WO2017056168A1

WO2017056168A1 - Information processing system and information processing method

Info

Publication number: WO2017056168A1
Application number: PCT/JP2015/077394
Authority: WO
Inventors: 純一宮越; 泰幸工藤
Original assignee: 株式会社日立製作所
Priority date: 2015-09-28
Filing date: 2015-09-28
Publication date: 2017-04-06
Also published as: JPWO2017056168A1; JP6363305B2

Abstract

This information processing system uses, as a model, a graph structure that is composed of both a plurality of vertices, each of which is associated with an event to be analyzed, and edges, each of which connects vertices and indicates a relationship between the events associated with these vertices, wherein a plurality of computers, each corresponding to a respective one of the plurality of vertices, are connected to one another in the same manner as the manner in which the vertices are connected by the edges, so as to be able to provide data to and receive data from one another, and wherein computers corresponding to adjacent vertices that are connected to each other by edges calculate a transition probability for each identifier held by the computer, in accordance with a predetermined algorithm based on the numbers of identifiers held by the respective computers, and update the number of identifiers held by the computer, on the basis of the calculation result.

Description

Information processing system and information processing method

The present invention relates to an information processing system and an information processing method for executing graph processing.

In order to efficiently design and operate social infrastructure and cities, etc., technology that processes data distributed in the real world and cyber space, analyzes and predicts the state of social infrastructure, etc., and controls elements that make up society is drawing attention. Yes.

The above-mentioned distributed data is composed of environmental sensing data such as temperature and humidity, log data related to machines such as automobiles, and log data related to humans and organizations such as mail and SNS. In addition, the processing content of such distributed data is optimally arranged with clustering processing that classifies the data and adds labels and indexes, machine learning processing, and elements (people, things, information, etc.) that constitute society. Control processing. The processing results regarding the distributed data obtained by these processes are expanded to distributed users and control targets. The user or the control target object determines, for example, the moving means and the moving direction and the control parameter according to the processing result.

The following technologies have been proposed as such technologies. In other words, physically distributed sensing data is aggregated and processed in a computer system via communication means such as the Internet, and the results of this processing are deployed to control objects to analyze and predict social infrastructure. The technique to perform (refer patent document 1) etc.

US Patent Application Publication No. 2013/0151536

However, in the conventional technique (Personalized Pagerank Algorithm) as described above, synchronization between the computation subjects is required when executing parallel computation, and processing cannot be performed unless the distributed data is aggregated in the parallel computer. Furthermore, the calculation result obtained by the calculation needs to be developed for each distributed control target. Therefore, very large-scale data that is difficult to collect in one place, data that takes time for aggregation and expansion, and is updated every moment cannot be processed by the conventional technology.

Therefore, an object of the present invention is to provide a technique that enables efficient calculation of data that cannot be collected in one place on a large scale or that is updated every moment.

The information processing system of the present invention that solves the above problem is a model of a graph structure that is composed of a plurality of vertices corresponding to an analysis target event and edges connecting the corresponding vertices according to the relationship between the corresponding events. A plurality of computers corresponding to the vertices and connected to each other so as to be able to exchange data, and an attribute representing one or more states with respect to the event of the vertices. And a display device that displays the number of identifiers held for each vertex in the storage device as a spatial distribution map based on the distribution of the vertices. Transition of identifiers between computers according to a predetermined algorithm based on the number of identifiers held between the computers connected by edges and corresponding to the adjacent vertices The rate is calculated, the number of identifiers held by each computer is updated according to the calculation result, and the update amount is determined according to a predetermined algorithm based on the feature amount indicating the amount of change in the number of identifiers due to the update. The number of identifiers is updated based on the update amount.

According to the present invention, it is possible to efficiently calculate large-scale data that cannot be collected in one place or data that is updated every moment.

FIG. 3 is a diagram illustrating an example of a graph structure model to be analyzed in the first embodiment. It is a figure which shows an example of the number of information elements on each vertex in Example 1. FIG. 1 is a block diagram of a network configuration including an information processing system in Embodiment 1. FIG. 3 is a flowchart illustrating an example of processing performed in information processing according to the first exemplary embodiment. 3 is a flowchart illustrating an example of reception processing according to the first exemplary embodiment. 6 is a flowchart illustrating an example of transmission processing according to the first exemplary embodiment. 6 is a flowchart illustrating an example of acquisition processing according to the first exemplary embodiment. It is explanatory drawing which shows the outline of the calculation order in Example 1. FIG. It is explanatory drawing which shows the outline of the calculation order in Example 1. FIG. It is a figure which shows the relationship between the calculation model in Example 1, and the number of loops. FIG. 10 is a block diagram illustrating an example of a configuration of a computer in an information processing system according to a second embodiment. It is a figure which shows graph structure data in case there exists a weight in the edge of the calculation model in Example 2. FIG. It is a figure which shows the relationship between the edge weight in Example 2, and delay time. It is a figure which shows an example of the transition time of the information element in Example 2. FIG. It is a figure which shows an example of the acquisition concept of the graph structure data according to the real social activity in Example 3. It is a figure which shows an example of the acquisition concept of the graph structure data according to the real social activity in Example 3. FIG. 10 is a block diagram illustrating an example of a configuration of an information processing system according to a third embodiment. 10 is a flowchart illustrating an example of a processing procedure in the information processing method according to the third embodiment. 12 is a flowchart illustrating an example of reception processing in the information processing method according to the third embodiment. 10 is a flowchart illustrating an example of transmission processing in the information processing method according to the third embodiment. 12 is a flowchart illustrating an example of a calculation result acquisition process in the information processing method according to the third embodiment. FIG. 10 is a block diagram illustrating an example of a configuration of an information processing system according to a fourth embodiment. FIG. 10 is a diagram illustrating an example of an access chart to a process data block according to a fourth embodiment. FIG. 10 is a diagram illustrating an example of a process relationship in the fourth embodiment. FIG. 10 is a diagram illustrating an outline of an information processing system according to a fourth embodiment. FIG. 10 is a diagram illustrating an example of an access chart of a process in the fourth embodiment. FIG. 20 is a diagram illustrating an example of processing to an information child storage area corresponding to a data block according to a fourth embodiment. It is a figure which shows the number of the information elements stored in the storage area of each information element in Example 4, and the relationship of time. It is a figure which shows the structure of the warehouse of the process target in Example 5. FIG. It is a figure which shows the concept of the pick-up operation | work in Example 5. FIG. It is a block diagram which shows the example of the mobile terminal which the computer installed in the shelf in Example 5 and an operator hold | maintain. FIG. 10 is a block diagram illustrating an example of a hardware configuration of a computer and a mobile terminal according to a fifth embodiment. It is a figure which shows the concept of the information processing in Example 5. FIG. FIG. 10 is a diagram illustrating a relationship between a computer installed on each shelf and a destination in Example 5. 14 is a flowchart illustrating an example of processing performed by a computer according to a fifth embodiment. 12 is a flowchart regarding calculation result acquisition processing according to the fifth embodiment. It is a figure which shows the example of an effect in Example 5. FIG. It is a figure which shows the concept in Example 6. FIG. It is a figure which shows the exchange example of the information element in Example 6. FIG. It is a figure which shows the concept which specifies the affiliation community of each user from the information child number of each vertex in Example 6. FIG. It is a figure which shows the concept in Example 7. FIG. It is a figure which shows the example of the transition probability table 1 in Example 8. FIG. It is a figure which shows the example of the transition probability table 2 in Example 8. FIG. It is a figure which shows the example of the transition probability table 3 in Example 8. FIG. It is a figure which shows the change of the information child total number in Example 9, and the change of distribution of the information child of the whole information processing system. 22 is a flowchart illustrating an example of processing performed by a computer according to a ninth embodiment. It is a flowchart which shows an example of the information element number control process in Example 9. 22 is a flowchart illustrating an example of information element splitting processing according to the ninth embodiment. 22 is a flowchart illustrating an example of information element fusion processing according to the ninth embodiment. It is a figure which shows the relationship between the power spectrum of the change of the number of information elements in Example 9, the mutual information amount of the time direction calculated on each vertex, and the number of information elements on each vertex. It is a figure which shows the concept in Example 9. FIG. 22 is a flowchart illustrating an example of processing of a computer in Example 10. 22 is a flowchart illustrating an example of processing performed by a computer according to an eleventh embodiment. FIG. 20 is a diagram illustrating an example of a determination algorithm according to the ninth embodiment.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. First, the concept of the technical idea of the information processing method in the present embodiment will be described based on the problems in the prior art. In the prior art, data analysis cannot be executed unless all the target data is collected in one place. When large-scale data, so-called big data, is the target of analysis, each piece of data scattered over a very wide area must be instant. Therefore, it is necessary to collect and analyze it efficiently without omission, and to process this analysis result in response to each object of data origin, which is particularly difficult to apply to frequently updated data.

Therefore, in the information processing method of the present embodiment, autonomous distributed data analysis that performs analysis for each distributed data is performed in order to solve such a problem. In this autonomous decentralized data analysis, each element (specifically, a computer) that manages the distributed data performs a predetermined calculation on the data of other elements adjacent to its own data, and performs the desired calculation for the entire element. It is a technique to do.

The phenomenon corresponding to the concept of autonomous decentralization described above is a phenomenon that is often seen in nature, for example, a reaction diffusion model in the field of biology is well known. Among the reaction-diffusion models, for example, in a model relating to the formation of zebra stripes, the zebra stripes are caused by the diffusion of proteins in each cell individually. In the present embodiment, such an autonomous distributed reaction diffusion model is replaced with a situation in which diffusion of objects and information is performed individually in each place and each element, and a technique applied to data analysis will be described.

First, the conceptual processing of the information processing system of the present invention will be described. FIG. 1 is a conceptual diagram of a graph structure model (hereinafter, calculation model 1) to be analyzed in the information processing system. The calculation model 10 performs calculation by diffusing information elements (Information particles) that are information units with respect to a graph structure including vertices 110 to 114 and edges 121 connecting these vertices. This is a calculation model with the number of information elements on .about.114 as a solution. In the first embodiment, the calculation target is a classification problem.

Here, an information element (identifier) that is a unit of information is defined. The information element is data having a state (attribute) variable. In the first embodiment, the number of states = 2 (that is, the data capacity is 1 bit), and the state u and the state v, respectively. That is, one vertex includes a plurality of information elements, and as described later, one vertex is associated with a state (attribute) according to the total number of information elements.

FIG. 1 illustrates an information element 132 in a state u and an information element 131 in a state v. Each information element diffuses along the side 121 (information element diffusion 140). The calculation result can be obtained from the number of information elements on each vertex. From the information child number table 150 (see FIG. 2) on each vertex, for example, the vertex A (vertex 110) has the information child number (state u) = 0 and the information child number (state v) = 2, and the information child number (state v) is maximized. The state of the maximum information element is made to correspond to the classification result. That is, if the state u is the classification result A and the state v is the classification result B, the vertex A becomes the classification result B. If a computer described later calculates in the same way at each vertex, all the vertices can be classified as either A or B. FIG. 2 is a diagram illustrating an example of the number of information elements on each vertex.

Subsequently, a configuration example of the information processing system 100 according to the first embodiment will be described. FIG. 3 is a block diagram illustrating an example of the configuration of the information processing system 100 according to the first embodiment. The information processing system 100 shown in FIG. 3 includes one or more computers 220-1 to 220-4, and these computers are connected by a network 1. Hereinafter, the computer 220 will be referred to unless the computer is particularly distinguished.

The computer 220 includes a CPU 221, a main storage device 222 configured by a volatile storage device such as a RAM, a storage 223 configured by an appropriate nonvolatile storage device such as a hard disk drive, an input / output device 224 such as a keyboard, a mouse, and a display, A network I / F 225 is included. In the computer 220 having such a configuration, the CPU 221 executes a program 226 stored in the main storage device 222 to implement necessary functions, performs overall control of the computer itself, and performs various determinations, computations, and control processes. Therefore, the function corresponding to the information processing method of the first embodiment corresponds to a function that is implemented by the execution of the program 226 by the computer 220 described above.

Next, the actual procedure of the information processing method according to the first embodiment will be described with reference to the drawings while reiterating the calculation model of FIG. 3 described above. FIG. 4 is a flowchart illustrating an example of a procedure in the information processing method according to the first embodiment. In this case, first, in step 311, each of the computers 220-1 to 220-4 stores data relating to the corresponding vertex in the above-described calculation model 10 that is divided for each computer and stored in its own storage 223 or the like. . For example, data stored in the storage 223 by the computer 220-1 is stored in a predetermined data area 230-1 in the storage 223, as exemplified in FIG. Here, the same symbol is attached | subjected supposing the calculation model 10 similar to FIG.

One graph structure data is distributed and stored in the data areas 230-1 to 230-4 of the plurality of storages 223, and the data areas 230-1 to 230-4 are connected via the network 1.

In the data area 230-1, the computer 220-1 stores information on the vertex 110, which is the vertex A, the vertex 111, which is the vertex B, and the connection destination of each vertex in the storage 223. That is, information on a vertex 113 that is a vertex D to which a vertex 110 that is a vertex A is connected, information about a vertex 110 that is a vertex A to which a vertex 111 that is a vertex B is connected, a vertex 112 that is a vertex C, and a vertex 113 that is a vertex D Is stored in the data area 230-1 of the storage 223.

Subsequently, in step 312, each computer 220 calculates in advance the number of information elements assigned to each vertex for all vertices included in the calculation model 10 acquired from each data area 230 in step 311 described above. Initialize with a fixed number. For example, when the predetermined number of the vertex 110 that is the vertex A is the state u = 0 and the state v = 2, the computer 220 determines the number of information elements assigned to the vertex A among the data stored in the storage 223. Are the number of information elements (state u) = 0 and the number of information elements (state v) = 2.

After step 312 described above, each computer 220 executes a certain number of loop processes (steps 313-1 to 313-2). Each computer 220 executes loop processing (steps 314-1 to 314-2) for all vertices for which data has been acquired in step 311 described above in the loop processing, and receives information elements (step 315). The transmission process (step 316) is executed. After the above two loop processes (steps 313-1 to 313-2 and steps 314-1 to 314-2) are completed, each computer 220 executes a calculation result acquisition process (step 317) to execute this flowchart. Exit.

Subsequently, the reception process (step 315), the transmission process (step 316), and the calculation result acquisition process (step 317) will be described in the flowchart described above. FIG. 5 is a flowchart showing an example of the reception process (step 315). Each computer 220 determines whether or not an information element has been received from another vertex after the start of the flowchart (step 411). If this determination is true (step 411: Y), the computer 220 updates the number of information elements storing data in the storage 223 (step 412). If the above determination is false (step 411: N), This flowchart is terminated.

FIG. 6 is a flowchart showing an example of the transmission process (step 316). Each computer 220 acquires the number of information elements on vertices connected by edges (referred to as adjacent vertices) from the vertices selected in the loop processing (step 314) regarding the vertices after the start of the flowchart (step 314). Step 512). Here, the number of information elements acquired by the computer 220 may be the number of information elements acquired in the past.

Thereafter, the computer 220 executes a loop process (steps 513-1 to 513-2) for the information element on the vertex selected in step 314 described above. In the loop, the computer 220 calculates a transition probability for the information element selected in step 314 described above (step 514). A formula for calculating the transition probability is shown below as Formula (1).

Where Ndj is the degree of vertex j. The left side P indicates the transition probability, and Pu indicates the transition probability of the information element in the state u. U and v represent different states of the information element. For example, in the case of a two-state (1 bit) information element, u is in a state = 0, and v is in a state = 1. Nu and Nv represent the number of information elements. For example, Nu is the number of information elements in state u, and Nv is the number of information elements in state v. Also, Σ in the third term and the fourth term is calculated for adjacent vertices in the neighborhood range. Nui and Nvj are the number of information elements of each state of the adjacent vertex j in the neighborhood range. F and g are functions, and α and β are positive constants.

Note that the computer 220 may calculate the transition probability by prediction based on the update history of past information elements. The computer 220 in such a prediction method uses the result obtained by inputting the test pattern in the above formula (1) in advance as correct data, and uses the update history of past information elements and the number of information elements at the vertex as input data. It is assumed that the model learned by a predetermined neural network is used for calculating the transition probability. By using this model, the computer 220 can replace the number of information elements at adjacent vertices with the update history of past information elements.

Subsequent to step 514 described above, in step 515, the computer 220 compares the transition probability calculated in step 514 described above with a predetermined threshold. As a result, for example, when the threshold value is 0.5 and 0.5 ≦ transition probability ≦ 1 (step 515: Y), that is, the computer 220 that has determined that the transition probability is high advances the process to step 516. On the other hand, for example, when 0 ≦ transition probability <0.5 (step 515: N), the computer 220 determines that the transition probability is low, and proceeds to the end (step 520).

In step 516, the computer 220 selects one vertex from the adjacent vertices for which the number of information elements has been obtained in step 512. The selection method here may be random or round robin. Next, in step 517, the computer 220 transmits the data (state) of the information element from the network I / F 125 to the adjacent vertex (corresponding computer) selected in step 516 described above. Further, in step 518, the computer 220 updates the number of information elements at the vertex (executes the number of information elements −1 because the information element is transmitted).

Subsequently, FIG. 7 shows a detailed flowchart regarding the calculation result acquisition process (step 317) in the flowchart of FIG. In step 611, the computer 220 compares the number of information elements of each state of each vertex stored in the storage 223 and selects the state having the maximum number of information elements after starting the flowchart. For example, when the number of states is 2 (u and v), the number of information elements in the state u on a certain vertex is 1, and the number of information elements in the state v is 2, the computer 220 selects the state v.

Next, in step 612, the computer 220 displays the result corresponding to the state selected in step 611 described above on the display device or the input / output device 224 of the predetermined computer 220 on the network 1. For example, when the result corresponding to the state v is the classification result B, it can be seen that the vertex belongs to the classification B (the state u is the community A). Other vertices can be calculated similarly.

Here, as a feature of this calculation model, the calculation result does not depend on the calculation order of the information elements. 4 and 6, in the vertex loop (step 314-1) and the information child loop (513-1) on the vertex, the calculation order of the vertex and the information child is free, For example, the calculation result does not change when a certain vertex A is processed and then the vertex B is processed, and when the vertex B is processed and then the vertex A is processed. That is, the calculation order is free.

8A and 8B show an outline of the calculation order when the calculation model 10 and the information processing system 100 in FIG. 3 are assumed. 8A and 8B show two flowcharts having different calculation orders, which are described as processing order 1 in FIG. 8A and processing order 2 in FIG. 8B, respectively. Further, the movement of data (information element) due to transmission / reception between vertices shows only movement related to the vertex D.

Among these, in the processing order 1 of FIG. 8A, the vertex D is connected to the vertex A, the vertex B, the vertex E, and the edge, and thus the information element moves between the vertices. In calculation order 1, before vertex D reception processing, transmission processing with the same cycle as vertex D and transmission processing with the same cycle as vertex E are executed for vertex A. However, since the transmission process of the same period as that of the vertex D is not yet executed for the vertex B, in the reception process of the vertex D, the movement 710 of the same period of information from the vertex A and the previous period from the vertex B An information element movement 712 and an information element movement 711 of the same period from the vertex E occur.

On the other hand, in the processing order 2 of FIG. 8B, since the transmission processing of the vertex A, the vertex B, and the vertex E is not executed before the reception processing of the vertex D, the information element of the previous cycle from the vertex A is moved. 750, the previous period information child movement 750 from vertex B, and the previous period information child movement 750 from vertex E occur. From this, the timing at which the information element moves differs between processing order 1 and processing order 2.

However, in both processing order 1 and processing order 2, after the processing is repeated for a certain period of time, it converges to the same calculation result. From the above, this calculation model shows that in the configuration example shown in FIG. 3, the calculation result does not change even if each computer 220 performs the calculation independently during parallel calculation. Even if synchronization and other cooperation cannot be executed due to a problem of delay of communication means between computers, the processing can be executed. Therefore, according to the present invention, it is possible to solve a classification problem for a target problem in which data is distributed over a wide area.

A more specific example is shown in FIG. FIG. 9 is a diagram showing the relationship between the calculation model and the number of loops. The example shown in FIG. 9 is a calculation model having 4096 vertices. In FIG. 9, the edges between the vertices are not shown because the figure becomes complicated. In FIG. 9, the vertices are dense near the center of the circle, and the density of the vertices decreases toward the outer periphery. In addition, in a wavy gap having no vertices crossing the circle, the distance between the vertices near the outer periphery of the circle is larger than the gap near the center of the circle. For this reason, when the classification based on the distance threshold, which is one of the conventional statistical methods, is performed, the wavy gap cannot be recognized and cannot be correctly classified.

On the other hand, since the method of the present invention performs classification in an autonomous and distributed manner, a gap in a high-density region near the center and a low-density gap near the outer periphery can be correctly recognized. In the calculation model shown in FIG. 9, the number of states of the information element is 2, and in initialization of the number of information elements, the information group (state u) is stored in the vertex group 810-1 and the information field (state is stored in the vertex group 810-2). Assume that 4096 × 8 are assigned to v). Further, at each vertex in FIG. 9, the density (color tone density from white to gray to black) is changed depending on whether the number of information elements is large in the state u and the state v is large.

In the state 810 of the loop count t in FIG. 9, the information elements are only diffused to a part of the upper end and the lower end of the graph structure data, but in the state 820 of the loop count t + n, the information elements are spread throughout. . However, in the state of the above loop count t + n, the classification accuracy is lowered near the gap at the center of the graph structure data. On the other hand, in the state 830 of the loop count t + 2n, a beautiful classification is performed.

As described above, according to the first embodiment, it is possible to efficiently calculate distributed data that cannot be collected in one place on a large scale or data that is updated every moment.

Subsequently, a case where each side defined in the calculation model of the first embodiment includes a weighting coefficient will be described as a second embodiment. In the second embodiment, a delay process based on the weighting coefficient of each side is added to the transmission process shown in the first embodiment. Therefore, the computer 220 has a configuration including a delay unit 911 that executes the corresponding process. Other configurations are the same as those in the first embodiment.

FIG. 10 is a block diagram illustrating an example of the configuration of the computer 910 in the information processing system according to the second embodiment. The computer 910 shown here corresponds to each of the computers 220-1 to 220-4 shown in FIG. 3 of the first embodiment, and modules having the same function are given the same reference numerals. The computer 910 according to the second embodiment newly includes a delay unit 911 with respect to the computers 220-1 to 220-4.

Based on this, FIG. 11 shows graph structure data 920 when weights are added to the sides of the calculation model. The graph structure data 920 includes a vertex A950, a vertex B951, and a vertex C952, a weight 960 having a value of 1 for the side between the vertices AB, and a value of 10 for the side between the vertices AC. Assume that a weight 961 exists.

In the graph structure data 920 loaded with weights, when the information element moves from the vertex A 950 to the vertex B 951, the computer 910 delays the transition of the information element by the delay unit 911 according to the weight 960 described above. FIG. 12 shows an example in which a relationship table 940 between edge weights and delay times is defined in advance.

FIG. 13 is a diagram showing an example of the transition time of the information element. That is, the computer 910 controls the transition time of the information element. For example, as shown in FIG. 13, when the transition time 930 of the information element moves from the vertex A 950 to the vertex B 951, the weight = 1, so that the delay time t = 1 (970). Similarly, when moving from the vertex A950 to the vertex C952, since the weight = 10 (961), the delay time is t = 10 (971). The relationship between the weight and the delay time may be given in advance in the table 940 as shown in FIG. 12, or may be calculated by the computer 910 using a predetermined mathematical formula.

Next, in step 311 in the flowchart of FIG. 4 shown in the first embodiment, that is, an example of an information processing system having a function of automatically acquiring data in accordance with real-world activities at the time of acquiring graph structure data explain. In the concept of the third embodiment, the graph structure data, which is a calculation model, is not calculated on the information processing system, and real-world activities are used as they are.

FIG. 14 is a diagram illustrating an example of an acquisition concept of graph structure data according to real social activities in the third embodiment. In FIG. 14, for example, when a human conversation is assumed as a real-world activity, the graph structure data as a calculation model is obtained from a log of a person's conversation (who and who, how many times, how long, and a conversation). Structured.

For example, when a person A (1001) and a person B (1002) have a conversation in an activity (1000) in the real world, the computer 220 that obtains the recorded data has the graph structure data 1010 expressing the frequency and time of the conversation. A side 1013 is generated between the vertex A (1011) corresponding to the person A and the vertex B (1012) corresponding to the person B.

The result 1020 obtained by analyzing the generated graph structure data 1010 using an information element is shown. The attribute of each vertex can be calculated according to the number of information elements on each vertex, and the vertex A 1021 and the vertex B 1022 are attributes of information elements expressed by black squares. 1420 and 1060 in FIG. 15 show the same analysis results, although the calculation means leading to the results are different.

The computer 220 performs data structuring 1030 by the above-described processing. Of course, real-world activities are not limited to human conversation, but also activities between objects (for example, communication between machines such as robots, automobiles, traffic lights, etc.), activities between persons and objects, and objects through persons. Activities (indirect communication between places and facilities via people who visit multiple facilities and shelves), SNS user interaction (message communication, e-mail, etc.) in the virtual space good. In this case, the computer 220 structures the activity in the real world into graph structure data, and then analyzes it in the same manner as the processing exemplified in the first embodiment.

On the other hand, if an information element is attached to an activity in the real world, which is input data for structuring the graph structure data, calculation using the input data itself during the structuring of the graph structure data is possible. FIG. 15 is a diagram illustrating a state 1050 in which information elements are exchanged in association with activities in the real world. If the person A and person B described above can hold an information child and can update the information child during a conversation between the person A and the person B, the computer 220 executes a calculation 1070 using real-world activities to calculate the information child. As an analysis result 1060 as a result, the number of information elements held in the real world, that is, person A and person B is obtained. In Examples 3 to 6, specific configurations corresponding to this concept will be described.

The configuration of the information processing system in Example 3 will be described below. FIG. 16 is a block diagram illustrating an example of the configuration of the information processing system 100 according to the third embodiment. The information processing system 100 illustrated here includes a device group 3001 and a device 3020 held or mounted on the vertex 3010. In this embodiment, an information processing system for processing the real world is described as an example, the device group 3001 is a group of people in the real world, and each device is a smart device held by a person or the like. Of course, the device group is not limited to a person but may be a mobile object such as a car, a smartphone attached to a machine, a device such as an embedded computer and a program attached to data. The problem to be analyzed is a vertex group classification problem. When each vertex is a person, for example, it is applied to community detection of a certain group.

In this case, the device 3020 as a computer includes a CPU 3021, a main storage device 3022 for storing a program 3026, a storage 3023, an input / output device 3024, and a network I / F 3025. The device 3020 has a neighborhood range 3030 that is recognized as communicable. In the example of FIG. 16, in the device 3020 associated with a person or device, a circle having a radius with a predetermined value centered on the device 3020 is the neighborhood range 3030. Such a proximity range 3030 can be calculated from the physical distance between devices, but the reach of radio waves transmitted by the network I / F 3025, the frequency of communication between devices (that is, between vertices having devices), for example, mail The number of exchanges may be in a range above a certain threshold.

Further, the device 3020 can communicate within the vicinity range 3030 using a network I / F 3025 or the like. In the example of FIG. 16, the vertex 3010 has a function of communicating with another device 3011 existing in the vicinity range 3030. Since the devices existing in the neighborhood range 3030 correspond to the vertices connected at the sides shown in the first embodiment, the actual social activities themselves become the sides. That is, the graph structure data of the first embodiment is not necessary. In the third embodiment, the device 3020 is called a vertex.

Next, an example of the processing procedure of the information processing method in the third embodiment will be described. FIG. 17 is a flowchart illustrating a processing procedure in the information processing method according to the third embodiment. Here, the information element which is a unit of information is the same as that defined in the first embodiment.

In this case, after starting the flowchart, each device 3020 first initializes the number of information elements of the corresponding vertex 3010 (step 3111). The processing content of step 3111 is to initialize the number of information elements with a predetermined number in the device 3020. For example, if the predetermined number of vertices A is state u = 0 and state v = 2, the device 3020 sets the number of information elements assigned to the vertex A to the number of information elements (state u) = 0, information element The number (state v) = 2. After that, in step 3112, the device 3020 activates the reception processing process, and in step 3113 activates the transmission processing process.

Next, in step 3114, the device 3020 activates a calculation result acquisition process. The device 3020 may execute the reception process (step 3112), the transmission process (step 3113), and the calculation result acquisition process (step 3114) in parallel.

Hereinafter, the reception process (step 3112) among the above-described processes will be described. The main difference between the reception process in the third embodiment and the reception process in the first embodiment is the addition of a process after a lapse of a fixed time. FIG. 18 is a flowchart illustrating an example of reception processing.

The device 3020 determines whether or not a predetermined time has elapsed in step 3211 after the process starts. If the determination is true (step 3211: Y), the device 3020 moves the process to step 3212.

In step 3212, the device 3020 determines whether an information element has been received from another device. If the determination is true (step 3212: Y), the number of information elements of the device itself is updated (information element) in step 3213. Equation -1). The process ends with an end interrupt or the like. Further, the process of step 3211 may be a process triggered by communication instead of a lapse of a fixed time.

Subsequently, the above-described transmission process (step 3113) will be specifically described. The main difference between the transmission process in the third embodiment and the transmission process in the first embodiment is the addition of a process after a certain period of time.

FIG. 19 is a flowchart showing an example of transmission processing. The device 3020 determines whether or not a predetermined time has elapsed in step 3311 after the process starts. If this determination is true (step 3311: Y), the device 3020 moves the process to step 3312. In step 3312, the device 3020 communicates with the vertex existing in the neighborhood range 3030 described above, and acquires the number of information elements held by the vertex.

Subsequent steps 3313-1 to 3313-2 are loop processing for each information element of the own vertex. In the loop processing, in step 3314, the device 3020 calculates the transition probability of the self information element. This step is the same as the processing in step 514 of the first embodiment.

Thereafter, in step 3315, the device 3020 compares the transition probability calculated in step 3314 described above with a predetermined threshold value. As a result, if the transition probability> the threshold value is true (step 3315: Y), the process proceeds to step 3316. For example, when the threshold value is 0.5 and 0 ≦ transition probability <0.5, the device 3020 advances the process to step 3311. On the other hand, if 0.5 ≦ transition probability ≦ 1, the process advances to step 3316. .

In step 3316, the device 3020 selects one vertex from the vertices in the neighborhood range 330. This selection method may be random or order (round robin). Thereafter, in step 3317, the device 3020 transmits the data (status) of the self-identifier from the network I / F 3025 to the vertex (device) selected in step 3316 described above.

Further, in step 3318, the device 3020 updates the number of information elements at its own vertex (since the information element is transmitted, the number of information elements −1 is executed). Also, the parameters of these processes (for example, coefficients of transition probability calculation formulas, threshold values, selection methods, and the like) may be different at the vertices.

Next, the above-described calculation result acquisition process (step 3114) will be described in detail. FIG. 20 shows a flowchart of calculation result acquisition processing. In this case, the device 3020 determines whether there is a result acquisition request from the input / output device 3024 in step 3411 after the calculation result acquisition processing is started. If this determination is true (step 3411: Y), the device 3020 executes step 3412 and step 3413. This step is the same as step 611 and step 612 in the first embodiment, and a description thereof will be omitted.

By the above processing, the classification problem for the vertices can be solved as in the first embodiment. Since the vertex in the third embodiment is a device distributed in the real world, that is, the classification problem for the device distributed in the real world can be efficiently solved without generating graph structure data.

Next, a fourth embodiment is shown as an example of an information processing system corresponding to a calculation model in which each vertex in the above-described third embodiment is data and an edge between the vertices is an access continuity between the data. In the fourth embodiment, when a plurality of processes are processed on a plurality of computers, a method for efficiently arranging data necessary for each process on the computers is provided.

FIG. 21 is a block diagram illustrating an example of the configuration of the information processing system according to the fourth embodiment. In the information processing system 4000 illustrated in FIG. 21, the computer 120-1 and the computer 120-2 are connected via the network 1, the process 120 is processed by the computer 120-1, and the data area of the computer 120-1 Assume that data blocks 1, 2, and 3 are stored in 130-1.

It is also assumed that the process 2 is processed by the computer 120-2 and the data blocks 4, 5, and 6 are stored in the data area 130-2 of the computer 120-2. The data block described above stores data necessary for the

processes

1 and 2 described above. The data areas 130-1 and 130-2 are stored on a storage (not shown), and desired data is stored from the storage data area 130 at a timing required for calculation by a program executed by the computer 120. It is transferred to the main memory (not shown).

FIG. 22 is a diagram showing an example of an access chart 4010 to each data block of each

process

1 and 2. In the access chart 4010 illustrated in FIG. 22, data blocks adjacent in the time direction in the time section T are related. Specifically, when the process 1 accesses the data block 1 and subsequently accesses the data block 2, the relationship between the data block 1 and the data block 2 is incremented by “+1”.

FIG. 23 is a diagram showing an example of the relationship 4020 calculated by integrating the relationships of the processes. In the table 4021 in FIG. 23, “2” of the value (4211) of the data block 1 in the row and the data block 2 in the column is continuously transferred to the data block 2 after the access to the data block 1 in the time interval T described above. This indicates that the number of accesses is “2”. Further, when the relationship table 4021 in FIG. 23 is represented as a graph, a relationship graph 4023 is obtained. The graph 4023 does not represent the side of the number of times “0”. From this result, the data blocks 1, 2, and 6 are originally the data area of the processing computer of the process 1, that is, the data area (130-1), and the data blocks 3, 4, and 5 are the data area of the processing computer of the process 2, that is, the data This indicates that it is preferably stored in the area (130-2).

FIG. 24 is a schematic diagram of an information processing system 4100 according to the fourth embodiment. The information processing system 4100 is a detailed diagram of the configuration of the storage area of the information processing system 4000 of FIG. The information processing system 4100 has information child storage areas 1 to 6 (4101-1 to 4101-6) corresponding to the data blocks in the data areas 130-1 and 130-2. This storage area has a function of storing one or more information elements.

Subsequently, an information processing method according to the fourth embodiment will be described. FIG. 25 is a diagram showing an access chart 4200 for each process. In the fourth embodiment, when each data block is accessed, processing is performed on the information child storage area corresponding to each data block. FIG. 26 is a diagram showing an example of processing 4300 for this information storage area. Here, when the process 1 accesses the data block 1, the information child stored in the area 1 is acquired as the process 4201 for the storage area 1 of the information child. That is, the number of information elements is subtracted.

In this example, for the number of information elements = 10 before the process, in the process 4210, five information elements stored in the area 1 are acquired (number of information elements = 10−5 = 5). ). Thereafter, since the process 1 accesses the data block 2, the process 4202 is performed on the storage area 2 of the information element corresponding to the data block. In this process, the five information elements acquired in the pre-process 4201 are added to the area 2 (4211).

Further, in the process, an information element is acquired from the area (4212). By repeating such processing, the information element is circulated between the storage areas corresponding to the respective data blocks. Data blocks that are highly relevant (successfully accessed continuously) are classified into the same cluster according to the distribution of information elements. By regularly moving data blocks between data areas of computers according to the distribution of information elements, highly relevant data blocks can be collected in the same computer.

Next, a calculation example of the fourth embodiment will be shown. FIG. 27 is a diagram showing the relationship between the number of information elements stored in the storage areas 1 to 6 of each information element and time. In FIG. 27, the initial state of the number of information elements stored in the storage area of each information element is shown in a table 4310.

Here, in the fourth embodiment, the number of states of the information element is 2 (state u and state v), and the above time change and the number of information elements in the table 4310 are calculated by the equation of state u−state v. Further, in the initial state, areas 1 to 3 on the data area 130-1 are information element u = 10, information element v = 0 (information element u-information element v = 10), and the above-described area on the data area 130-2. 4 to 6 are information child u = 0 and information child v = 10 (information child u−information child v = −10). The number of information elements in each area is updated every time the data block is accessed, and changes with time as shown in FIG.

When attention is paid to the time Tp in FIG. 27, the magnitudes of the number of states of the information elements in the storage area 3 and the storage area 6 are reversed. Therefore, in the initial state, the clusters are the areas 1 to 3 and the areas 4 to 6. However, at the time Tf, the clusters are changed to the clusters of the

areas

1, 2, and 6 and the

areas

3, 4, and 5. This is the preferred data block arrangement described above in relation 4020 between data blocks in FIG.

Next, regarding the pick-up work of the luggage in the warehouse, an information processing method when dealing with the problem of rearranging the shelves in the warehouse so as to shorten the flow line distance of the worker who performs the work is shown as Example 5. . FIG. 28 exemplifies a configuration of a warehouse 5000 to be processed. The warehouse 5000 has a plurality of areas inside. The warehouse 5000 in FIG. 28 has four areas, area A (5010-1) to area D (5010-4).

In each of these areas A to D, a plurality of shelves 5011-A1 to 5011-D1 are arranged in the corresponding area. In the example of FIG. 28, the shelf 5011-A1 is illustrated as one of the shelves placed in the area A (5010-A), but it is assumed that a plurality of other shelves are arranged. Similarly to the area A, the areas B to D have a plurality of shelves 5011. Further, each shelf 5011 can hold a plurality of luggage 5012. In the example of FIG. 28, an example in which the packages 5012-A1-1 and 5012-A1-2 are arranged on the shelf A-1 (5011-A1) is illustrated. Similarly, a plurality of packages 5012 can be arranged on other shelves 5011.

Subsequently, the pick-up operation of the luggage 5012 arranged on each shelf 5011 will be described. FIG. 29 is a diagram showing the concept of the pickup work. In FIG. 29, an operator 5100 picks up a baggage 5012 placed on each shelf 5011 according to a predetermined baggage list 5110. In the fifth embodiment, the packages to be picked up are packages 5012-A1-1, 5012-B1-2, and 5012-D1-1.

The worker 5100 visits the shelves 5011-A1, 5011-B1, and 5011-D1 on which the respective packages 5012 are placed according to the list 5110. In that case, the movement route of the worker is like a movement route 5120. Here, the order of picking up the luggage 5012 is not defined.

Next, a method for reducing the movement distance of the worker 5100 by the information processing method of the fifth embodiment when there are a plurality of workers 5100 in the above-described pick-up work of the luggage 5012 will be described. FIG. 30 is a block diagram showing a computer installed on a shelf and a mobile terminal possessed by an operator 5100. In this embodiment, a computer 5210 is installed on each shelf 5011, and each worker 5100 has a mobile terminal 5220.

FIG. 31 is a block diagram showing a configuration example of the computer 5210 and the mobile terminal 5220.

The computer 5210 and the mobile terminal 5220 have the same configuration. The computer 5210 and the mobile terminal 5220 are a CPU 1021, a main storage device 1022 constituted by a volatile storage device such as a RAM, a storage 1023 constituted by an appropriate nonvolatile storage device such as a hard disk drive, an input / output such as a keyboard, a mouse, and a display. A device 1024 and a network I / F 1025 are included. The CPU 1021 provides a predetermined function by executing a program 1026 stored in the main storage device 1022, performs overall control of the computer itself, and performs various determinations, calculations, and control processes.

Subsequently, the concept of the information processing method of this embodiment will be described. FIG. 32 is a diagram illustrating the concept of information processing in the fifth embodiment. In FIG. 32, after state 1 (5301) in which the worker 5100 has acquired the luggage from the shelf B-1 (5011-B1), the worker 5100 described above from the shelf B-1 (5011-B1) to the shelf D-1 (5011-D1) is moved to state 2 (5302), and then the operator 5100 obtains the luggage 5012 from shelf B-1 (5011-B1) and reaches a state 3 (5303). An example is shown.

In the state 1 (5301) described above, when the worker 5100 acquires the luggage 5012 from the shelf B-1, the mobile terminal 5220 possessed by the worker 5100 and the computer 5210-B1 installed on the shelf B-1 Communicate between each other and update each other's number of information elements.

Further, when the worker 5100 moves the shelf 5011 in the state 2 and acquires the luggage from another shelf 5011-D1 in the state 3, the worker 5100 is installed on the shelf D-1 and the mobile terminal 5220 possessed by the worker 5100. Communication is performed with the computer 5210-D1, and the number of information elements is updated. That is, the information element is moved from the shelf B-1 to the shelf D-1 via the worker 5100. By carrying out such processing by a plurality of workers, the information element circulates between shelves.

After that, the computer 5210 of each shelf calculates the movement destination of each shelf from the number of information elements that it has. For example, if there are two types of information elements (state u and state v), the destination area is A when state u> state v, and the destination area when state u <= state v. B.

FIG. 33 is a diagram illustrating the relationship between the computers 5210 installed on the shelves 5011 and the movement destinations of the shelves 5011 according to the fifth embodiment. FIG. 33 shows a table 5400 describing the relationship between the computers 5210 installed on each shelf 5011 and the destination of the shelf 5011. From this table 5400, it is recommended that the shelf A-1 and the shelf B-2 are installed in the region B, and the shelf A-2, the shelf B-1, and the shelf B-3 are installed in the region A.

After that, the worker 5100 moves according to the above-described movement destination area. This movement may be executed by a self-propelled mechanism (not shown) of the shelf 5011 that receives an instruction from the computer 5210, or a robot (not shown) that moves the shelf 5011. Further, the execution timing of the movement of the shelf 5011 may be every night or every other day according to an instruction from the computer 5210.

Next, a flowchart of processing of the computer 5210 in the fifth embodiment will be described. FIG. 34 is a flowchart showing an example of processing performed by each computer 5210.

Each computer 5210 initializes the number of information elements after the start of the flowchart (step 5511). Thereafter, the reception processing (step 5512) and transmission processing (step 5513) processes of the computer 5210 are started. Thereafter, the computer 5210 activates a calculation result acquisition process (step 5514). Each of these processes (steps 5511 to 5513) can be the same as each process (steps 3111 to 3114) described in FIG.

FIG. 35 is a flowchart regarding calculation result acquisition processing (step 5514). In step 5611 in this flowchart, each computer 5210 compares the number of information elements of each state held by itself, and selects the state having the maximum number of information elements. For example, when the number of states is 2 (u and v), the number of information elements in the state u on a certain vertex is 1, and the number of information elements in the state v is 2, the computer 5210 selects the state v.

Thereafter, in step 5612, the computer 5210 determines a movement destination area corresponding to the state selected in step 5611 described above, and displays it on a predetermined display device connected to the network or the like or its own input / output device 1024. For example, when the state v is the maximum state, the corresponding region is the region B (the state u is the region A).

FIG. 36 is a diagram showing the effect of the fifth embodiment. In FIG. 36, the movement distance reduction effect of the operator 5100 by the above-described simulation of the shelf movement by the computer 5210 is shown. In this example, when the movement distance of the worker 5100 in the initial state 5700 is 1, the pick-up operation is performed in the warehouse in the convergence state 5701 when the shelf is moved based on the analysis result according to the fifth embodiment. The distance traveled by the worker was 0.08, and the reduction effect was 92%.

As described above, according to the fifth embodiment, the arrangement of the shelves can be optimized from the flow line of the worker 5100 who picks up the luggage from the shelves.

Next, an information processing method for dealing with a problem of detecting user clustering, that is, a community in an exchange service between a plurality of users such as a social network service will be described as a sixth embodiment. Here, an information processing system is assumed in which each vertex is a user and the side between the vertices is an exchange between users in the above-described third embodiment. Here, the exchange corresponds to, for example, transmission / reception of mail, transmission / reception of a message, visit to a personal page or posting.

FIG. 37 is a diagram illustrating the concept of the sixth embodiment. In FIG. 37, it is assumed that user 1 (6011) and user 2 (6012) transmit messages (6020) through the exchange

service using terminals

6031 and 6032 respectively held by the users.

In this case, the user 1 operates the terminal 1 (6031) that he / she owns, and transmits a message to the user 2 from the message transmission screen 6040, for example. This message is delivered to the terminal 2 of the user 2 via the information processing system of the exchange service such as a server, and for example, a message reception screen 6041 is displayed on the corresponding terminal.

The configurations of the

terminals

6031 and 6032 are the same as the configuration of the device 3020 shown in FIG.

FIG. 38 is a diagram illustrating an example of exchanging information elements in the sixth embodiment. In FIG. 38, it is assumed that the information element is stored in a terminal owned by each user or a storage area of each user of the information processing service providing the above-described exchange service. In this case, the initial state is state 1 (6101).

Thereafter, when the user 1 (6011) transmits a message to the user 2 (6012), the corresponding

terminals

6031 and 6032 add an information element to the message. An example of the number of information elements 6113 added to the message by the

terminals

6031 and 6032 in the state 2 (6102) is shown. In this example, five information elements are added from the terminal 6031 of the user 1.

Thereafter, in the state 3 (6103), when the terminal 2 32 of the user 2 receives the above message, the number of information elements 6112 stored in the terminal 6032 of the user 2 is updated based on the information element added to the above message. To do.

FIG. 39 is a diagram illustrating the concept of identifying the community to which each user belongs from the number of information elements at each vertex according to the sixth embodiment. Vertices A to E in the figure correspond to each terminal.

From the above series of procedures, the information element circulates between users via messages. As shown in the third embodiment and FIG. 39, the community to which each user (= vertex) belongs is specified from the number of information elements of each vertex updated in this way (the state at time t + 2 in FIG. 39). The community specified in this way can be applied, for example, by extracting frequently used keywords from public text information of users belonging to the same community and marketing to users belonging to the corresponding community. From the above, in the sixth embodiment, the community to which the user on the exchange service belongs can be detected.

Next, in the warehouse 5000 shown in the above-described fifth embodiment, a form for simulating the optimal arrangement of the shelves 5011 when the luggage list 5110 of each worker 5100 in FIG.

In the above-described fifth embodiment, the example in which the optimal arrangement of the shelves 5011 (A1 to D1) is performed according to the movement history of the worker 5100 in the warehouse 5000 is shown. In the seventh embodiment, for example, if there is sufficient time from when the computer 5210 obtains the package list 5110 to the shipment of the package in the corresponding package list 5110, the computer 5210 obtains the graph structure data from the package list 5110. Create and execute the same calculation as the information processing system 100 shown in the first embodiment.

The computer 5210 calculates the movement destination of the shelf 5011 from the calculation result (classification result) based on the relationship table between the classification result (number of information elements) and the movement destination area shown in the fifth embodiment. After the calculation of the movement destination of the shelf 5011 by the computer 5210, before the worker 5100 actually starts the pickup operation of the luggage 5012 in the warehouse 5000, the shelf 5011 is moved according to the movement destination of the shelf 5011 calculated by the computer 5210. Deploy. The arrangement means of the shelf 5011 is the same as that of the fifth embodiment. Therefore, the seventh embodiment corresponds to a method in which the computer 5210 generates graph structure data from one or more package lists 5110.

The concept that the computer 5210 generates graph structure data from the above-described package list will be described. FIG. 40 is a diagram illustrating the concept of the seventh embodiment. In this case, for example, there are three

package lists

7001, 7002, and 7003, and it is assumed that the worker 5100 picks up the package 5012 in the order of the numbers in the lists 7001 to 7003. Further, assuming that each shelf 5011 is a vertex, the trajectory of movement of the worker 5100 in the warehouse 5000 can be represented as a graph 7010 of the trajectory of the worker 5100.

The value added to each side of the graph 7010 is the number of times the worker 5100 has passed. When the edges of the graph 7010 are normalized by the maximum number of passages described above, a graph 7011 of the trajectory of the standardized worker 5100 is obtained. This graph 7011 is graph structure data in which a weight is added to each side.

By processing such graph structure data by the same information processing system as in the first and second embodiments, it is possible to classify the vertices, that is, the shelves 5011, and the movement destination corresponding to the classification result is the method of the fifth embodiment. By calculating in step S1, it is possible to calculate the optimum movement destination of each shelf 5011.

Next, an embodiment in which the computer of the information processing system has the transition probability formula (formula (1)) in the first embodiment as a transition probability table will be described as an eighth embodiment.

FIG. 41 is a diagram showing an example of the transition probability table 1 (8001). FIG. 42 is a diagram illustrating an example of the transition probability table 2 (8002). FIG. 43 is a diagram illustrating an example of the transition probability table 3 (8003).

The computer of the information processing system compares the number of information elements (u, v) of its own vertex with the number of information elements (ΣNju, ΣNjv) of its own vertex against the transition probability tables 8001 to 8003, and the corresponding values in the table. The transition probability can be determined by specifying. The values in each of the above tables 8001 to 8003 are obtained by experimentally calculating values for obtaining a target result by a computer executing a simulation in advance.

Next, in the information processing system defined by the calculation model of the first embodiment, when an error occurs in processing performed by the computer 220 or communication between the computers 220 and the data of the information child fluctuates, the total number of information children is An example of suppressing fluctuation will be described as a ninth embodiment. The configuration of the information processing system is the same as that of the information processing system 100 of the first embodiment.

Describe the effect of the error on this information processing system. As shown in the first embodiment, the information processing system communicates and exchanges information elements between the computers 220 to realize calculation of the entire information processing system.

However, if there is an error, the entire information processing system may not be calculated as desired. For example, in a simple example, when an information element is moving between a first computer and a second computer, an error occurs between the computers 220 at a certain rate, and the information element is lost. The total number of information elements in the entire information processing system decreases with time, and eventually the information elements do not move and a desired processing result cannot be obtained.

Conversely, when the total number of information elements of the entire system increases with time, a desired processing result cannot be obtained. When the total number of information elements increases, for example, a case in which a function for increasing the number of decreased information elements is built in can be considered.

FIG. 44 is a diagram illustrating a change in the distribution of information elements in the entire information processing system when the total number of information elements is changed in the ninth embodiment. State 1 (9001) to state 4 (9004) in FIG. 44 show the distribution of information elements. Also, in this figure, vertices are arranged in a two-dimensional grid, and information elements are prepared in two states (u and v), and at each vertex, the number of information elements in state u> the number of information elements in state v The distribution is gray, otherwise black.

44, the processing of the classification problem shown in the first embodiment can be realized in the state 2 (9002). However, when the total number of information elements decreases, the state 1 (9001) is obtained, and when the total number of information elements increases, the state 3 (9003) or the state 4 (9004) is obtained and a desired processing result cannot be obtained. Therefore, a method for keeping the total number of information elements within a predetermined amount or within a predetermined range in an information processing system that performs processing in an environment having an error will be described.

As shown in FIG. 44, the distribution of information elements is different in the number of information elements in this information processing system. Therefore, by combining the method for detecting the state and the method for controlling the number of information elements according to each state at each vertex, the total number of information elements in the entire system can be maintained at a constant amount.

45 to 48 show flowcharts in which the control processing of the number of information elements is added. First, FIG. 45 shows a flowchart in which the information element number control process 9101 is added to the flowchart of FIG. 3 (FIG. 3) shown in the first embodiment. The information element number control process (9101) is executed after the reception process (315) at each vertex. The other processes are the same as those in the first embodiment shown in FIG.

Next, FIG. 46 shows a flowchart of the information element number control process (9101). In this process, first, the computer 220 obtains the time change data of the number of information elements of the vertexes to be calculated from the main storage device 222 (9201).

The computer 220 predicts the system state using the state prediction algorithm based on the acquired time change data (9202). In the ninth embodiment, any one of the states 1 to 4 is predicted. State 1 means that the number of information elements is small, state 2 and state 3 increase in order, and state 4 is the largest.

The computer 220 determines the next process according to the predicted state. In the case of the state 1, the information element splitting process (9210) is executed, in the case of the state 2, nothing is performed, and in the case of the state 3 or the state 4, the information element fusion process is executed (9211). Thereafter, this process ends.

Also, the processing (split, merge, do nothing) corresponding to each state 1 to 4 can be changed according to the purpose. In the ninth embodiment, since the state 2 is a desired state, the state 1 having a smaller number of information elements than the state 2 is split, and the

states

3 and 4 having a larger number of information elements are fusion processes.

Here, the state prediction (9202) and the branch of the processing of each state (9203) will be specifically described. As a feature quantity used for state prediction, for example, a mutual information quantity and a power spectrum are used.

FIG. 49 shows the relationship between the number of information elements, the mutual information calculated in the time direction on each vertex, and the power spectrum of the change in the number of information elements on each vertex. The computer 220 stores the number of information elements during a predetermined time T (time change data), and calculates a mutual information amount and a power spectrum at the time T.

The mutual information amount is calculated from two information amounts in the time direction, and each information amount is calculated from the occurrence probability of the state of each vertex within a certain time T. The probability of occurrence is, for example, when the information element has two states (u and v) and the state of each vertex is also two states (u and v are determined by the number of information elements on the vertex). Can be calculated from the ratio of the state u within the time T, and the probability Pv can be calculated from the ratio of the state v.

The amount of information tends to decrease as the probabilities become the same. For example, when Pu is 0 or Pv is 0, the information amount is also 0. Mutual information has the same tendency.

Also, the power spectrum is calculated from the amount of change of the information element on each vertex. Each computer 220 performs fast Fourier transform on the amount of change of the information element on the apex, and a power spectrum is calculated as the power value of the AC component obtained by removing the DC component from the conversion result. The power spectrum is an index that takes a low value when the amount of information elements does not change, or when the change amount does not have a constant period such as noise, and takes a large value when it repeatedly increases and decreases at a constant period.

In FIG. 49, Iave (9502) indicates the average value of the mutual information amount on each vertex, Idif (9501) indicates the maximum and minimum difference of the mutual information amount on each vertex, and PWac indicates the power spectrum (9503). Indicates.

From FIG. 49, it is also possible to use the flow rate of the information element as the feature quantity used for state prediction. The flow rate indicates the magnitude of change in the number of information elements per certain time. For example, the total number of information elements transmitted within a certain time T (the amount of information elements flowing out when viewed from the top) or the number of received information elements This is the total number (inflow of information elements when viewed from the apex) or the difference between the maximum number and minimum number of information elements within a certain time T.

FIG. 53 shows an example of the algorithm 530 described in pseudo code for determining the states 1 to 4.

Algorithm 530 shows a state in which x1 indicates an average value Iave (9502) of mutual information, x3 indicates a power spectrum (9503), and y is predicted. By the method as described above, the state of the entire information processing system is predicted at each vertex.

State 1 is a state in which the number of information elements is small and the state of each vertex is stationary (the change amount of the information element is 0). For example, the mutual information amount is 0 (any occurrence probability of each state is 0). It has become). In this case, the maximum and minimum differences and the power spectrum of the mutual information amount are also zero. State 2 is determined when determination of state 1 is false and the mutual information amount is smaller than a predetermined threshold (th1). This is because in state 2, the amount of information is particularly large near the boundary, but the state does not change as a whole and the amount of information is small.

Next, the determination of state 3 is determined as state 3 when the power spectrum is equal to or greater than the threshold (th2). This represents that in state 3, the pattern is periodically changed. The other state is determined as state 4.

With the above algorithm 530, the computer 220 predicts states 1 to 4 corresponding to the total number of information elements from the mutual information amount and the power spectrum.

In addition, although the example which uses a mutual information amount and a power spectrum as a feature-value used for state prediction was shown, the amount of change of the number of information elements within a predetermined time may be used. Mutual information can be used as an example of the amount of change in the number of information elements within a predetermined time. A power spectrum can be used as an example of the amount of change in the number of information elements within a predetermined time. Further, as an example of the amount of change in the number of information elements within a predetermined time, the difference between the maximum value and the minimum value of the number of information elements can be used.

Next, FIG. 47 shows a flowchart of information element splitting processing (9210). In the splitting process, first, the computer 220 selects one information element from among the information elements stored at the vertex to be processed (9301). The method of selecting an information element is a method of selecting at random from the information elements stored at its own vertex, or the information element stored at its own vertex is randomly selected from the information elements. Or the like (for example, if there are three state u information elements and two state v information elements at the vertices), a method of selecting at random from the state u information elements may be employed as appropriate.

Thereafter, the computer 220 executes a process of generating a new information element based on the selected information element (9302). As an example of the generation process, the computer 220 generates one information element having the same state as the selected information element. As described in the first embodiment, the information element has a state variable, for example, u and v in the case of two states. Therefore, when the state of the selected information element is u, the state of the generated information element is also u.

Through the above processing, the computer 220 can generate a new information element and add it to the information element of the information processing system.

FIG. 48 is a flowchart of information element fusion processing (9211). In the fusion process, first, the computer 220 selects one information element from among the information elements stored at the vertex to be processed (9401). The selected information element is called a fusion source information element.

The method of selecting one information element from among the information elements is a random selection method or the number of information elements stored at its own vertex, which is selected at random from the information elements having the most states (for example, If there are three state u information elements and two state v information elements at the vertices, a method of randomly selecting from among the state u information elements, or the newest information element among the received information elements A selection method or the like may be adopted as appropriate.

Thereafter, the computer 220 selects an information child at the fusion destination based on the information child at the fusion source (9402). In selecting the merged information element, an information element having the same state as the selected information element is selected at random.

Thereafter, the computer 220 performs a fusion process based on the fusion source information element and the fusion destination information element in the process 9401 (9403). In the fusion process, the data of the merged information child is deleted. That is, the information element at the fusion destination disappears, the vertex at the fusion source is retained, and the number of information elements in the own vertex decreases by one.

Through the above processing, the computer 220 can delete the merged information child and delete it from the information child of the information processing system.

By processing according to the above method, the computer 220 can keep the total number of information elements in the entire information processing system within a certain range (within a range where desired processing is performed) without aggregating all information elements. Become.

FIG. 50 shows a conceptual diagram of the ninth embodiment. When the number of information elements in the entire information processing system decreases, the state shifts to state 1 (9001) at each vertex, and the computer 220 performs information element division processing (9210) to increase the number of information elements. On the other hand, when the number of information elements in the entire information processing system increases, the state shifts to

states

3 and 4 at each vertex, and the computer 220 performs information element fusion processing, thereby reducing the number of information elements. In this way, the information processing system adjusts the total number of information elements by the information element splitting or fusion process or maintaining the information elements according to the prediction result of the total number of information elements (states 1 to 4). 2 can be maintained.

Next, the tenth embodiment shows a configuration example in the case where there are a plurality of states of the entire information processing system using the management method of the total number of information elements described in the ninth embodiment. FIG. 51 is a flowchart in which the number of states of the entire system shown in FIG. 46 of the ninth embodiment is changed to a plurality (n).

In the tenth embodiment, it is assumed that the number of states is n and the number of information elements is arranged in ascending order from state 1 (state 1 is the smallest).

51 are the same as those in FIG. 46 of the ninth embodiment. In step 10001 of FIG. 51, the computer 220 first acquires the target state from the main storage device 222. The target state represents a state that realizes desired processing of the entire information processing system, and can be set from a management computer (not shown). Alternatively, it may be preset in the computer 220.

In the tenth embodiment, a case where the state 3 is preset as the target state will be described. When the target state is state 3, the subscript codes of each state are arranged in ascending order of information elements, so states 1 and 2 have a small number of information elements, and states 4 to n have a large number of information elements. Determined.

In this process, it is determined whether to split the information element assigned to each of the states 1 to n, the fusion process, or no operation. For example, in states 1-2, the splitting / fusion processing (10010, 10011) of the information element is determined as splitting processing, and in states 4-n (10012, 10013), the fusion processing is determined.

Thereafter, according to the determination, the computer 220 performs either one of the split / fusion processing of the information element (10010 to 10013). Here, the splitting process and the fusion process are the same as the processing contents described in FIGS. 47 and 48 of the ninth embodiment. The algorithm related to the conditional branch in step 10001 is stored in the main storage device 222 of the computer 220 where the vertex is processed, and the condition can be changed by changing the algorithm via the network. . Note that the algorithm related to the conditional branch in step 10001 is the same as the algorithm 530 of the ninth embodiment, and states 1 to n corresponding to the total number of information elements are predicted from the mutual information amount and the power spectrum.

As described above, even when the number of states is n, the number of information elements is adjusted according to the prediction result of the total number of information elements (states 1 to n) in the same manner as in the ninth embodiment. State 3 (target state) can be maintained.

Next, Example 11 shows a configuration obtained by adding the processing of the entire information processing system to the configuration described in Example 10. FIG. 52 is a flowchart illustrating an example of processing performed by the computer according to the eleventh embodiment. This flowchart is obtained by adding the processing of the entire information processing system to the processing of FIG. 51 described in the tenth embodiment.

In the eleventh embodiment, it is assumed that the number of states of the entire information processing system is n as in the tenth embodiment, and the state 1 to the state n are arranged in ascending order from a state with few information elements.

In the eleventh embodiment, processing (10011, 10021, 10031, 10041) for the entire information processing system in each state is added after the split / fusion processing (10010, 10020, 10030, 10040) of the information element in each state. It has been. Each of these processes can be performed differently according to the states 1 to n of the entire information processing system.

Furthermore, the computer 220 of the eleventh embodiment is provided with a moving device capable of changing the physical position, and can be moved by processing according to the states 1 to n of the entire information processing system. Further, consider a case where the communication error between the computers 220 depends on the physical distance between the computers 220, and the error rate increases when the distance is long and decreases when the distance is short.

The number of states of the information processing system is 4 (n = 4), and the desired state is 2. At this time, if the state of the entire information processing system is state 1, that is, the total number of information elements is small and the number of information elements is increased, the computer 220 causes the information element division / fusion process 10010 to increase. Execute the splitting process of information.

After that, in the processing of state 1 (10011), for the purpose of improving the communication error considered to be the cause of the decrease in the number of information elements, the aggregation is determined for the computer 220 to which the mobile device is added. The agglomeration may be a method of gathering at predetermined points by moving the computers 220 so as to approach each other, a method of approaching the adjacent computers 220, or the like. By aggregating the plurality of computers 220, the distance between the computers is shortened, the communication error rate is improved, and desired processing can be realized for the entire system.

On the other hand, when the information processing system as a whole is in

states

3 and 4, since the total number of information elements is large and the number of information elements is reduced, the computer 220 performs information element split / fusion processing (10030, 10040). ) Integration of information elements.

Thereafter, the computer 220 determines whether or not a communication error is still acceptable in the processing of states 3 and 4 (10031 and 10041), and makes a spread decision to the computer 220 to which the mobile device is added. The spread is to move the computers 220 away from each other. As an effect of the configuration disclosed in the eleventh embodiment, for example, when the computer 220 with the moving device is a robot, a plurality of robots have an effect of keeping the distance between each other within a certain range.

In the above example, the case where the communication error differs depending on the distance is shown. For example, in the network I / F 225, a communication network such as a proximity infrared communication with a high communication error rate and a wireless LAN with a low communication error rate is used. It may be a method of switching, or a method of switching a communication path with a high error rate to a communication path with a low error rate.

As described above, according to the eleventh embodiment, in addition to the effects of the tenth embodiment, it is possible to control the mobile device according to the occurrence state of the communication error and the communication state. It becomes possible to move the computer 220 to a position where the occurrence rate is low.

As described above, according to the information processing system and the information processing method of the first to eleventh embodiments, it is possible to efficiently calculate large-scale data that cannot be collected in one place or data that is updated every moment.

記載 At least the following will be made clear by the description in this specification. That is, in the information processing systems according to the first to eleventh embodiments, each computer uses the identifier attribute having the largest number among the identifiers (information elements) stored by itself as to the vertex as the attribute of the corresponding vertex. It may be determined. According to this, event clustering can be efficiently performed for events corresponding to each vertex.

In the information processing system described above, each of the computers uses, as the algorithm, the number of identifiers associated with each vertex stored by itself and the number of identifiers associated with each vertex stored by the adjacent computer as variables. A mathematical function for calculating the transition probability of the corresponding identifier from the own computer to another adjacent computer may be held by a predetermined function, and the transition probability may be calculated using the mathematical formula. According to this, it is possible to perform an efficient and accurate update process with respect to the number of identifiers that are the basis for clustering each event.

Further, in the information processing system described above, each of the computers, as the algorithm, depends on a relationship between the number of identifiers for each vertex stored by itself and the number of identifiers for each vertex stored by the adjacent computer. It is also possible to hold a table that predetermines the transition probability of the corresponding identifier from the own computer to another adjacent computer, and calculate the transition probability using the table. According to this, it becomes possible to perform a more efficient and accurate update process with respect to the number of identifiers that are the basis of clustering of each event.

In the information processing system described above, when each computer updates the number of identifiers according to the calculation result of the transition probability, it maintains the total number of identifiers stored in itself and the adjacent computer. It is good also as what updates. According to this, it is possible to perform an efficient and accurate update process for communication of identifiers.

In the above information processing system, one computer may correspond to a plurality of vertices in the graph structure. According to this, it becomes possible to execute the information processing method of the present invention in a server device that controls a plurality of vertices, that is, a plurality of events.

Further, in the information processing system including a plurality of terminals that transmit and receive messages via a network and a management computer in the first to eleventh embodiments, each terminal replaces message transmission and reception via the network, When each terminal is physically close to each other, direct communication may be performed between the corresponding terminals, and the number of identifiers stored in the corresponding terminals may be updated. According to this, it becomes possible to perform processing corresponding to not only a wide-area communication line such as the Internet but also a mode of message exchange by means such as various proximity wireless communication.

In addition, this invention is not limited to each above-mentioned Example, Various modifications are included. For example, the above-described embodiments are described in detail for easy understanding of the present invention, and are not necessarily limited to those having all the configurations described. Further, a part of the configuration of one embodiment can be replaced with the configuration of another embodiment, and the configuration of another embodiment can be added to the configuration of one embodiment. In addition, any of the additions, deletions, or substitutions of other configurations can be applied to a part of the configuration of each embodiment, either alone or in combination.

In addition, each of the above-described configurations, functions, processing units, processing means, and the like may be realized by hardware by designing a part or all of them with, for example, an integrated circuit. In addition, each of the above-described configurations, functions, and the like may be realized by software by the processor interpreting and executing a program that realizes each function. Information such as programs, tables, and files that realize each function can be stored in a memory, a hard disk, a recording device such as an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.

Also, the control lines and information lines indicate what is considered necessary for the explanation, and not all the control lines and information lines on the product are necessarily shown. Actually, it may be considered that almost all the components are connected to each other.

Claims

A graph structure composed of a plurality of vertices corresponding to the event to be analyzed and edges connecting the corresponding vertices according to the relationship between the corresponding events is used as a model, corresponding to each of the vertices, Correspondingly, a plurality of computers connected to each other so as to be able to exchange data,
A storage device connected to each of the computers and holding an identifier including an attribute representing one or more states for the vertex event;
A display device that displays the number of identifiers held for each vertex in the storage device as a spatial distribution map based on the distribution of the vertices;
Each of the computers is connected to the computer connected by the edge and corresponds to the adjacent vertex by a predetermined algorithm based on the number of identifiers held by each other, and the identifier transition probability between the computers is determined. Calculate, update the number of identifiers held by each computer according to the calculation result, determine the update amount according to a predetermined algorithm based on the feature amount indicating the amount of change in the number of identifiers due to the update, and An information processing system, wherein the number of identifiers is updated based on an update amount.
The information processing system according to claim 1,
The feature amount is
An information processing system characterized in that it is a mutual information amount calculated based on a change amount of the number of identifiers within a predetermined time.
The information processing system according to claim 1,
The feature amount is
An information processing system characterized by being a power spectrum calculated based on a change amount of the number of identifiers within a certain time.
The information processing system according to claim 2,
The feature amount is
An information processing system characterized in that a difference between a maximum value and a minimum value of the mutual information amount.
The information processing system according to claim 1,
Each of the computers is
Based on one or more feature amounts, the state of each vertex is classified into a predetermined number of groups, and either increase, decrease or maintenance of the number of identifiers is selected according to the classified group. Information processing system.
The information processing system according to claim 5,
The group is classified as one of the first state to the fourth state,
When the feature amount is 0, the state of each vertex is classified into a first state where it is stationary,
When the mutual information amount calculated from the two information amounts in the time direction is smaller than a predetermined first threshold value, it is classified into the second state,
When the power spectrum calculated from the amount of change is greater than or equal to a second threshold, classify into the third state,
An information processing system, wherein the states other than the first to third states are classified into a fourth state.
The information processing system according to claim 6,
Each of the computers is
In the first state, an increase in the number of identifiers is selected to perform information element splitting processing,
In the second state, choose to maintain the number of identifiers;
In the third or fourth state, the information processing system is characterized by selecting a decrease in the identifier and executing an information element fusion process.
A graph structure composed of a plurality of vertices corresponding to the event to be analyzed and edges connecting the corresponding vertices according to the relationship between the corresponding events is used as a model, corresponding to each of the vertices, Multiple computers connected to each other so that data can be exchanged
Storing an identifier including an attribute representing one or more states for the event of the vertex in a storage device connected to each of the computers;
The number of identifiers held for each vertex in the storage device is output to the display device as a spatial distribution map based on the vertex distribution,
Each computer has a transition algorithm of identifiers between the computers by a predetermined algorithm based on the number of identifiers held between the computers connected by the edges and corresponding to the adjacent vertices. Calculate, update the number of identifiers held by each computer according to the calculation result, determine the update amount according to a predetermined algorithm based on the feature amount indicating the amount of change in the number of identifiers due to the update, and An information processing method, wherein the number of identifiers is updated based on an update amount.
The information processing method according to claim 8,
The feature amount is
An information processing method, wherein the mutual information amount is calculated based on a change amount of the number of identifiers within a predetermined time.
The information processing method according to claim 8,
The feature amount is
An information processing method, wherein the information is a power spectrum calculated based on a change amount of the number of identifiers within a predetermined time.
An information processing method according to claim 9,
The feature amount is
An information processing method characterized by being a difference between a maximum value and a minimum value of the mutual information amount.
The information processing method according to claim 8,
Each of the computers classifies the state of each vertex into a predetermined number of groups based on one or more feature quantities, and increases, decreases or maintains the number of identifiers according to the classified group. An information processing method characterized by selecting.
An information processing method according to claim 12,
Each of the computers classifies the group into any one of a first state to a fourth state;
When the feature amount is 0, the state of each vertex is classified into a first state where it is stationary,
When the mutual information amount calculated from the two information amounts in the time direction is smaller than a predetermined first threshold value, it is classified into the second state,
When the power spectrum calculated from the amount of change is greater than or equal to a second threshold, classify into the third state,
An information processing method, wherein the states other than the first to third states are classified into a fourth state.
An information processing method according to claim 13,
Each of the computers selects an increase in the number of identifiers in the first state and executes an information element splitting process, and in the second state selects to maintain the number of identifiers, In the third or fourth state, an information process is performed by selecting the decrease of the identifier and executing information element fusion processing.