clustering of hierarchical data using an inference model distributed from the information processing server 20.

[0052] In addition, the information processing apparatus 10 according to the present embodiment performs learning using data belonging to each cluster and an inference model corresponding to the cluster, and transmits an intermediate result to the information processing server 20.

[0053] The information processing apparatus 10 according to the present embodiment may be, for example, a personal computer, a smartphone, a tablet, a game machine, a wearable device, or the like.

[0054] (Information processing server 20) The information processing server 20 according to the present embodiment generates an inference model corresponding to a set cluster and distributes it to a plurality of information processing apparatuses 10.

[0055] In addition, the information processing server 20 according to the present embodiment receives intermediate results corresponding to each cluster from a plurality of information processing apparatuses 10, and updates each inference model based on the intermediate results.

[0056] The information processing server 20 according to the present embodiment distributes information related to the update of each inference model to a plurality of information processing apparatuses 10.

[0057] (Network 30) The network 30 according to the present embodiment mediates communication between the information processing apparatus 10 and the information processing server 20.

[0058] <<1.3. Configuration Example of Information Processing Apparatus 10>> Next, a configuration example of the information processing apparatus 10 according to this embodiment will be described in detail.

[0059] FIG. 2 is a block diagram showing a configuration example of the information processing apparatus 10 according to this embodiment.

[0060] As shown in FIG. 2, the information processing apparatus 10 according to this embodiment may include a sensor unit 110, a learning unit 120, and a communication unit 130.

[0061] (Sensor Unit 110) The sensor unit 110 according to this embodiment collects various sensor information.

[0062] The sensor information collected by the sensor unit 110 may be used as an element (feature amount) of the hierarchical data.

[0063] For this purpose, the sensor unit 110 may include various sensors for collecting sensor information that can be used as an element of the hierarchical data.

[0064] On the other hand, when the sensor information is not included in the hierarchical data, the information processing apparatus 10 may not include the sensor unit 110.

[0065] (Learning Unit 120) The learning unit 120 according to this embodiment clusters hierarchical data based on a plurality of inference models distributed from the information processing server 20.

[0066] In addition, the learning unit 120 according to this embodiment performs learning using the inference model corresponding to each cluster.

[0067] As described above, the hierarchical data according to this embodiment may include information for specifying the main element and logs collected or generated in association with the main element.

[0068] Further, the learning unit 120 may cluster hierarchical data related to different main elements based on a plurality of inference models.

[0069] The functions of the learning unit 120 according to the present embodiment are realized by various processors.

[0070] Details of the functions of the learning unit 120 according to the present embodiment will be described separately.

[0071] (Communication unit 130) The communication unit 130 according to the present embodiment communicates with the information processing server 20 via the network 30.

[0072] The communication unit 130 receives, for example, an inference model, information related to the update of the inference model, etc. from the information processing server 20.

[0073] Further, the communication unit 130 transmits the intermediate result generated by the learning unit 120 to the information processing server 20.

[0074] Note that the communication unit 130 according to the present embodiment may communicate with another device different from the information processing server 20.

[0075] Also, in this case, the communication unit 130 may generate and store a log related to communication with another device.

[0076] The log related to communication with another device may be used as part of the hierarchical data.

[0077] The configuration example of the information processing apparatus 10 according to the present embodiment has been described above. Note that the above configuration described with reference to FIG. 2 is merely an example, and the configuration of the information processing apparatus 10 according to the present embodiment is not limited to such an example.

[0078] The information processing apparatus 10 according to the present embodiment may further include, for example, an input unit that receives input of information by a user, a display unit that displays various information, and the like.

[0079] In addition, when the information processing apparatus 10 includes an input unit, the input information may be used as part of the hierarchical data.

[0080] The configuration of the information processing apparatus 10 according to the present embodiment can be flexibly deformed according to specifications and operations.

[0081] <<1.4. Configuration Example of Information Processing Server 20>> Next, a configuration example of the information processing server 20 according to the present embodiment will be described in detail.

[0082] FIG. 3 is a block diagram showing a configuration example of the information processing server 20 according to the present embodiment.

[0083] As shown in FIG. 3, the information processing server 20 according to the present embodiment may include a learning unit 210 and a communication unit 220.

[0084] (Learning Unit 210) The learning unit 210 according to the present embodiment generates a plurality of inference models respectively corresponding to a plurality of clusters.

[0085] In addition, the learning unit 210 according to the present embodiment updates the plurality of inference models based on the plurality of intermediate results received by the communication unit 220.

[0086] The functions of the learning unit 210 according to the present embodiment are realized by various processors.

[0087] Details of the functions of the learning unit 210 according to the present embodiment will be described separately.

[0088] (Communication Unit 220) The communication unit 220 according to the present embodiment communicates with a plurality of information processing apparatuses 10 via the network 30.

[0089] The communication unit 220 transmits, for example, the inference model generated by the learning unit 210 and information related to the update of the inference model to the plurality of information processing apparatuses 10.

[0090] In addition, the communication unit 220 receives intermediate results from a plurality of information processing apparatuses 10.

[0091] The configuration example of the information processing server 20 according to the present embodiment has been described above. Note that the above configuration described with reference to FIG. 3 is merely an example, and the configuration of the information processing server 20 according to the present embodiment is not limited to such an example.

[0092] The information processing server 20 according to the present embodiment may further include, for example, an input unit that receives input of information by a user, a display unit that displays various types of information, and the like.

[0093] The configuration of the information processing server 20 according to the present embodiment can be flexibly modified according to specifications and operations.

[0094] <<1.5. Details of functions>> Next, the functions of each of the information processing apparatus 10 and the information processing server 20 according to the present embodiment will be described in detail.

[0095] As described above, the information processing apparatus 10 according to the present embodiment clusters the hierarchical data held by itself using a plurality of inference models distributed from the information processing server 20, and transmits the intermediate results generated by learning for each cluster to the information processing server 20.

[0096] In addition, the information processing server 20 according to the present embodiment updates the inference model corresponding to each cluster based on the intermediate results for each cluster received from a plurality of information processing apparatuses 10, and distributes the information related to the update to each of the information processing apparatuses 10.

[0097] In order to realize the above-described processing, it is required that the information processing apparatus 10 and the information processing server 20 share a model corresponding to the hierarchical data.

[0098] FIG. 4 is a diagram showing an example of a model corresponding to the hierarchical data according to the present embodiment.

[0099] On the left side of FIG. 4, an example of a graphical model corresponding to hierarchical data is shown, and on the right side of FIG. 4, an example of a generative model corresponding to hierarchical data is shown, respectively.

[0100] In each model shown in FIG. 4, η, θ, and κ correspond to the global set of the information processing apparatus 10, the information processing apparatus 10, and the main element, respectively.

[0101] However, each model shown in FIG. 4 is merely an example, and the graphical model and the generative model according to the present embodiment may be appropriately designed according to the characteristics of the hierarchical data, the characteristics of the inferred label (objective variable), and the like.

[0102] Subsequently, with reference to FIGS. 5 and 6, the flow of the process executed by the system 1 according to the present embodiment will be described in detail.

[0103] FIG. 5 is a schematic diagram for explaining the flow of the process executed by the system 1 according to the present embodiment.

[0104] Also, FIG. 6 is a flowchart showing the flow of the process executed by the system 1 according to the present embodiment.

[0105] In FIG. 5, the processing by the information processing apparatuses 10a and 10b is illustrated. However, as described above, the system 1 according to the present embodiment may include three or more information processing apparatuses 10.

[0106] Also, in FIG. 5, a case where the information processing server 20 generates three inference models M1 to M3 corresponding to three clusters C1 to C3 respectively is illustrated. However, the number of clusters and inference models according to the present embodiment is not limited to such an example.

[0107] The number of clusters and inference models according to the present embodiment may be appropriately designed according to the characteristics of the hierarchical data, the characteristics of the inferred label (objective variable), and the like.

[0108] First, as shown in FIG. 6, the information processing server 20 initializes the inference model (S100).

[0109] In the case of the example shown in FIG. 5, the information processing server 20 initializes the inference models M1 to M3.

[0110] Next, the information processing server 20 distributes information related to the inference model (S101).

[0111] In the case of the example shown in FIG. 5, the information processing server 20 may transmit all the information constituting the inference models M1 to M3.

[0112] Next, each of the information processing apparatuses 10 clusters the hierarchical data using the inference model (S102).

[0113] In the case of the example shown in FIG. 5, the information processing apparatus 10a classifies the hierarchical data D1 into any one of the clusters C1 to C3 corresponding to the inference models M1 to M3, respectively, using the distributed inference models M1 to M3.

[0114] Similarly, the information processing apparatus 10b classifies the hierarchical data D2 into any one of the clusters C1 to C3 corresponding to the inference models M1 to M3, respectively, using the distributed inference models M1 to M3.

[0115] Next, each of the information processing apparatuses performs learning for each cluster (S103).

[0116] In the case of the example shown in FIG. 5, the information processing apparatus 10a performs learning for each of the clusters C1 to C3, and generates intermediate results w 11 ~w 31 corresponding to the clusters C1 to C3, respectively.

[0117] Similarly, the information processing apparatus 10b performs learning for each of the clusters C1 to C3, and generates intermediate results w 12 ~w 32 corresponding to the clusters C1 to C3, respectively.

[0118] Next, each of the information processing apparatuses 10 transmits the intermediate result to the information processing server 20 (S104).

[0119] In the case of the example shown in FIG. 5, the information processing apparatus 10a transmits the intermediate results w 11 ~w 31 to the information processing server.

[0120] Similarly, the information processing apparatus 10b transmits the intermediate results w 12 ~w 32 to the information processing server 20.

[0121] Next, the information processing server 20 groups the intermediate results received for each cluster and updates the inference model corresponding to each cluster (S105).

[0122] In the case of the example shown in FIG. 5, the information processing server 20 calculates w1 from the intermediate results w 11 and w 12 and updates the inference model M1.

[0123] Similarly, the information processing server 20 calculates w2 from the intermediate results w 21 and w 22 and updates the inference model M2.

[0124] Similarly, the information processing server 20 calculates w3 from the intermediate results w 31 and w 32 and updates the inference model M3.

[0125] Subsequently, the information processing server 20 determines whether or not each inference model has converged (S106).

[0126] If the information processing server 20 determines that each inference model has converged (S106: Yes), the system 1 ends the series of processes.

[0127] On the other hand, if the information processing server 20 determines that each inference model has not converged (S106: Yes), the information processing server 20 returns to step S101 and distributes the information related to the inference model.

[0128] In the case of an example shown in FIG. 5, the information processing server 20 may transmit, for example, w1, w2, and w3 to the information processing apparatuses 10a and 10b.

[0129] When the information processing server 20 returns to step S101, the information processing apparatus 10 and the information processing server 20 repeatedly execute the subsequent processing.

[0130] As described above, an example of the processing flow by the system 1 according to the present embodiment has been described in detail.

[0131] Subsequently, the information transmitted and received between the information processing apparatus 10 and the information processing server 20 according to the present embodiment will be described in more detail.

[0132] FIG. 7 is a diagram showing an example of clustering of hierarchical data according to the present embodiment.

[0133] In the case of an example shown in FIG. 7, hierarchical data having main element IDs: ME1, main element ID: ME2, and main element ID: ME3 is classified into the cluster C1.

[0134] Further, hierarchical data having main element ID: ME4 and main element ID: ME5 is classified into the cluster C2.

[0135] Further, hierarchical data having main element ID: ME6, main element ID: ME7, and main element ID: ME8 is classified into the cluster C2.

[0136] Here, the main element ID is an example of information for specifying a main element.

[0137] Further, the hierarchical data includes, in addition to the main element ID, logs collected or generated in association with the main element.

[0138] The above logs may include, for example, feature amounts and labels (objective variables).

[0139] In an example shown in FIG. 7, the feature amount has five elements of x n1 ~x n5 , but this is merely an example, and the number of elements of the feature amount according to the present embodiment is not limited to such an example.

[0140] The intermediate result according to the present embodiment may be a value calculated from the feature amounts and labels included in the hierarchical data.

[0141] When the information processing apparatus 10 performs clustering of hierarchical data as shown in FIG. 7, the intermediate results w 11 , w 21 , w 31 corresponding to the clusters C1 to C3 may be set as follows.

[0142] w 11 ={A1, b1} w 21 ={A2, b2} w 31 ={A3, b3}

[0143] Here, A k and b k may be values calculated based on the feature amounts and labels belonging to the cluster Ck, respectively.

[0144] Examples of the calculation of A k and b k are shown below.

[0145] A1 = A(x 11 , x 12 , x 13 , x 14 , x 15 , y1, x 21 , x 22 , x 23 , x 24 , x 25 , y2, x 31 , x 32 , x 33 , x 34 , x35 , y3) b1 = b(x 11 , x 12 , x 13 , x 14 , x 15 , y1, x 21 , x 22 , x 23 , x 24 , x 25 , y2, x 31 , x 32 , x 33 , x 34 , x 35 , y3)

[0146] A2 = A(x 41 , x 42 , x 43 , x 44 , x 45 , y4, x 51 , x 52 , x 53 , x 54 , x 55 , y5) b2 = b(x 41 , x 42 , x 43 , x 44 , x 45 , y4, x 51 , x 52 , x 53 , x 54 , x 55 , y5)

[0147] A3 = A(x 61 , x 62 , x 63 , x 64 , x 65 , y6, x 71 , x 72 , x 73 , x 74 , x 75 , y7, x 81 , x 82 , x 83 , x 84 , x 85 , y8) b3 = b(x 61 , x 62 , x 63 , x 64 , x 65 , y6, x 71 , x 72 , x 73 , x 74 , x 75 , y7, x 81 , x 82 , x 83 , x 84 , x 85 , y8)

[0148] As described above, the information processing apparatus 10 according to the present embodiment may calculate an intermediate result without using information for specifying a main element.

[0149] Also, according to the above calculation, the larger the number of hierarchical data belonging to the cluster Ck, the more A k and b k it becomes difficult to restore the original feature amount x ij and the label y i from them.

[0150] From these facts, according to the generation of the intermediate result according to the present embodiment, it becomes possible to effectively enhance the privacy protection performance.

[0151] Subsequently, the information related to the update of the inference model transmitted from the information processing server 20 to the information processing apparatus 10 will be described in more detail.

[0152] Here, it is assumed that inference models M1 to M3 corresponding to clusters C1 to C3 are generated, respectively.

[0153] In this case, the information processing server 20 may calculate the information w1 to w3 related to the update of each of the inference models M1 to M3, for example, as follows.

[0154] w1=(w 11 , w 12 , w 13,…,w 1n ) w2 = (w 21 , w 22 , w 23 ,…, w 2n ) w3 = (w 31 , w 32 , w 33 ,…, w 3n )

[0155] According to the above calculations, the larger the number n of information processing devices that have calculated the intermediate results, the more difficult it is to restore the original feature quantity x k and the label y ij , and it becomes possible to effectively enhance the privacy protection performance. i

[0156] On the other hand, the information processing device 10 can perform more accurate inference by receiving the information w1~w3 related to the update of each of the inference models M1~M3.

[0157] For example, when inferring the label y9 of the feature quantity x9 belonging to the cluster C3, the information processing device 10 may calculate f(w3, x9).

[0158] Thus, the learning unit 120 of the information processing device 10 according to the present embodiment can infer a label based on a feature quantity and an inference model.

[0159] <<1.6. Application Examples>> Next, specific examples will be given and described regarding the application of the system 1 according to the present embodiment.

[0160] For example, the main elements according to the present embodiment may be various devices that communicate with the information processing device 10.

[0161] In this case, the information processing device 10 may use various logs collected along with the communication with the device as hierarchical data.

[0162] In the following, an example will be described in which the main element according to this embodiment is an access point that communicates with the information processing apparatus 10.

[0163] FIG. 8 is a diagram for explaining the operation of the system 1 when the main element according to this embodiment is the access point 40.

[0164] Note that in FIG. 8, the operation when the information processing apparatus 10 is a smartphone is illustrated.

[0165] In this example, the information processing apparatus 10 collectively holds a communication log L40 for each access point 40.

[0166] The communication log L40 is used as hierarchical data associated with information identifying the access point 40.

[0167] At this time, each communication log L40 includes feature amounts x1 to x n and a label y.

[0168] Feature amounts x1 to x n may include, for example, received radio wave intensity, CCA (Clear Channel Assessment) busy time, and the like.

[0169] Also, as the label y, an index representing the communication quality related to the access point 40 may be used.

[0170] Each of the information processing apparatuses 10 clusters the communication log L40 including the feature amounts x1 to x n and the label y using a plurality of inference models distributed from the information processing server 20.

[0171] For example, in the case of an example shown in FIG. 8, the information processing apparatus 10a classifies the communication log L40a for three records corresponding to the access point 40a and the communication log L40b for two records corresponding to the access point 40b into the cluster C1.

[0172] Further, the information processing apparatus 10a classifies the communication log L40c for two records corresponding to the access point 40c into the cluster C2.

[0173] Similarly, the information processing apparatus 10b classifies the communication log L40a for two records corresponding to the access point 40a into the cluster C1, and classifies the communication log L40c for two records corresponding to the access point 40c into the cluster C2.

[0174] In this way, the information processing apparatus 10 according to the present embodiment may perform clustering so that the hierarchical data related to the same main element is classified into the same cluster.

[0175] According to this, learning of features according to a pair of a predetermined type of access point 40, a predetermined type of information processing apparatus 10 and a predetermined type of access point 40, and more accurate inference are realized.

[0176] Also, at the same time, the information processing apparatus 10 may perform clustering so that the hierarchical data related to different main elements is classified into the same cluster.

[0177] According to this, the number of clusters can be suppressed, and the efficiency of learning is realized.

[0178] After performing the above-described clustering, each of the information processing apparatuses 10 according to the present embodiment transmits the intermediate result calculated as described above to the information processing server 20 and receives information related to the update of the inference model.

[0179] Thereafter, each of the information processing apparatuses 10 according to the present embodiment infers the label y using the collected feature amount x and the inference model.

[0180] For example, the information processing apparatus 10 can infer the communication quality when using a certain access point 40 from the received radio wave intensity from the access point 40 using the inference model.

[0181] Also, in this case, the information processing apparatus 10 may perform control such as connecting to the access point 40 with the highest inferred communication quality among the plurality of access points 40, which infers that the communication quality satisfies a predetermined condition, or connecting to the access point 40 with the highest inferred communication quality.

[0182] As described above, the operation in the case where the main element according to the present embodiment is a device that communicates with the information processing apparatus 10 has been described.

[0183] Next, an example in the case where the main element according to the present embodiment is a product category will be described.

[0184] The main element according to the present embodiment does not necessarily have to be a device.

[0185] FIG. 9 is a diagram for explaining the operation of the system 1 in the case where the main element according to the present embodiment is the product category GC.

[0186] Note that in FIG. 9, the operation in the case where the information processing apparatus 10 is a game device and the product is a game is illustrated.

[0187] In this example, the information processing apparatus 10 collectively holds the purchase log Lgc for each game category GC.

[0188] The purchase log Lgc is used as hierarchical data associated with information identifying the game category GC.

[0189] At this time, each purchase log Lgc includes feature amounts x1 to x n and a label y.

[0190] For the feature amounts x1 to x n For example, a game maker, a sales ranking, etc. may be adopted.

[0191] Also, as the label y, an index related to the purchase of a game belonging to the category GC (for example, whether it has been purchased or reserved) may be used.

[0192] Each of the information processing apparatuses 10 clusters the purchase log Lgc including the feature amounts x1 to x and the label y as described above using a plurality of inference models distributed from the information processing server 20. n and the label y.

[0193] For example, in the case of an example shown in FIG. 9, the information processing apparatus 10a classifies the purchase log Lgc1 for 3 records corresponding to the game category GC1 and the purchase log Lgc2 for 2 records corresponding to the game category GC2 into the cluster C1.

[0194] In addition, the information processing apparatus 10a classifies the purchase log Lgc3 for 2 records corresponding to the game category GC3 into the cluster C2.

[0195] Similarly, the information processing apparatus 10a classifies the purchase log Lgc1 for 2 records corresponding to the game category GC1 into the cluster C1, and classifies the purchase log Lgc3 for 2 records corresponding to the game category GC3 into the cluster C3.

[0196] Each of the information processing apparatuses 10 according to the present embodiment transmits the intermediate result calculated as described above to the information processing server 20 after performing the clustering as described above, and receives information related to the update of the inference model.

[0197] After that, each of the information processing apparatuses 10 according to the present embodiment infers the label y using the collected feature amount x and the inference model.

[0198] For example, the information processing apparatus 10 can infer the possibility that a user purchases a certain game using the inference model.

[0199] In this case, the information processing apparatus 10 may perform control such as explicitly presenting to the user a game whose purchase possibility exceeds a threshold value, or arranging it in an online shop so as to be easily noticeable to the user.

[0200] Next, an example in the case where the main element according to this embodiment is a person will be described.

[0201] In this case, the label y may be, for example, an index representing the physical or mental state of a person.

[0202] FIG. 10 is a diagram for explaining the operation of the system 1 when the main element according to this embodiment is a person and the label y is an index related to the health state of the person.

[0203] Note that in FIG. 10, the operation when the information processing apparatus 10 is installed in a medical institution is illustrated.

[0204] In the case of this example, the information processing apparatus 10 collectively holds the examination logs Lpe for each person P.

[0205] The feature amounts x1 to x included in the examination log Lpe n include, for example, various examination results such as blood pressure and heart rate, and symptoms.

[0206] Also, the label y may be a diagnosis result by a doctor.

[0207] Each of the information processing apparatuses 10 uses a plurality of inference models distributed from the information processing server 20 to cluster the examination log Lpe including the feature amounts x1 to x n and the label y.

[0208] For example, in the case of an example shown in FIG. 10, the information processing apparatus 10a classifies the examination log Lpe1 for 3 records corresponding to the person P1 and the examination log Lpe2 for 2 records corresponding to the person P2 into the cluster C1.

[0209] Also, the information processing apparatus 10a classifies the examination log Lpe3 for 2 records corresponding to the person P3 into the cluster C2.

[0210] Similarly, the information processing device 10a classifies the medical examination logs Lpe4 for two records corresponding to the person P4 into the cluster C1, and classifies the medical examination logs Lpe5 for two records corresponding to the person P5 into the cluster C3.

[0211] After performing the above clustering, each of the information processing devices 10 according to the present embodiment transmits the intermediate result calculated as described above to the information processing server 20 and receives information related to the update of the inference model.

[0212] After that, each of the information processing devices 10 according to the present embodiment infers the label y using the collected feature amount x and the inference model.

[0213] For example, the information processing device 10 can infer the health status of a person based on new examination results related to the person using the inference model.

[0214] According to this, it becomes possible to tentatively determine the health status of a person without performing an actual diagnosis by a doctor.

[0215] Next, the case where the label y is an index representing the emotion of a person will be described.

[0216] The index representing the emotion of the person includes, for example, the user's expression.

[0217] FIG. 11 is a diagram for explaining the operation of the system 1 when the main element according to the present embodiment is a person and the label y is the expression of the person.

[0218] In FIG. 11, the operation when the information processing device 10 is a robot that communicates with the user is illustrated.

[0219] In this example, the information processing device 10 collectively holds the shooting logs Lpp for each person.

[0220] Feature amounts x1 to x included in the shooting log Lpp nExamples include the photographed image, the positions of the respective parts on the face, the sizes of the respective parts, and the like.

[0221] Also, the label y may be various inferred expressions.

[0222] Each of the information processing apparatuses 10 clusters the shooting log Lpp including the feature amounts x1 to x and the label y as described above using a plurality of inference models distributed from the information processing server 20. n And the label y.

[0223] For example, in the case of an example shown in FIG. 11, the information processing apparatus 10a classifies the shooting log Lpp1 for three records corresponding to the person P1 and the shooting log Lpp2 for two records corresponding to the person P2 into the cluster C1.

[0224] Also, the information processing apparatus 10a classifies the shooting log Lpp3 for three records corresponding to the person P3 into the cluster C2.

[0225] Similarly, the information processing apparatus 10a classifies the shooting log Lpp4 for three records corresponding to the person P4 into the cluster C1, and classifies the shooting log Lpp5 for three records corresponding to the person P5 into the cluster C3.

[0226] After performing the clustering as described above, each of the information processing apparatuses 10 according to the present embodiment transmits the intermediate result calculated as described above to the information processing server 20, and receives information related to the update of the inference model.

[0227] Thereafter, each of the information processing apparatuses 10 according to the present embodiment infers the label y using the collected feature amount x and the inference model.

[0228] For example, the information processing apparatus 10 can infer the expression of a person from an image having the person as a subject using the inference model.

[0229] In addition, each of the information processing apparatuses 10 according to the present embodiment may perform control such as changing the behavior toward the user according to the inferred facial expression.

[0230] <2. Hardware Configuration Example> Next, a hardware configuration example common to the information processing apparatus 10 and the information processing server 20 according to an embodiment of the present disclosure will be described.

[0231] FIG. 12 is a block diagram showing a hardware configuration example of an information processing apparatus 90 according to an embodiment of the present disclosure.

[0232] The information processing apparatus 90 may be an apparatus having a hardware configuration equivalent to that of the information processing apparatus 10 and the information processing server 20.

[0233] As shown in FIG. 19, the information processing apparatus 90 includes, for example, a processor 871, a ROM 872, a RAM 873, a host bus 874, a bridge 875, an external bus 876, an interface 877, an input device 878, an output device 879, a storage 880, a drive 881, a connection port 882, and a communication device 883. Note that the hardware configuration shown here is an example, and some of the components may be omitted. Further, the information processing apparatus 90 may further include components other than those shown here.

[0234] (Processor 871) The processor 871 functions as, for example, an arithmetic processing unit or a control unit, and controls the overall operation or a part of the operation of each component based on various programs recorded in the ROM 872, the RAM 873, the storage 880, or the removable storage medium 901.

[0235] (ROM 872, RAM 873) The ROM 872 is a means for storing programs read by the processor 871, data used for arithmetic operations, and the like. In the RAM 873, for example, programs read by the processor 871 and various parameters that change as appropriate when executing the programs are temporarily or permanently stored.

[0236] (Host bus 874, bridge 875, external bus 876, interface 877) The processor 871, ROM 872, and RAM 873 are interconnected via a host bus 874 capable of high-speed data transmission, for example. On the other hand, the host bus 874 is connected to an external bus 876 with a relatively low data transmission speed via a bridge 875, for example. Further, the external bus 876 is connected to various components via an interface 877.

[0237] (Input device 878) For the input device 878, for example, a mouse, keyboard, touch panel, button, switch, lever, etc. are used. Further, as the input device 878, a remote controller (hereinafter referred to as a remote control) capable of transmitting a control signal using infrared rays or other radio waves may also be used. In addition, the input device 878 includes a voice input device such as a microphone.

[0238] (Output device 879) The output device 879 is a device capable of notifying the user visually or auditorily of the acquired information, such as a display device such as a CRT (Cathode Ray Tube), LCD, or organic EL, an audio output device such as a speaker or headphones, a printer, a mobile phone, or a facsimile. Further, the output device 879 according to the present disclosure includes various vibration devices capable of outputting a tactile stimulus.

[0239] (Storage 880) The storage 880 is a device for storing various data. As the storage 880, for example, a magnetic storage device such as a hard disk drive (HDD), a semiconductor storage device, an optical storage device, or a magneto-optical storage device is used.

[0240] (Drive 881) The drive 881 is a device that reads information recorded on a removable storage medium 901 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, or writes information to the removable storage medium 901.

[0241] (Removable storage medium 901) The removable storage medium 901 is, for example, a DVD medium, a Blu-ray (registered trademark) medium, an HD DVD medium, various semiconductor memory media, etc. Of course, the removable storage medium 901 may be, for example, an IC card equipped with a non-contact IC chip, or an electronic device, etc.

[0242] (Connection port 882) The connection port 882 is a port for connecting an external connection device 902 such as, for example, a USB (Universal Serial Bus) port, an IEEE1394 port, a SCSI (Small Computer System Interface), an RS-232C port, or an optical audio terminal.

[0243] (External connection device 902) The external connection device 902 is, for example, a printer, a portable music player, a digital camera, a digital video camera, or an IC recorder, etc.

[0244] (Communication device 883) The communication device 883 is a communication device for connecting to a network, and is, for example, a communication card for wired or wireless LAN, Bluetooth (registered trademark), or WUSB (Wireless USB), a router for optical communication, a router for ADSL (Asymmetric Digital Subscriber Line), or a modem for various communications, etc.

[0245] <3. Summary> As described above, the information processing apparatus 10 according to an embodiment of the present disclosure clusters hierarchical data based on a plurality of inference models distributed from the information processing server 20, and includes a learning unit 120 that performs learning using the corresponding inference model for each cluster.

[0246] Also, the information processing apparatus 10 according to an embodiment of the present disclosure includes a communication unit 130 that transmits the intermediate results generated for each cluster in the above learning by the learning unit 120 to the information processing server 20.

[0247] According to the above configuration, it is possible to achieve both privacy protection and high inference accuracy.

[0248] As described above, the preferred embodiments of the present disclosure have been described in detail with reference to the accompanying drawings, but the technical scope of the present disclosure is not limited to such examples. It is obvious that those having ordinary knowledge in the technical field of the present disclosure can conceive of various modification examples or correction examples within the scope of the technical idea described in the claims, and it is naturally understood that these also belong to the technical scope of the present disclosure.

[0249] Also, each step related to the processing described in this specification does not necessarily need to be processed in time series in the order described in the flowchart or sequence diagram. For example, each step related to the processing of each device may be processed in an order different from the described order or may be processed in parallel.

[0250] In addition, a series of processes performed by each device described in this specification may be realized using any of software, hardware, and a combination of software and hardware. The programs that make up the software are, for example, provided inside or outside each device and are pre-stored in a non-transitory computer readable medium that can be read by a computer. Then, each program is read into the RAM when executed by a computer, for example, and is executed by various processors. The storage medium is, for example, a magnetic disk, an optical disk, a magneto-optical disk, a flash memory, or the like. Also, the above computer program may be distributed via a network, for example, without using a storage medium.

[0251] In addition, the effects described in this specification are merely illustrative or exemplary and not limiting. That is, the technology according to the present disclosure may exhibit other effects apparent to those skilled in the art from the description in this specification, together with or instead of the above effects.

[0252] Note that the following configurations also belong to the technical scope of the present disclosure. (1) A learning unit that clusters hierarchical data based on a plurality of inference models distributed from an information processing server and performs learning using the corresponding inference model for each cluster; A communication unit that transmits intermediate results generated for each cluster in the learning by the learning unit to the information processing server; comprising The hierarchical data includes information identifying main elements and logs collected or generated in association with the main elements. An information processing apparatus. (2) The learning unit clusters the hierarchical data related to different main elements based on a plurality of the inference models. The information processing apparatus according to (1) above. (3) The learning unit performs clustering so that the hierarchical data related to the same main element is classified into the same cluster. The information processing apparatus according to (1) or (2) above. (4) The learning unit performs clustering so that the hierarchical data related to different main elements is classified into the same cluster. The information processing apparatus according to any one of (1) to (3) above. (5) The log includes feature amounts and labels. The information processing apparatus according to any one of (1) to (4) above. (6) The intermediate result includes values calculated from the feature amounts and the labels. The information processing apparatus according to (5) above. (7) The learning unit infers the label based on the feature amounts and the inference model. The information processing apparatus according to (5) or (6) above. (8) The communication unit receives information related to the inference model updated based on the intermediate results received by the information processing server from a plurality of devices, and delivers it to the learning unit. The information processing apparatus according to any one of (1) to (7) above. (9) The main element includes a device that communicates with the communication unit. The information processing apparatus according to any one of (5) to (7) above. (10) The main element includes an access point that communicates with the communication unit. The information processing apparatus according to (9) above. (11) The label includes an index representing the communication quality related to the access point. The information processing apparatus according to (10) above. (12) The main element includes a category of a product. The information processing apparatus according to any one of (5) to (7) above. (13) The label includes an indicator related to the purchase of a product belonging to the category, The information processing apparatus according to (12) above. (14) The main element includes a person, The information processing apparatus according to any one of (5) to (7) above. (15) The label includes an indicator representing the physical or mental state of a person, The information processing apparatus according to (14) above. (16) The label includes an indicator representing the emotion of a person, The information processing apparatus according to (14) above. (17) The processor, Cluster hierarchical data based on a plurality of inference models distributed from an information processing server, and perform learning using the corresponding inference model for each cluster, Send the intermediate results generated for each cluster in the learning to the information processing server, Including, The hierarchical data includes information identifying a main element and logs collected or generated in association with the main element, Information processing method. (18) A computer, A learning unit that clusters hierarchical data based on a plurality of inference models distributed from an information processing server and performs learning using the corresponding inference model for each cluster, A communication unit that sends the intermediate results generated for each cluster in the learning by the learning unit to the information processing server, Comprising, The hierarchical data includes information identifying a main element and logs collected or generated in association with the main element, Information processing apparatus, A non-transitory computer-readable storage medium storing a program for causing a computer to function as the above. (19) A learning unit that generates a plurality of inference models respectively corresponding to a plurality of clusters, A communication unit that transmits information related to the plurality of inference models generated by the learning unit to a plurality of information processing apparatuses, and comprising, The communication unit receives, from the plurality of information processing apparatuses, intermediate results generated by learning based on hierarchical data clustered based on the plurality of inference models and the inference models corresponding to each cluster, The learning unit updates the plurality of inference models based on the plurality of intermediate results, The hierarchical data includes information for identifying main elements and logs collected or generated in association with the main elements, An information processing server. (20) A processor, generating a plurality of inference models respectively corresponding to a plurality of clusters, transmitting information related to the generated plurality of inference models to a plurality of information processing apparatuses, receiving, from the plurality of information processing apparatuses, intermediate results generated by learning based on hierarchical data clustered based on the plurality of inference models and the inference models corresponding to each cluster, updating the plurality of inference models based on the plurality of intermediate results, including, The hierarchical data includes information for identifying main elements and logs collected or generated in association with the main elements, An information processing method. (21) A computer, a learning unit that generates a plurality of inference models respectively corresponding to a plurality of clusters, a communication unit that transmits information related to the plurality of inference models generated by the learning unit to a plurality of information processing apparatuses, and comprising, The communication unit receives, from a plurality of the information processing apparatuses, intermediate results generated by learning based on hierarchical data clustered based on a plurality of the inference models and the inference models corresponding to each cluster. The learning unit updates a plurality of the inference models based on a plurality of the intermediate results. The hierarchical data includes information for identifying main elements and logs collected or generated in association with the main elements. An information processing server A non-transitory storage medium readable by a computer storing a program for causing the computer to function as such.

Explanation of Signs

[0253] 1 System 10 Information processing apparatus 110 Sensor unit 120 Learning unit 130 Communication unit 20 Information processing server 210 Learning unit 220 Communication unit 40 Access point

Claims

1. Generating a plurality of inference models corresponding to a plurality of clusters; Based on intermediate results generated by clustering hierarchical data obtained by each information processing device transmitted from a plurality of information processing devices according to a plurality of inference models and learning for each cluster, updating the plurality of inference models; Distributing the plurality of inference models to the plurality of information processing devices; including An information processing method executed by a server system.

2. The hierarchical data includes sensor data as one element. The information processing method according to Claim 1.

3. The sensor data is obtained by a sensor included in the information processing device. The information processing method according to Claim 2.

4. The hierarchical data includes information for specifying a main element and logs collected or generated in association with the main element. The information processing method according to Claim 1.

5. The hierarchical data includes logs collected along with communication. The information processing method according to Claim 1.

6. The hierarchical data includes information for specifying an access point through which the information processing device communicates and logs associated with the access point. The information processing method according to Claim 5.

7. The log includes at least one of a feature amount related to the radio wave intensity of communication or a label representing communication quality. The information processing method according to Claim 5.

8. Further including distributing information related to updating of the inference model to the plurality of information processing devices. including The information processing method according to Claim 1.

9. Receiving intermediate results obtained using an inference model updated using information related to updating of the inference model distributed from the server system from the plurality of information processing devices. including The information processing method according to Claim 1.

10. The main element includes a device that communicates with the information processing device. The information processing method according to Claim 4.

11. A learning unit that generates a plurality of inference models corresponding to a plurality of clusters; A communication unit that distributes the plurality of inference models to the plurality of information processing devices; comprising The learning unit updates the plurality of inference models based on intermediate results generated by clustering hierarchical data obtained by each information processing device transmitted from the plurality of information processing devices according to the plurality of inference models and learning for each cluster. A server system.

12. The hierarchical data includes sensor data as one element. The server system according to claim 11.

13. The sensor data is obtained by a sensor included in the information processing apparatus. The server system according to claim 12.

14. The hierarchical data includes information identifying a main element and logs collected or generated in association with the main element. The server system according to claim 11.

15. The hierarchical data includes logs collected along with communication. The server system according to claim 11.

16. The hierarchical data includes information identifying an access point through which the information processing apparatus performs communication and logs associated with the access point. The server system according to claim 15.

17. The log includes at least one of a feature quantity related to the radio wave intensity of communication or a label representing communication quality. The server system according to claim 15.

18. The communication unit receives intermediate results obtained using an updated inference model updated using information related to updating of the inference model distributed from the server system, from a plurality of information processing apparatuses. The server system according to claim 11.

19. The main element includes an apparatus that communicates with the information processing apparatus. The server system according to claim 14.

Images