Information processing apparatus, information processing method, and program

Abstract

To retrieve parameter values for optimization calculation processes further efficiently.SOLUTION: An information processing apparatus comprises an optimization unit, a control unit, and a retrieval unit. The optimization unit performs an optimization calculation process for obtaining an optimal solution by repeatedly performing a solution-finding process for obtaining a solution to an optimization problem using one or more set parameters. The control unit sets a calculation time for the optimization calculation process for each of a plurality of retrieval times obtained by dividing the designated time. The retrieval unit sets parameters for each of the retrieval times that are different from those for the other retrieval times, causes the optimization unit to perform the optimization calculation process within the set calculation time using the set parameters, and repeatedly performs a retrieval process to retrieve optimal values of the parameters using Bayesian optimization by using the optimal solution obtained by the optimization calculation process.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] An embodiment of the present invention relates to an information processing device, an information processing method, and a program. [Background technology]

[0002] Known techniques for solving combinatorial optimization problems include optimization techniques using a simulated bifurcation machine (SBM) and quantum annealers. In these techniques, a search process is sometimes performed to find optimal values ​​for multiple parameters used in the calculation in order to obtain a better solution. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2017-073106 [Patent Document 2] Japanese Patent Application Laid-Open No. 2017-167749 [Patent Document 3] Japanese Patent Application Publication No. 2019-145010 [Patent Document 4] Japanese Patent Publication No. 2022-142914 Summary of the Invention [Problem to be solved by the invention]

[0004] An object of the present invention is to provide an information processing device, an information processing method, and a program that can more efficiently search for parameter values ​​for optimization calculation processing. [Means for solving the problem]

[0005] An information processing device according to an embodiment includes an optimization unit, a control unit, and a search unit. The optimization unit executes an optimization calculation process to find an optimal solution by repeatedly executing a solution-finding process to find a solution to an optimization problem using one or more set parameters. The control unit sets a calculation time for the optimization calculation process for each of a plurality of search times obtained by dividing a specified time. The search unit sets parameters for each of the plurality of search times that are different from those for the other search times, causes the optimization unit to execute the optimization calculation process within the set calculation time using the set parameters, and repeatedly executes a search process to search for optimal values ​​of the parameters through Bayesian optimization using the optimal solution obtained by the optimization calculation process. [Brief explanation of the drawings]

[0006] [Figure 1] FIG. 1 is a block diagram illustrating an example of a configuration of an information processing apparatus according to an embodiment. [Figure 2] FIG. 2 is a flowchart illustrating an example of information processing according to the embodiment. [Figure 3] FIG. 3 is a flowchart showing an example of information processing in the first modification. [Figure 4] FIG. 4 is a flowchart showing an example of information processing in the second modification. [Figure 5] FIG. 5 is an explanatory diagram illustrating an example of the hardware configuration of an information processing device. DETAILED DESCRIPTION OF THE INVENTION

[0007] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of an information processing apparatus according to the present invention will be described in detail below with reference to the accompanying drawings.

[0008] As a search process for searching for optimal values ​​of parameters in an optimization calculation process, for example, a parameter search technique using Bayesian optimization has been proposed. In such parameter search techniques, based on the Bayesian optimization idea of ​​"searching for a probabilistically good point from the observation points," an "evaluation value (observation point)" is compared for a "combination (point) of parameters of the calculation conditions." The parameters of the calculation conditions are then updated according to the Bayesian optimization algorithm, and the evaluation value is acquired again. If the calculation accuracy indicated by the evaluation value improves, the parameters are updated, and if the accuracy does not improve, the parameters are not updated. This process is repeated.

[0009] As an optimization calculation process (a calculation process for solving an optimization problem), a technology has been proposed in which a calculation process for finding a solution to an optimization problem (hereinafter referred to as a solution-finding process) is repeatedly executed within a given calculation time, the solution is updated (converged) so as to approach an optimal solution, and the converged solution is output as the optimal solution. An example of such an optimization calculation process is described below. - Optimization calculation processing using simulated bifurcation machines Optimization calculation processing using quantum annealing

[0010] The parameters for the optimization calculation process are searched for, for example, as follows. The search process is repeated to update the parameters and perform the optimization calculation process. In each search process, an optimization calculation process is performed using the updated parameters, and the solution to be evaluated (optimal solution) is obtained. In one optimization calculation process (search process), multiple solution processes are repeatedly executed, and the converged solution is obtained as the solution to be evaluated (optimal solution).

[0011] The total processing time including multiple search processes is calculated by multiplying the number of search process iterations by the calculation time for each search process. Therefore, in order to reduce the total processing time and search for parameters more efficiently, it is necessary to reduce the number of iterations or the calculation time for each iteration.

[0012] In parameter search technology using Bayesian optimization, the more times the search process is repeated, the higher the probability (search accuracy) of obtaining good points (parameter combinations). Also, if the number of times the search process is repeated is small, there is a possibility that appropriate parameters will not be found, so reducing the number of iterations may not be desirable.

[0013] In this embodiment, parameter search processing using Bayesian optimization can be performed more efficiently within a limited time, and parameters that can provide a solution for the optimization calculation processing with higher accuracy can be searched for. In this embodiment, the optimization calculation processing is performed to repeatedly execute the solution-finding processing as described above to obtain a converged optimal solution.

[0014] 1 is a block diagram showing an example of the configuration of an information processing device 100 according to an embodiment. As shown in FIG. 1, the information processing device 100 includes a reception unit 101, a control unit 102, a search unit 103, an optimization unit 104, an output control unit 105, and a storage unit 121.

[0015] The receiving unit 101 receives input of various information used in the information processing device 100. For example, the receiving unit 101 receives calculation data used in the optimization calculation process by the optimization unit 104 and the search process by the search unit 103. The calculation data may be any data according to the optimization calculation process and search process to be applied, and includes, for example, the following data: Data defining the optimization problem (constraints, etc.) -Data to be used as input for optimization calculation Optimization calculation process parameters Search process parameters

[0016] The parameters of the optimization calculation process input as calculation data are, for example, the following parameters. - The specified time that represents the upper limit of the overall processing time for the optimization calculation process The upper limit of the number of iterations of the solution process that can be repeated within one search process (threshold TH_A)

[0017] The specified time (total processing time of the optimization calculation process) is specified by, for example, a user as an upper limit of the time required to solve the optimization problem. This embodiment can be realized as a form in which a search process (optimization calculation process) is executed multiple times within the specified time, and an optimal solution to the optimization problem is obtained while finding optimal values ​​of parameters, and output to the requester (user, etc.). This can be interpreted as a command from the user to execute a single optimization calculation process, but typically, multiple optimization calculation processes are executed within the information processing device 100 in response to repeated search processes.

[0018] The search process parameters include, for example, the following parameters: Upper limit of the number of search repetitions (number of searches) (threshold TH_B)

[0019] The receiving unit 101 may verify the received data, convert the received data, etc. Data verification is, for example, processing to confirm whether the data is appropriate as a target for optimization calculation processing.

[0020] The control unit 102 controls the parameter search process using Bayesian optimization. For example, the control unit 102 creates a schedule for the search process and controls the search process in accordance with the schedule.

[0021] The control unit 102 first divides the specified time specified by the user or the like into multiple search times. The search time does not correspond to the calculation time per search process, but corresponds to the time required to repeatedly execute multiple search processes. The number of divisions may be any value as long as it is two or more. The following mainly describes an example in which the specified time is divided into two search times.

[0022] The specified time is specified by, for example, a length of time (e.g., 10 seconds). The control unit 102 divides the specified time into two search times, for example, according to a predetermined ratio. In the following, an example in which the specified time is divided at a 1:1 ratio will be described, but the ratio is not limited to 1:1 and may be any value.

[0023] If 10 seconds is specified as the specified time, the control unit 102 divides the specified time of 10 seconds into a first half search time of 5 seconds and a second half search time of 5 seconds. The first half search time and the second half search time correspond to an early search time (first search time) and a late search time (second search time), respectively. Hereinafter, a search time will be referred to as a phase, and the first half search time and the second half search time will sometimes be referred to as a first half phase and a second half phase, respectively.

[0024] Next, the control unit 102 sets the calculation time of the optimization calculation process executed in the search process for each of the multiple search times (phases). For example, the control unit 102 sets a smaller calculation time value for the earlier search time among the multiple search times. When dividing the specified time into two search times (phases), the control unit 102 sets the calculation time value for the first phase to be smaller than the calculation time value for the second phase. For example, if the first and second phases are each 5 seconds, the control unit 102 sets the calculation time for the optimization calculation process of the first phase to 0.1 seconds and the calculation time for the optimization calculation process of the second phase to 2.5 seconds.

[0025] Dividing the specified time into multiple search times and setting a calculation time for each search time corresponds to creating a search process schedule. An example of the effect of setting a smaller calculation time value for a shorter search time will be described later.

[0026] The search unit 103 executes a parameter search process using Bayesian optimization under the control of the control unit 102. For example, the search unit 103 sets different parameters for each of a plurality of search times. The search unit 103 causes the optimization unit 104 to execute an optimization calculation process using the set parameters so as to complete the process within the set calculation time, and repeatedly executes the search process using an optimal solution obtained by the optimization calculation process.

[0027] Note that the parameters searched using Bayesian optimization are different from the parameters of the optimization calculation process input as calculation data. The parameters searched are determined depending on the type of optimization calculation process to be applied. For example, when an optimization calculation process using a simulated bifurcation machine is used, the following parameters correspond to the parameters searched using Bayesian optimization. Time step size for calculation Coefficients for constraints

[0028] The search unit 103 causes the optimization unit 104 to execute the optimization calculation process so as to complete it within the set calculation time within the scheduled search time. For example, if the first half phase is 5 seconds, the search unit 103 causes the optimization unit 104 to execute the optimization calculation process so as to complete it within a calculation time of 0.1 seconds. In this case, the maximum number of iterations of the optimization calculation process (search process) is 50 (5 / 0.1). Furthermore, if the second half phase is 5 seconds, the search unit 103 causes the optimization unit 104 to execute the optimization calculation process so as to complete it within a calculation time of 2.5 seconds. In this case, the maximum number of iterations of the optimization calculation process (search process) is 2 (5 / 2.5).

[0029] A threshold (second threshold) representing an upper limit of the number of searches may be set separately from the number of repetitions determined from the search time and the calculation time. The threshold TH_B described above corresponds to this threshold. For example, when the number of repetitions of the search process within the search time (number of searches) reaches the threshold TH_B, the search unit 103 ends the search process even if it is within the search time, and if there is a next phase (e.g., a latter phase), transitions to processing of the next phase.

[0030] The threshold value TH_B is, for example, a value that is predetermined as the number of searches required to find appropriate parameters. By using the threshold value TH_B, it is possible to suppress repetition of ineffective search processes and search for parameters more efficiently. Different values ​​of the threshold value TH_B may be set for each of a plurality of search times.

[0031] The optimization unit 104 executes the optimization calculation process in accordance with the instructions of the search unit 103. For example, the optimization unit 104 executes the optimization calculation process so as to complete it within a set calculation time using one or more parameters set by the search unit 103. The optimization unit 104 outputs the optimal solution obtained within the set calculation time to the search unit 103.

[0032] The optimization unit 104 may terminate the optimization calculation process when the number of iterations of the solution-finding process reaches a threshold (first threshold). The threshold TH_A described above corresponds to the threshold in this case. For example, when the number of iterations of the solution-finding process within the calculation time of the optimization calculation process reaches the threshold TH_A, the optimization unit 104 terminates the optimization calculation process even if the calculation time is still within the calculation time.

[0033] The threshold value TH_A is, for example, a value that is predetermined as the number of iterations that can find an appropriate optimal solution. By using the threshold value TH_A, it is possible to suppress the repetition of ineffective solution-finding processes and to search for parameters more efficiently. Different values ​​of the threshold value TH_A may be set for each of multiple search times.

[0034] The optimization calculation process by the optimization unit 104 may be executed in parallel by multiple devices according to a specified parallelism level. In such a case, the parallelism level may be included in the parameters to be searched.

[0035] The output control unit 105 controls the output of various information used in the information processing device 100. For example, the output control unit 105 outputs the optimal solution obtained by the optimization unit 104 as a processing result. The output control unit 105 may output the processing result in a format that can be interpreted by a user who requested the execution of the optimization calculation process, or the like.

[0036] At least a part of each of the above units (reception unit 101, control unit 102, search unit 103, optimization unit 104, and output control unit 105) may be realized by one or more processing units. Each of the above units is realized, for example, by one or more processors. For example, each of the above units may be realized by having a processor such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit) execute a program, that is, by software. Each of the above units may be realized by a processor such as a dedicated IC (Integrated Circuit), that is, by hardware. Each of the above units may be realized by a combination of software and hardware. When multiple processors are used, each processor may realize one of the units, or may realize two or more of the units.

[0037] The storage unit 121 stores various types of information used in the information processing device. For example, the storage unit 121 stores various types of information (such as calculation data) accepted by the accepting unit 101 and information (such as intermediate data and processing results) output by the processes of each unit.

[0038] The storage unit 121 can be configured from any commonly used storage medium such as a flash memory, a memory card, a RAM (Random Access Memory), an HDD (Hard Disk Drive), and an optical disk.

[0039] The information processing device 100 may be physically configured as one device, or may be physically configured as multiple devices. For example, the information processing device 100 may be built on a cloud environment. Furthermore, each unit in the information processing device 100 may be distributed and provided in multiple devices. For example, the optimization unit 104 may be provided in a device (such as another cloud) different from the other components.

[0040] Next, information processing by the information processing device 100 of the embodiment will be described. The information processing is processing for finding an optimal solution to an optimization problem requested by a user or the like. In the information processing, optimal values ​​of parameters are searched for by a search processing, while an optimal solution is calculated by an optimization calculation processing. Figure 2 is a flowchart showing an example of information processing in the embodiment.

[0041] The receiving unit 101 receives calculation data input by a user or the like (step S101). The calculation data includes the specified time and the like, as described above.

[0042] The control unit 102 divides the specified time into a plurality of phases and determines a schedule for parameter search (step S102). For example, the control unit 102 divides the specified time of 10 seconds into a first half phase of 5 seconds and a second half phase of 5 seconds.

[0043] The control unit 102 determines whether the phase has changed (step S103). For example, the control unit 102 determines that the phase has changed at the first processing after dividing into a plurality of phases or when a search time corresponding to each phase has elapsed.

[0044] If the phase is changed (step S103: Yes), the control unit 102 sets the calculation time corresponding to the current phase (the changed phase) (step S104). For example, if the current phase is the first half phase, the control unit 102 sets the calculation time to 0.1 seconds, and if the current phase is the second half phase, the control unit 102 sets the calculation time to 2.5 seconds.

[0045] After the calculation time is set (step S104), and if the phase has not been changed (step S103: No), the control unit 102 causes the search unit 103 to execute a search process in accordance with the determined schedule (search time) and the set calculation time (steps S105 to S109).

[0046] First, the search unit 103 determines parameters for the optimization calculation process by Bayesian optimization (step S105). In the first iteration, the search unit 103 determines the parameters to be initial values ​​(for example, random values). In the second iteration and thereafter, the parameters for the optimization calculation process are updated according to Bayesian optimization using the evaluation value of the optimal solution obtained up to step S109 so as to improve the accuracy of the evaluation value.

[0047] The optimization unit 104 executes the optimization calculation process using the determined parameters (step S106). Note that steps S106 to S108 correspond to one iteration of the solution-finding process that is repeatedly executed in the optimization calculation process. That is, steps S106 to S108 are repeatedly executed in the optimization calculation process.

[0048] The optimization unit 104 determines whether the number of calculations in the repeatedly executed solution-finding process has reached a threshold value TH_A (step S107). If the number of calculations has reached the threshold value TH_A (step S107: Yes), the optimization unit 104 ends the optimization calculation process and returns to step S105.

[0049] If the number of calculations has not reached the threshold value TH_A (step S107: No), the optimization unit 104 determines whether the calculation time for the optimization calculation process has been reached (step S108). If the calculation time for the optimization calculation process has not been reached (step S108: No), the optimization unit 104 returns to step S106 and repeats the solution-finding process.

[0050] If the calculation time for the optimization calculation process has been reached (step S108: Yes), the search unit 103 determines whether the number of searches has reached a threshold TH_B (step S109). If the number of searches has not reached the threshold TH_B (step S109: No), the search unit 103 returns to step S105 and repeats the process.

[0051] If the number of searches reaches the threshold value TH_B (step S109: Yes), the control unit 102 determines whether the elapsed time from the start of processing reaches a specified time (step S110). The start of processing refers to, for example, the start of the first search processing (start of step S103). If the elapsed time has not reached the specified time (step S110: No), the control unit 102 returns to step S103 and repeats the processing.

[0052] If the elapsed time reaches the designated time (step S110: Yes), the output control unit 105 outputs the processing result (step S111) and ends the information processing.

[0053] Next, an example of the effect of setting a smaller value for the calculation time of the optimization calculation process as the search time becomes shorter will be described.

[0054] Decreasing the calculation time value of the optimization calculation process (e.g., 0.1 seconds) means decreasing the number of iterations of the solution-finding process executed in the optimization calculation process. Conversely, increasing the calculation time value (e.g., 2.5 seconds) means increasing the number of iterations of the solution-finding process executed in the optimization calculation process.

[0055] Depending on the optimization calculation process used, the earlier the solution is found, the faster the solution converges, and the slower the solution converges in the later stages. Therefore, in stages where the solution converges quickly, as many search processes as possible (e.g., 50 times) are performed in a short calculation time, enabling a rough search for the optimal parameter values. In stages where the solution converges slowly, fewer search processes (e.g., two times) are performed in a longer calculation time, enabling a more accurate search for the optimal parameter values.

[0056] In the parameter search process using Bayesian optimization, it is sufficient to be able to compare the evaluation values ​​of the solutions as described above, and it is not necessary to obtain a highly accurate solution. Also, as described above, if the optimization calculation process has a fast convergence rate at an early stage of the iterations, a solution with acceptable accuracy can be obtained even with a short calculation time. Therefore, the faster the search time, the more likely it is that parameters that can obtain a more accurate solution from the optimization calculation process can be searched for, even if the calculation time for the optimization calculation process is set to a small value.

[0057] As described above, by setting a smaller value for the calculation time of the optimization calculation process for faster search times, the average calculation time per search process can be reduced without reducing the number of searches, making it possible to search for parameter values ​​more efficiently.

[0058] In the above example, the control unit 102 divided the specified time into two search times (first phase, second phase) and set a smaller calculation time for the first phase than for the second phase. Similarly, when dividing the specified time into three or more phases, the control unit 102 may set a smaller calculation time value for the optimization calculation process for shorter search times. For example, the control unit 102 may divide the specified time into three phases, the first phase, the second phase, and the third phase, in order from earliest to latest, and set the calculation time for each phase so that the first phase, the second phase, and the third phase have smaller values ​​in that order.

[0059] The method for setting the calculation time is not limited to a method in which the calculation time value of the optimization calculation process is reduced as the search time is faster. For example, a method in which the calculation time is set according to the convergence speed of the solution of the applied optimization calculation process (solution-finding process) may be used. For example, the control unit 102 may set a small calculation time value for a search time (phase) corresponding to a stage in which the convergence speed of the solution is fast, and set a large calculation time value for a search time (phase) corresponding to a stage in which the convergence speed of the solution is slow.

[0060] (Variation 1) In Modification 1, when the number of times that the solution is not updated by the solution-finding process reaches a specified value (for example, 100 times), the optimization unit 104 terminates the optimization calculation process even if the calculation time is still within the time limit. This process prevents the repetition of ineffective solution-finding processes, enabling more efficient parameter search. Different specified values ​​may be set for each of multiple search times.

[0061] 3 is a flowchart showing an example of information processing in Modification 1. This modification differs from the information processing in the above embodiment (FIG. 2) in that step S207 is added. Steps S201 to S206 are the same as steps S101 to S106 in the above embodiment, and steps S208 to S212 are the same as steps S107 to S111 in the above embodiment, so their description will be omitted.

[0062] The optimization unit 104 determines whether the number of times the solution is not updated by the solution-finding process has reached a specified value (step S207). If the number of times the solution is not updated has not reached the specified value (step S207: No), the optimization unit 104 continues the process from step S208 onwards. If the number of times the solution is not updated has reached the specified value (step S207: Yes), the optimization unit 104 ends the optimization calculation process and returns to step S205.

[0063] In this way, in this modification, even if the calculation time has not yet been reached, if the solution is not updated, the optimization calculation process (search process) can be terminated, and the search time can be further shortened. Therefore, the number of searches can be increased beyond the number of searches expected from the length of each phase (search time) and the calculation time.

[0064] (Variation 2) In addition to the function (as in the above embodiment) of dividing a specified time into multiple search times (phases) and setting a calculation time for each phase, Modification 1 also has a function of terminating (aborting) the optimization calculation process depending on the number of times the solution is not updated. Even with only the latter function (the function of terminating the optimization calculation process depending on the number of times the solution is not updated), parameter values ​​can be searched for more efficiently. In Modification 2, an example configured in this way will be described.

[0065] In Modification 2, the control unit 102 does not have a function to divide the specified time into multiple search times (phases), but sets the calculation time for the specified time as one search time. For example, the control unit 102 sets the calculation time to 1 second for a specified time (search time) of 10 seconds.

[0066] The search unit 103 of this modification causes the optimization unit 104 to execute the optimization calculation process so that it is completed within a set calculation time. The search unit 103 of this modification has the same functions as those of the above embodiment, except that it executes the search process for one search time instead of multiple search times.

[0067] Similar to the first modification, the optimization unit 104 of this modification further has a function of terminating the optimization calculation process when the number of times the solution is not updated by the solution-finding process reaches a specified value, even if the calculation time is still within the limit.

[0068] 4 is a flowchart showing an example of information processing in Modification 2. Step S301 is the same processing as step S201 in Modification 1, and therefore a description thereof will be omitted.

[0069] In the second modification, the control unit 102 sets a calculation time for one search time (specified time) (step S302).

[0070] Steps S303 to S308 are the same as steps S205 to S210 in the first modification, and therefore a description thereof will be omitted.

[0071] If the number of searches reaches the threshold value TH_B (step S308: Yes), the control unit 102 determines whether the elapsed time from the start of the process reaches the specified time (step S309). If the elapsed time does not reach the specified time (step S309: No), the control unit 102 returns to step S303 and repeats the process.

[0072] Step S310 is the same process as step S212 in the first modification, and therefore a description thereof will be omitted.

[0073] In step S305, similarly to step S207 in Modification 1, it is determined whether the number of times the solution is not updated reaches a specified value (step S305). If the specified value is reached (step S305: Yes), the optimization unit 104 terminates the optimization calculation process even if the calculation time is still within the specified value. For example, if the solution converges before the set calculation time of 1 second (for example, 0.1 seconds), the optimization calculation process (search process) can be terminated midway through the calculation time, and the number of searches can be increased accordingly.

[0074] As described above, according to the embodiment, parameter values ​​for the optimization calculation process can be searched for more efficiently.

[0075] Next, the hardware configuration of the information processing apparatus according to the embodiment will be described with reference to Fig. 5. Fig. 5 is an explanatory diagram showing an example of the hardware configuration of the information processing apparatus according to the embodiment (and the modified example).

[0076] The information processing apparatus of the embodiment includes control devices such as a CPU (Central Processing Unit) 51-1 and a GPU (Graphics Processing Unit) 51-2, storage devices such as a ROM (Read Only Memory) 52 and a RAM (Random Access Memory) 53, a communication I / F 54 that connects to a network and performs communication, and a bus 61 that connects each unit. Note that the CPU 51-1 and the GPU 51-2 may be integrated into one unit that has the functions of both.

[0077] The programs executed by the information processing apparatus according to the embodiment are provided in advance in the ROM 52 or the like.

[0078] The program executed by the information processing device of the embodiment may be configured to be provided as a computer program product by being recorded in an installable or executable format on a computer-readable recording medium such as a CD-ROM (Compact Disk Read Only Memory), a flexible disk (FD), a CD-R (Compact Disk Recordable), or a DVD (Digital Versatile Disk).

[0079] Furthermore, the program executed by the information processing apparatus of the embodiment may be stored on a computer connected to a network such as the Internet and provided by being downloaded via the network. Also, the program executed by the information processing apparatus of the embodiment may be provided or distributed via a network such as the Internet.

[0080] The programs executed by the information processing device of the embodiment can cause a computer to function as each unit of the information processing device described above. In this computer, the CPU 51-1 and the GPU 51-2 can read the programs from a computer-readable storage medium onto a main storage device and execute the programs.

[0081] Although several embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be embodied in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included within the scope and spirit of the invention, and are also included in the scope of the invention and its equivalents as defined in the claims. [Explanation of symbols]

[0082] 100 Information processing device 101 Reception 102 Control section 103 Search Department 104 Optimization Department 105 Output control section 121 Storage section

Claims

1. an optimization unit that executes an optimization calculation process to obtain an optimal solution by repeatedly executing a solution-finding process to obtain a solution to an optimization problem using one or more set parameters; a control unit that sets a calculation time for the optimization calculation process for each of a plurality of search times obtained by dividing a designated time; a search unit that sets, for each of the plurality of search times, the parameters different from those of the other search times, causes the optimization unit to execute the optimization calculation process within the set calculation time using the set parameters, and repeatedly executes a search process that searches for optimal values ​​of the parameters by Bayesian optimization using the optimal solution obtained by the optimization calculation process; An information processing device comprising:

2. the control unit sets the value of the calculation time to a smaller value as the search time among the plurality of search times becomes shorter; The information processing device according to claim 1 .

3. the control unit sets a value of the calculation time for a first search time corresponding to an earlier time and a value of the calculation time for a second search time corresponding to a later time out of the two search times obtained by dividing the specified time into two, the first search time corresponding to an earlier time and the second search time corresponding to a later time, to be smaller than the value of the calculation time for the second search time. The information processing device according to claim 1 .

4. the optimization unit terminates the optimization calculation process when the number of times the solution is not updated by the solution-finding process reaches a specified value. The information processing device according to claim 1 .

5. the optimization unit terminates the optimization calculation process when the number of iterations of the solution-finding process reaches a first threshold value. The information processing device according to claim 1 .

6. the search unit terminates the search process when a search count representing the number of times the search process has been repeated reaches a second threshold value. The information processing device according to claim 1 .

7. an optimization unit that executes an optimization calculation process to obtain an optimal solution by repeatedly executing a solution-finding process to obtain a solution to an optimization problem using one or more set parameters; a search unit that sets different parameters for each repetition, causes the optimization unit to execute the optimization calculation process using the set parameters, and repeatedly executes a search process to search for optimal values ​​of the parameters by Bayesian optimization using the optimal solution obtained by the optimization calculation process, the optimization unit terminates the optimization calculation process when the number of times the solution is not updated by the solution-finding process reaches a specified value. Information processing device.

8. An information processing method executed by an information processing device, an optimization step of executing an optimization calculation process to obtain an optimal solution by repeatedly executing a solution-finding process to obtain a solution to the optimization problem using one or more set parameters; a control step of setting a calculation time for the optimization calculation process for each of a plurality of search times obtained by dividing a designated time; a search step of repeatedly executing a search process in which, for each of the plurality of search times, the parameters different from those of the other search times are set, the optimization calculation process is executed within the set calculation time using the set parameters, and the optimal solution obtained by the optimization calculation process is used to search for optimal values ​​of the parameters by Bayesian optimization; An information processing method including:

9. On the computer, an optimization step of executing an optimization calculation process to obtain an optimal solution by repeatedly executing a solution-finding process to obtain a solution to the optimization problem using one or more set parameters; a control step of setting a calculation time for the optimization calculation process for each of a plurality of search times obtained by dividing a designated time; a search step of repeatedly executing a search process in which, for each of the plurality of search times, the parameters different from those of the other search times are set, the optimization calculation process is executed within the set calculation time using the set parameters, and the optimal solution obtained by the optimization calculation process is used to search for optimal values ​​of the parameters by Bayesian optimization; A program to execute.

Images