Evaluation index value calculation device, evaluation index value calculation method, and evaluation index value calculation program

The evaluation index value calculation device addresses the inflexibility of existing search methods by calculating and integrating evaluation index values based on specified ranges and types within graph structures, improving the optimization of molecular compounds and ICT systems.

JP7581704B2Active Publication Date: 2024-11-13NEC CORP

Patent Information

Application Number
JP2020141565
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Filing Date
2020-08-25
Publication Date
2024-11-13
Estimated Expiration
2040-08-25

AI Technical Summary

Technical Problem

Existing search methods in multi-objective reinforcement learning for graph structures, such as molecular compounds and ICT systems, fail to adapt to different evaluation indexes specified at varying ranges within the graph, leading to inefficiencies in optimizing search models.

Method used

An evaluation index value calculation device and method that calculates and integrates evaluation index values based on specified ranges and types within the graph, using a function corresponding to the type of evaluation index, allowing for flexible and comprehensive evaluation value output.

Benefits of technology

Enables the output of a comprehensive evaluation value that accurately reflects the specified range and type of evaluation index, enhancing the flexibility and effectiveness of search models in optimizing graph structures.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007581704000001
    Figure 0007581704000001
  • Figure 0007581704000002
    Figure 0007581704000002
  • Figure 0007581704000003
    Figure 0007581704000003
Patent Text Reader

Abstract

To provide an evaluation index value calculation device with which it is possible to output an overall evaluation value in accordance with the range of search states where an evaluation index is designated or the type of evaluation index.SOLUTION: An evaluation index value calculation device 20 comprises: a calculation unit 21 which, when a graph where an evaluation index is designated in at least a portion of it is inputted, calculates the value of evaluation index in one or more ranges of the graph where an evaluation index may possibly be designated for each of the ranges, respectively, using a prescribed function that corresponds to the type of evaluation index; and a selection unit 22 for selecting a value that corresponds to the range of the graph for which the evaluation index is designated, among one or more of calculated evaluation indices.SELECTED DRAWING: Figure 7
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to an evaluation index value calculation device, an evaluation index value calculation method, and an evaluation index value calculation program. [Background technology]

[0002] Multi-objective reinforcement learning, a branch of machine learning, is applied to the problem of optimizing graph structures that represent information, such as those found in molecular compounds and ICT (Information and Communication Technology) systems. The main objective of multi-objective reinforcement learning is to learn a cost function model that enables a search that satisfies multiple evaluation indicators.

[0003] In the above problems, particularly in problems where the search state is represented by a graph, the evaluation index represents a type of feature calculated from the graph structure. The evaluation index may be specified for a part or the whole graph.

[0004] In general, a number of different functions correspond to each evaluation index, i.e., the input values ​​to and output values ​​from the functions are different for each type of evaluation index.

[0005] Fig. 8 is an explanatory diagram showing an example of an evaluation index defined for the entire graph representing a molecular compound and a function corresponding to the evaluation index. Fig. 8 shows an example of the evaluation index. In the example shown in Fig. 8, QED (Quantitative Estimate of Druglikeness), which represents druglikeness, is adopted as the evaluation index.

[0006] The graph enclosed by the dashed line in FIG. 8 represents an example of a molecular compound. The function f QED receives the graph shown in FIG. 8 as input and performs numerical calculation processing on the input graph. In the example shown in FIG. 8, the function f QED outputs a numerical value s as the result of numerical computation on the graph.

[0007] Non-Patent Document 1 describes a method of searching for a molecular structure that optimizes multiple evaluation indexes by performing multi-objective reinforcement learning on a graph representing a molecular compound as shown in Fig. 8. The values ​​of the evaluation indexes used in the method described in Non-Patent Document 1 are values ​​calculated from the structure of the molecular compound, such as druglikeness and molecular weight. Each evaluation index is specified as an index for the entire graph structure.

[0008] In addition, in the method described in Non-Patent Document 1, a single scalar value obtained by converting the values ​​of the above-mentioned multiple evaluation indexes is used as a teacher label to learn a cost function model for search.

[0009] Fig. 9 is an explanatory diagram showing an example of an evaluation index defined for the entire graph expressing an ICT system and a function corresponding to the evaluation index. Fig. 9 shows an example of the evaluation index.

[0010] The graph enclosed by the dashed line in Fig. 9 represents an example of an ICT system. The graph in Fig. 9 represents the functional configuration of the system.

[0011] Specifically, each vertex in the graph shown in Fig. 9 represents an IT / NW (Information Technology / Network) device such as an application, middleware, a server, a router, a firewall, etc. Also, each edge in the graph shown in Fig. 9 represents a connection relationship between IT / NW devices such as HTTP (Hypertext Transfer Protocol) communication, TCP (Transmission Control Protocol) communication, IP (Internet Protocol) communication, Ethernet (registered trademark) communication, etc.

[0012] In the example shown in Fig. 9, the cost defined in the ICT system is the evaluation index. cost is a function that calculates the defined cost from the graph. In the example shown in FIG. 9, the function f costoutputs the cost s calculated from the graph.

[0013] Fig. 10 is an explanatory diagram showing an example of an evaluation index defined for a part of a graph expressing an ICT system and a function corresponding to the evaluation index. Fig. 10 shows an example of the evaluation index.

[0014] The graph enclosed by the dashed line in Fig. 10 represents an example of an ICT system. The graph illustrated in Fig. 10 is the same as the graph enclosed by the dashed line in Fig. 9.

[0015] In addition, the function f bandwidth is a function that receives as input the graph shown in the dashed frame in Fig. 10 and edges within the designated range candidate shown in the dashed frame inside the graph in Fig. 10, and calculates the bandwidth of the received edges. The designated range candidate shown in Fig. 10 is a range in which an evaluation index may be specified.

[0016] That is, in the example shown in FIG. 10, the band defined for a particular side is the evaluation index. Also, the function f bandwidth is a function that calculates the bandwidth defined for an edge. In the example shown in FIG. 10, the function f bandwidth outputs the band s calculated from the edges within the specified range candidate.

[0017] Note that the candidate designated range may be a range surrounding the vertices of a graph, a range wider than the range surrounding a subgraph, or the like, other than the range surrounding the edges of the graph shown in FIG.

[0018] Non-Patent Document 2 describes a method for deriving a system configuration that optimizes an evaluation index by performing reinforcement learning on a graph representing an ICT system as shown in Figures 9 to 10. The evaluation index used in the method described in Non-Patent Document 2 is a functional requirement.

[0019] Functional requirements are requirements for the functions required of a system. For example, functional requirements may include "analyzing images captured by a camera using analysis software" and "monitoring the results of the analysis."

[0020] The method described in Non-Patent Document 2 derives a system configuration that satisfies input functional requirements by searching for a system configuration that optimizes an evaluation index. The presence or absence of a functional requirement being satisfied is expressed by a value of 0 or 1. [Prior art documents] [Non-patent literature]

[0021] [Non-Patent Document 1] J. You, B. Liu, R. Ying, V. Pande, and J. Leskovec, “Graph Convolutional Policy Network for Goal-Directed Molecular Graph Generation,” Neural Information Processing Systems, 2018. [Non-Patent Document 2] Takashi Maruyama, Takuya Kuwahara, Yutaka Yakuwa, Takayuki Kuroda, and Yoichi Sato, "Improvement of efficiency of search by reinforcement learning in search-based network design derivation method," IEICE Technical Report, vol. 118, no. 483, ICM2018-71, pp. 123-128, March 2019. Summary of the Invention [Problem to be solved by the invention]

[0022] The types of evaluation indices used in the search methods described in the above Non-Patent Documents 1 and 2 are not changed for each search. In addition, the search model used in the search methods described in Non-Patent Documents 1 and 2 is trained to directly output a scalar value in which multiple evaluation indices are integrated.

[0023] Therefore, even if the evaluation index of the optimization target during search is different from the evaluation index used to calculate the value of the teacher data during learning, the search model calculates a predicted scalar value during search that integrates the evaluation indexes used during learning.

[0024] In addition, the search model used in the search methods described in Non-Patent Documents 1 and 2 is a model that is used on the premise that the specified range, which is the range in which an evaluation index is specified for a search state, is fixed to a range that surrounds the entire graph.

[0025] That is, the above-described searching model cannot handle evaluation metrics that are specified as part of a graph, because the searching model is not configured to allow for the input of a specified range that encloses a part of a graph.

[0026] For the above reasons, the search methods described in Non-Patent Documents 1 and 2 have difficulty in performing a search according to the type of evaluation index specified during a search or a specified range that encloses a part of a graph.

[0027] Therefore, one of the objectives of the present invention is to provide an evaluation index value calculation device, an evaluation index value calculation method, and an evaluation index value calculation program that can output an overall evaluation value according to the range of search states to which the evaluation index is specified and the type of evaluation index. [Means for solving the problem]

[0028] In an embodiment of the present invention, when a graph having an evaluation index specified in at least a part thereof is input, Among the elements that make up Evaluation metrics may be specified Is in range One or more designation range candidate In Regarding elements The value of the evaluation index is calculated by using a predetermined function corresponding to the type of evaluation index. designation range candidate A calculation section calculates the values ​​of one or more evaluation indices for each evaluation index, and among the calculated values, the evaluation indexes for which are specified in the graph are calculated. A range is a specified range. range Calculated from the candidates and a selection unit for selecting a value.

[0029] In an embodiment of the present invention, the evaluation index value calculation method includes: Among the elements that make up Evaluation metrics may be specified Is in range One or more designation range candidate In Regarding elements The value of the evaluation index is calculated by using a predetermined function corresponding to the type of evaluation index. designation range candidate The values ​​of one or more evaluation indices are calculated for each item, and the evaluation indices are displayed in the graph. A range is a specified range. range Calculated from the candidates Select a value.

[0030] In an embodiment of the present invention, the evaluation index value calculation program is a program for calculating an evaluation index value when a graph having an evaluation index specified at least in a part thereof is input to a computer. Among the elements that make up Evaluation metrics may be specified Is in range One or more designation range candidate In Regarding elements The value of the evaluation index is calculated by using a predetermined function corresponding to the type of evaluation index. designation range candidate The calculation process calculates the value of one or more evaluation indicators for each evaluation indicator, and the calculated value of one or more evaluation indicators for which an evaluation indicator is specified in the graph A range is a specified range. range Calculated from the candidates A selection process is performed to select a value. Effect of the Invention

[0031] According to the present invention, a comprehensive evaluation value can be output according to the range of search states in which the evaluation index is specified and the type of evaluation index. [Brief description of the drawings]

[0032] [Figure 1] 1 is a block diagram showing an example of the configuration of an evaluation index value calculation device according to an embodiment of the present invention. [Diagram 2] 10 is an explanatory diagram showing an example of a calculation process performed by an evaluation index value prediction unit 110. FIG. [Diagram 3]10 is an explanatory diagram showing an example of a selection process performed by an evaluation index value selection unit 120. FIG. [Figure 4] 11 is an explanatory diagram showing an example of a conversion process performed by an evaluation index value integrating unit 130. FIG. [Diagram 5] 10 is a flowchart showing the operation of a comprehensive evaluation value output process by the evaluation index value calculation device 100 of this embodiment. [Figure 6] FIG. 1 is a schematic block diagram showing an example of the configuration of a computer in which an evaluation index value calculation device 100 of this embodiment is implemented. [Figure 7] 1 is a block diagram showing an overview of a system verification program generating device according to the present invention; [Figure 8] FIG. 2 is an explanatory diagram showing an example of an evaluation index defined for an entire graph representing a molecular compound and a function corresponding to the evaluation index. [Figure 9] 1 is an explanatory diagram showing examples of evaluation indices defined for the entire graph representing an ICT system and functions corresponding to the evaluation indices; [Figure 10] 1 is an explanatory diagram showing examples of evaluation indices defined for a part of a graph representing an ICT system and functions corresponding to the evaluation indices; DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0033] [Configuration Description] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, an embodiment of the present invention will be described with reference to the drawings. Fig. 1 is a block diagram showing an example of the configuration of an evaluation index value calculation device according to an embodiment of the present invention.

[0034] As mentioned above, there are methods for exploratory optimization of graph structures representing information such as molecular compounds and ICT systems, using a pre-trained cost function model to satisfy a specified evaluation index.

[0035] The evaluation index value calculation device 100 of this embodiment handles a cost function model and a learning method for the cost function model when an evaluation index to be referenced when optimizing the structure is assigned to a part of the graph structure and there is a possibility that the evaluation index may change for each search.

[0036] As shown in FIG. 1, an evaluation index value calculation device 100 of this embodiment includes an evaluation index value prediction unit 110, an evaluation index value selection unit 120, and an evaluation index value integration unit .

[0037] 1, the evaluation index value prediction unit 110 is communicably connected to a graph input device 200. Also, as shown in FIG 1, the evaluation index value integration unit 130 is communicably connected to a comprehensive evaluation index value output device 300.

[0038] The evaluation index value prediction unit 110 has a function of calculating the value of an evaluation index for a graph input from the graph input device 200 as a predicted value.

[0039] Specifically, the evaluation index value prediction unit 110 receives a graph to which evaluation indexes are assigned as input. Next, the evaluation index value prediction unit 110 calculates, for each type of evaluation index, the values ​​of evaluation indexes for vertices and edges in a candidate specified range, which is a range in which an evaluation index may be specified, among the vertices and edges constituting the input graph.

[0040] That is, the evaluation index value prediction unit 110 calculates the value of the evaluation index for the designated range candidate in the input graph. Note that there may be a plurality of designated range candidates in one graph.

[0041] Fig. 2 is an explanatory diagram showing an example of calculation processing by the evaluation index value prediction unit 110. Fig. 2 shows calculation processing for a graph in which "delay" is specified as an evaluation index. The evaluation index value prediction unit 110 calculates the value of the evaluation index for the input graph.

[0042] The graph enclosed by the dashed line in Fig. 2 represents an ICT system in which "delay" is specified as an evaluation index. In addition, the function f delay is a function that calculates the value of the evaluation index "delay".

[0043] Moreover, the evaluation index values ​​s1 and s2 enclosed by dashed lines in FIG. 2 are delay values ​​calculated from the designated range candidates A and B that respectively enclose the edges that make up the input graph.

[0044] The evaluation index value prediction unit 110 receives the graph shown in FIG. 2 in which "delay" is specified as the evaluation index. Next, the evaluation index value prediction unit 110 predicts the value of each delay for the designated range candidates A and B, which are ranges in which "delay" may be specified as the evaluation index, using a function f delay Next, the evaluation index value prediction unit 110 outputs the evaluation index values ​​s1 and s2 as the calculation results.

[0045] The candidate range may be determined in advance for each type of evaluation index. For example, the candidate range for the evaluation index of "bandwidth" may be defined in advance to be set only for edges that represent communication corresponding to the IP layer among edges that constitute a graph representing an ICT system.

[0046] The value of the evaluation index may be composed of multiple values ​​calculated based on each of the multiple evaluation indexes. The evaluation index value prediction unit 110 may include in the calculation result at least the value of the evaluation index calculated from the range in which the evaluation index in the graph is specified.

[0047] Furthermore, the function used in the evaluation index value prediction unit 110 may be any function that can calculate the value of the evaluation index. For example, the function may be a machine learning model such as a linear regression model or a graph neural network model.

[0048] When the function is a machine learning model, the machine learning model is preferably a model capable of calculating the value of the evaluation index, which is learned by a method such as reinforcement learning or supervised learning. Furthermore, the evaluation index value prediction unit 110 may calculate the value of the evaluation index by a simulator, actual measurement, or the like.

[0049] The evaluation index value selection unit 120 has the function of selecting the evaluation index value corresponding to the type of evaluation index and the specified range specified in the input graph from the evaluation index values ​​calculated by the evaluation index value prediction unit 110.

[0050] Specifically, the evaluation index value selection unit 120 selects, from the predicted values ​​calculated by the evaluation index value prediction unit 110, a value that corresponds to a specified range, in which the evaluation index is specified, in the graph input to the evaluation index value prediction unit 110.

[0051] Fig. 3 is an explanatory diagram showing an example of the selection process by the evaluation index value selection unit 120. Fig. 3 shows the selection process when a graph in which "delay" and "bandwidth" are specified as evaluation indexes is input to the evaluation index value prediction unit 110.

[0052] The evaluation index values ​​s1 and s2 enclosed by dashed lines in FIG. 3 are delay values ​​calculated from the designated range candidates C and D, which are ranges in which "delay" may be designated as the evaluation index. The evaluation index value prediction unit 110 predicts the delay by using a function f delay The evaluation index values ​​s1 and s2 are calculated using the above.

[0053] 3 are band values ​​calculated from the designated range candidates E and F, which are ranges in which a "band" may be designated as an evaluation index. The evaluation index value prediction unit 110 calculates a function f corresponding to the "band" type of evaluation index. bandwidth The evaluation index values ​​t1 and t2 are calculated using the above.

[0054] The graphs enclosed by dashed lines in Fig. 3 are graphs in which "delay" and "bandwidth" are specified as the evaluation index. Specifically, the specified range candidate D is a specified range in which "delay" is specified as the evaluation index. Information indicating that the specified range candidate D is a specified range in which "delay" is specified as the evaluation index is added to the value s2 of the evaluation index.

[0055] Moreover, the designated range candidate E is a designated range in which "band" is specified as the evaluation index. Information indicating that the designated range candidate E is a designated range in which "band" is specified as the evaluation index is added to the value t1 of the evaluation index.

[0056] The evaluation index value selection unit 120 refers to the information on the evaluation index and the designated range added to the value of the evaluation index calculated by the evaluation index value prediction unit 110. Next, the evaluation index value selection unit 120 selects, from the evaluation index values ​​s1, s2, t1, and t2, the evaluation index values ​​s2 and t1 that correspond to the designated range candidates D and E, which are the designated ranges given to the graph shown in Fig. 3. Next, the evaluation index value selection unit 120 outputs the selected evaluation index values.

[0057] As described above, the evaluation index value prediction unit 110 may calculate the value of the evaluation index from at least one of the designated range candidates that corresponds to the designated range given to the input graph.

[0058] The value of the evaluation index selected by the evaluation index value selection unit 120 may be determined in advance. For example, the evaluation index value selection unit 120 may select the value of the evaluation index corresponding to the designated range candidate to which a predetermined condition is assigned.

[0059] The evaluation index value integration unit 130 has a function of converting the value of the evaluation index selected by the evaluation index value selection unit 120 into a scalar value.

[0060] Specifically, the evaluation index value integrating unit 130 receives as input the predicted values ​​of the evaluation indexes selected by the evaluation index value selecting unit 120. Next, the evaluation index value integrating unit 130 outputs a single scalar value that represents the overall evaluation value for the graph input to the evaluation index value predicting unit 110.

[0061] Fig. 4 is an explanatory diagram showing an example of conversion processing by the evaluation index value integrating unit 130. Fig. 4 shows processing in which the evaluation index value integrating unit 130 converts the value of the evaluation index selected by the evaluation index value selecting unit 120 into a scalar value.

[0062] 4, the evaluation index value integration unit 130 receives the evaluation index values ​​s2 and t1 as inputs. Next, the evaluation index value integration unit 130 multiplies each value by scalar values ​​w1 and w2, and adds w1s2 and w2t1.

[0063] Next, the evaluation index value integration unit 130 inputs the calculated value w1s2+w2t1 to the overall evaluation index value output device 300. The scalar value w1s2+w2t1 represents the linear sum of s2 and t1. The overall evaluation index value output device 300 outputs the input scalar value as the overall evaluation value for the graph.

[0064] The conversion process to scalar values ​​by the evaluation index value integrating unit 130 may be any process that converts the evaluation index values ​​into predetermined scalar values, even if the evaluation index values ​​are input in any order. In the conversion process by the evaluation index value integrating unit 130, a product, a maximum value, a minimum value, an average value, a variance, etc. may be used in addition to a linear sum.

[0065] Furthermore, the conversion process to a scalar value by the evaluation index value integration unit 130 may be any process that converts the evaluation index value into an arbitrary scalar value when the order in which the evaluation index values ​​are input is fixed.

[0066] As described above, when a graph in which an evaluation index is specified at least in part is input, the evaluation index value prediction unit 110 of this embodiment calculates the value of the evaluation index in one or more ranges in which an evaluation index may be specified in the graph, for each range, using a predetermined function corresponding to the type of evaluation index.

[0067] Furthermore, the evaluation index value selection unit 120 of this embodiment selects, from the calculated values ​​of one or more evaluation indexes, a value that corresponds to a range in which the evaluation index is specified in the graph.

[0068] Furthermore, the evaluation index value integrating unit 130 of this embodiment outputs the scalar value into which the value of the selected evaluation index is converted as the overall evaluation value of the evaluation indexes specified in the input graph. The evaluation index value integrating unit 130 converts the values ​​of the multiple evaluation indexes into a scalar value, for example, by taking the linear sum of the values ​​of the multiple selected evaluation indexes.

[0069] The evaluation index value prediction unit 110, evaluation index value selection unit 120, and evaluation index value integrating unit 130 of this embodiment are realized, for example, by a CPU (Central Processing Unit) of a computer that operates according to an evaluation index value calculation program.

[0070] In the above case, the CPU reads the evaluation index value calculation program from a program recording medium such as a program storage device of the computer. Then, the CPU operates as the evaluation index value prediction unit 110, the evaluation index value selection unit 120, and the evaluation index value integration unit 130 according to the read evaluation index value calculation program.

[0071] [Explanation of operation] Hereinafter, the operation of the evaluation index value calculation device 100 of this embodiment will be described with reference to Fig. 5. Fig. 5 is a flowchart showing the operation of the overall evaluation value output process by the evaluation index value calculation device 100 of this embodiment.

[0072] 5 shows a process in which the evaluation index value calculation device 100 outputs a comprehensive evaluation value of the evaluation indexes from a graph to which the evaluation indexes are assigned. Note that in the following explanation, explanations of matters that have already been explained will be omitted as appropriate.

[0073] First, a graph to which evaluation indexes have been assigned is input from the graph input device 200 to the evaluation index value prediction unit 110 of the evaluation index value calculation device 100 (step S101).

[0074] Next, the evaluation index value prediction unit 110 calculates the evaluation index values ​​of the designated range candidates in the input graph as predicted values ​​for each type of evaluation index assigned to the input graph (step S102). The evaluation index value prediction unit 110 inputs the calculated predicted values ​​to the evaluation index value selection unit 120.

[0075] Next, the evaluation index value selection unit 120 selects, from the predicted values ​​calculated in step S102, a predicted value that corresponds to the specified range in the graph input in step S101 (step S103). The evaluation index value selection unit 120 inputs the selected predicted value to the evaluation index value integrating unit 130.

[0076] Next, the evaluation index value integration unit 130 converts the predicted value selected in step S103 into a scalar value (step S104). Next, the evaluation index value integration unit 130 outputs the scalar value obtained by conversion in step S104 as a comprehensive evaluation value (step S105). After outputting the comprehensive evaluation value, the evaluation index value calculation device 100 ends the comprehensive evaluation value output process.

[0077] [Effect description] The evaluation index value calculation device 100 of this embodiment is a device that calculates a comprehensive evaluation value according to the range of search states in which evaluation indexes are specified and the type of evaluation index.

[0078] The evaluation index value calculation device 100 includes an evaluation index value prediction unit 110 that, when a graph to which evaluation indexes have been assigned is input, calculates the values ​​of the evaluation indexes within the range in which the evaluation indexes in the graph may be specified for each type of evaluation index using a learning model or the like.

[0079] The evaluation index value calculation device 100 also includes an evaluation index value selection unit 120 that selects a predicted value calculated according to an evaluation index specified for a graph.

[0080] The evaluation index value calculation device 100 also includes an evaluation index value integration unit 130 that converts the values ​​of the selected evaluation indexes into scalar values, thereby outputting a comprehensive evaluation value of the evaluation indexes specified in the input graph.

[0081] The evaluation index value calculation device 100 of this embodiment can output a comprehensive evaluation value according to the type and specified range of evaluation indexes assigned to the input graph. This is because the evaluation index value prediction unit 110 calculates a predicted value according to the type of evaluation index specified in the input graph, and the evaluation index value selection unit 120 can select a predicted value that corresponds to the range in which the evaluation index is specified from among the calculated predicted values.

[0082] The evaluation index value calculation device 100 of this embodiment can calculate the value of the changed evaluation index even if the range of evaluation indices or the type of evaluation index specified for a graph representing a molecular compound, an ICT system, etc. is changed at the start of the search.

[0083] When the evaluation index value calculation device 100 is used, particularly in a problem in which the structure of a graph representing a molecular compound, an ICT system, or the like is exploratory, even if the evaluation index referenced during the search is changed, a solution based on the changed evaluation index is derived.

[0084] An industrial application example of this embodiment will be described below. The evaluation index value calculation device 100 of this embodiment can be used to optimize the structure of molecular compounds in the drug discovery industry. Specifically, a graph representing a molecular compound to which an evaluation index is assigned is input to the evaluation index value calculation device 100.

[0085] In addition, the evaluation index value prediction unit 110 uses a graph neural network as a function for calculating the value of the evaluation index, and the evaluation index value integrating unit 130 uses a linear sum for the conversion process to a scalar value, so that the evaluation index value calculation device 100 is used to optimize the structure of a molecular compound.

[0086] In particular, the evaluation index calculation device 100 of this embodiment can be applied to the problem of generating compounds in the field of drug discovery described in Non-Patent Document 1.

[0087] The evaluation index value calculation device 100 of this embodiment can also be used to optimize the structure of a graph representing an ICT system. Specifically, a graph representing an ICT system to which evaluation indexes are assigned is input to the evaluation index value calculation device 100.

[0088] In addition, the evaluation index value prediction unit 110 uses a graph neural network as a function for calculating the value of the evaluation index, and the evaluation index value integration unit 130 uses a linear sum for the conversion process to a scalar value, so that the evaluation index value calculation device 100 is used to optimize the structure of the graph representing the ICT system.

[0089] In particular, the evaluation index value calculation device 100 of this embodiment is applicable to searches in which multiple performance requirements other than functional requirements are treated as evaluation indexes in the search problem described in Non-Patent Document 2.

[0090] 6 is a schematic block diagram showing an example of the configuration of a computer in which the evaluation index value calculation device 100 of this embodiment is implemented. The computer 10 shown in FIG. 6 includes a CPU 11, a main memory device 12, an auxiliary memory device 13, an interface 14, an input device 15, and an output device 16.

[0091] The evaluation index value calculation device 100 of this embodiment is implemented in a computer 10. An evaluation index value calculation program that realizes the operation of the evaluation index value calculation device 100 is stored in an auxiliary storage device 13.

[0092] The CPU 11 reads out the evaluation index value calculation program from the auxiliary storage device 13, loads it into the main storage device 12, and executes the processes described in the above embodiments in accordance with the loaded evaluation index value calculation program. The main storage device 12 is, for example, a RAM (Random Access Memory).

[0093] The auxiliary storage device 13 is, for example, a non-transient tangible medium. Examples of non-transient tangible media include a magnetic disk, a magneto-optical disk, a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), and a semiconductor memory that are connected via an interface 14.

[0094] The input device 15 has a function of inputting data and processing commands, and is, for example, a keyboard or a mouse.

[0095] The output device 16 has a function of outputting data, and is, for example, a display device such as a liquid crystal display device, or a printing device such as a printer.

[0096] Furthermore, when the evaluation index value calculation program is distributed to the computer 10 via a communication line, the computer 10 may load the distributed program into the main storage device 12 and execute the above-mentioned processing.

[0097] In addition, some or all of the components may be realized by a general-purpose or dedicated circuit, a processor, or a combination of these. These may be configured by a single chip, or may be configured by multiple chips connected via a bus. Some or all of the components may be realized by a combination of the above-mentioned circuits, etc. and a program.

[0098] When some or all of the components are realized by a plurality of information processing devices, circuits, etc., the plurality of information processing devices, circuits, etc. may be centrally or distributed. For example, the information processing devices, circuits, etc. may be realized as a client-server system, a cloud computing system, etc., in a form in which each is connected via a communication network.

[0099] Next, an overview of the present invention will be described. Fig. 7 is a block diagram showing an overview of a system verification program generation device according to the present invention. An evaluation index value calculation device 20 according to the present invention includes a calculation unit 21 (e.g., evaluation index value prediction unit 110) that, when a graph in which an evaluation index is specified at least in a part is input, calculates the value of the evaluation index in one or more ranges (e.g., candidate specified ranges) in which the evaluation index may be specified in the graph, for each range, using a predetermined function corresponding to the type of evaluation index, and a selection unit 22 (e.g., evaluation index value selection unit 120) that selects a value corresponding to the range in which the evaluation index is specified in the graph (e.g., the specified range) from the calculated values ​​of the one or more evaluation indexes.

[0100] With such a configuration, the evaluation index value calculation device can output a comprehensive evaluation value according to the range of search states in which the evaluation indexes are specified and the type of evaluation index.

[0101] In addition, the evaluation index value calculation device 20 may be provided with an output section (e.g., evaluation index value integration section 130) that outputs a scalar value converted from the value of the selected evaluation index as an overall evaluation value of the evaluation indexes specified in the input graph.

[0102] The output unit may convert the values ​​of the multiple evaluation indexes into a scalar value by taking a linear sum of the values ​​of the multiple selected evaluation indexes.

[0103] With such a configuration, the evaluation index value calculation device can output a scalar value obtained by converting the value of the evaluation index as a comprehensive evaluation value.

[0104] The predetermined function may be a learning model, and the learning model may be a graph neural network.

[0105] With such a configuration, the evaluation index value calculation device can calculate the value of the evaluation index using a model that has been trained to be able to calculate the value of the evaluation index.

[0106] The graph may also be a graph representing an ICT system.

[0107] With such a configuration, the evaluation index value calculation device is used for the task of optimizing the structure of a graph representing an ICT system. [Explanation of symbols]

[0108] 10. Computers 11 CPU 12 Main storage 13 Auxiliary storage device 14 Interface 15 Input Devices 16 Output Devices 20, 100 Evaluation index value calculation device 21 Calculation section 22 Selection section 110 Evaluation index value prediction unit 120 Evaluation index value selection unit 130 Evaluation index value integration unit 200 Graph input device 300 Comprehensive evaluation index value output device

Claims

1. a calculation unit that, when a graph having an evaluation index specified for at least a part of the graph is input, calculates values ​​of the evaluation index for elements in one or more designated range candidates, which are ranges in which the evaluation index may be specified among elements constituting the graph, for each of the designated range candidates by using a predetermined function corresponding to the type of the evaluation index; a selection unit that selects, from among the calculated values ​​of the one or more evaluation indices, a value calculated from the designated range candidate that corresponds to a designated range in which the evaluation index is designated in the graph. An evaluation index value calculation device comprising:

2. an output unit that outputs a scalar value obtained by converting the value of the selected evaluation index as a comprehensive evaluation value of the evaluation index specified in the input graph; 2. The evaluation index value calculation device according to claim 1.

3. The output unit converts the values ​​of the evaluation indexes into the scalar value by taking a linear sum of the values ​​of the selected evaluation indexes.

3. The evaluation index value calculation device according to claim 2.

4. The predetermined function is a learning model. The evaluation index value calculation device according to claim 1 .

5. The learning model is a graph neural network.

5. The evaluation index value calculation device according to claim 4.

6. The graph represents an ICT (Information and Communication Technology) system. The evaluation index value calculation device according to any one of claims 1 to 5.

7. A computer comprising: When a graph having an evaluation index specified for at least a portion thereof is input, the value of the evaluation index for elements in one or more designated range candidates, which are ranges in which the evaluation index may be specified among the elements constituting the graph, is calculated for each of the designated range candidates using a predetermined function corresponding to the type of the evaluation index; Among the calculated values ​​of the one or more evaluation indices, a value calculated from the designated range candidate corresponding to a designated range in which the evaluation index is designated in the graph is selected.

2. An evaluation index value calculation method comprising the steps of:

8. A computer comprising: The scalar value obtained by converting the value of the selected evaluation index is output as a comprehensive evaluation value of the evaluation index specified in the input graph. The evaluation index value calculation method according to claim 7.

9. On the computer, a calculation process in which, when a graph having an evaluation index specified in at least a part thereof is input, values ​​of the evaluation index relating to elements in one or more designated range candidates, which are ranges in which the evaluation index may be specified among elements constituting the graph, are calculated for each of the designated range candidates by using a predetermined function corresponding to the type of the evaluation index; A selection process for selecting a value calculated from the designated range candidate corresponding to a designated range in which the evaluation index is designated in the graph, from among the calculated values ​​of the one or more evaluation indexes. An evaluation index value calculation program for executing the above.

10. On the computer, Execute an output process to output the scalar value obtained by converting the value of the selected evaluation index as a comprehensive evaluation value of the evaluation index specified in the input graph.

10. The evaluation index value calculation program according to claim 9.

Citation Information

Patent Citations

  • Search method, search program and search apparatus

    JP2019056960A

Cited By

  • Graph information calculation apparatus, graph information calculation method, and graph information calculation program

    JP2022159592A