Data processing apparatus, method, and computer-readable recording medium storing program

By generating and calculating the transition destination candidates of multi-valued variables in the data processing device, the calculation complexity problem of Ising machine in the case of multi-valued variables is solved, efficient parallel experimental calculation is realized, and the calculation of combination optimization problems is improved efficiency.

CN119004938APending Publication Date: 2024-11-22FUJITSU LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410555303.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2023-05-22
Filing Date
2024-05-07
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

When searching for solutions to the allocation problem using Ising machine, the number of instances of multi-valued variables increases, resulting in increased computational complexity, making it difficult to conduct parallel experiments, affecting the calculation efficiency.

Method used

A data processing device is designed to store evaluation function information and generate transformation destination candidates for multi-valued variables using the processor, calculate the change amount of relevant evaluation functions, specify the multi-valued variable that accepts the transformation, and convert its value into candidate values. , realize parallel experimental calculation.

Benefits of technology

Through parallel experimental calculations, the calculation efficiency of the combination optimization problem represented by multi-valued variables is improved, the memory capacity requirement is reduced, and the data movement complexity is simplified.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119004938A_ABST
    Figure CN119004938A_ABST
Patent Text Reader

Abstract

The present invention relates to a data processing apparatus, a method, and a computer-readable recording medium storing a program. The data processing apparatus includes: a memory configured to store evaluation function information indicating an evaluation function of a combinatorial optimization problem represented by using a plurality of multivalued variables; and a processor coupled to the memory, the processor configured to perform a process including: generating transition destination candidates that are part of respective transitionable ranges of values of the plurality of multivalued variables based on the evaluation function information; calculating, for each of the plurality of multi-valued variables, an amount of change in a value of an evaluation function associated with the transition to the transition destination candidate based on the evaluation function information; specifying a multi-valued variable that accepts a transition from among a plurality of multi-valued variables based on the amount of change; and converting the value of the specified multi-valued variable into the value of the transition destination candidate.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments discussed herein relate to a data processing apparatus, a non-transitory computer-readable recording medium storing a program, and a data processing method. Background Art

[0002] There is an Ising machine using an Ising-type evaluation function as a device for calculating large-scale discrete optimization problems that von Neumann computers are not good at. In the case of using an Ising machine, the combinatorial optimization problem is converted into an Ising model representing the spin behavior of a magnetic material. The Ising machine then searches for the state of the Ising model in which the value of the Ising-type evaluation function is a local minimum through a Markov chain Monte Carlo method, such as a simulated annealing method or a replica exchange method (also known as a parallel tempering method, etc.). The state that reaches the minimum value of the local minimum value of the evaluation function is regarded as the optimal solution. The state of the Ising model can be represented by a combination of the values ​​of multiple state variables. As the value of each state variable in the state variable, 0 or 1 can be used.

[0003] For example, the Ising type evaluation function is defined by the following formula (1).

[0004] [Mathematical formula 1]

[0005]

[0006] The first term on the right side is used to combine the product of the values ​​of two state variables (0 or 1) and the weight value (indicating the strength of the correlation between the two state variables) for all combinations of all state variables of the Ising model without omission or repetition. Let the state variable with i as the identification number be represented by x i Indicates that j is used as the state variable of the identification number by x j The weight value indicating the magnitude of the correlation between the state variables that make i and j the identification numbers is represented by W ij The second term on the right side is obtained by adding the product of the bias coefficient and the state variable for each identification number. The bias coefficient for identification number = i is given by b i instruct.

[0007] In addition, the following formula (2) represents the relationship between x i The change in the value of the evaluation function associated with the change in the value of i ).

[0008] [Mathematical formula 2]

[0009]

[0010] In formula (2), when x i When Δx changes from 1 to 0, iis -1; and when the state variable x i When Δx changes from 0 to 1, i is 1. Note that h i is called the local domain, and ΔE i is by adding h i Multiply by Δx i The sign (+1 or -1) of

[0011] For example, when ΔE i When the noise value is less than that obtained based on the value of the random number and the temperature parameter, the Ising machine performs the operation by repeatedly inverting x i The solution is searched by the process of and updating the local domain.

[0012] Since ΔE i Calculation, determine whether to reverse x i Multiple processing of the values ​​of can be performed in parallel, so multiple state variables can be tested in parallel.

[0013] Incidentally, some combinatorial optimization problems, such as allocation problems, can be represented using multi-valued variables. In the allocation problem, the allocation state indicating to which allocation destination a specific element is allocated can be represented by a multi-valued variable. In the past, a technique for searching for a solution to the allocation problem using an Ising machine has been proposed.

[0014] Examples of the related art include Japanese Laid-Open Patent Publication No. 2023-1055 and Japanese Laid-Open Patent Publication No. 2022-165250 disclosed as the related art. Summary of the invention

[0015] Technical issues

[0016] In the technique of searching for a solution to the assignment problem by using an Ising machine, since the assignment state as a multi-valued variable is converted into x 0 or 1 i , so x i The number of instances of increases sometimes. i The number of instances increases when the i Calculation of ΔE i The movement of data becomes complicated and in some cases parallel experiments become difficult.

[0017] In one mode, an embodiment aims to enable combinatorial optimization problems represented by multi-valued variables to be computed by parallel trials.

[0018] Solution to the problem

[0019] According to one aspect of an embodiment, a data processing device is provided, which includes: a memory configured to store evaluation function information indicating an evaluation function of a combinatorial optimization problem represented by using multiple multi-valued variables; and a processor coupled to the memory, which is configured to perform processing including: based on the evaluation function information, generating a transition destination candidate as a part of a corresponding transformable range of values ​​of multiple multi-valued variables; based on the evaluation function information, for each of the multiple multi-valued variables, calculating a change in the value of the evaluation function associated with the transition to the transition destination candidate; based on the change, specifying a multi-valued variable to be transformed from the multiple multi-valued variables; and transforming the value of the specified multi-valued variable into the value of the transition destination candidate.

[0020] Beneficial effects of the present invention

[0021] In one mode, according to an embodiment, a combinatorial optimization problem represented by multi-valued variables may be calculated by parallel trials. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a diagram showing an exemplary data processing apparatus according to the first embodiment;

[0023] Figure 2 is a diagram showing an example of transition destination candidates of n multi-valued variables;

[0024] Figure 3 is a diagram showing an exemplary data processing apparatus according to the second embodiment;

[0025] Figure 4 is a diagram showing an example of a current state and states of transition destination candidates;

[0026] Figure 5 is a diagram of an example of matrix elements to be read from F memory;

[0027] Figure 6 is a diagram showing an example of matrix elements to be read from the D memory;

[0028] Figure 7 is a timing chart showing an operation example of the data processing device of the second embodiment;

[0029] Figure 8 is a flowchart showing an example of a processing procedure of a data processing device according to the second embodiment;

[0030] Fig. 9 is a diagram showing an exemplary data processing apparatus according to a third embodiment;

[0031] Fig.10is a timing chart showing an operation example of the data processing apparatus of the third embodiment;

[0032] Fig.11 is a diagram showing an example of current states of a plurality of transition destination candidates and states of the transition destination candidates;

[0033] Fig.12 is a diagram showing an example of matrix elements to be read from the D memory;

[0034] Fig.13 is a diagram showing an exemplary data processing apparatus according to a fifth embodiment;

[0035] Fig.14 is a diagram showing an example of matrix elements to be read from a ΔD memory;

[0036] Fig.15 is a timing chart showing an operation example of the data processing apparatus of the fifth embodiment; and

[0037] Fig.16 is a diagram showing an example of hardware of a computer as an example of a data processing device. DETAILED DESCRIPTION

[0038] Hereinafter, modes for carrying out the embodiments will be described with reference to the drawings.

[0039] The data processing apparatus according to each embodiment described below searches for a solution to a combinatorial optimization problem represented by a plurality of multi-valued variables. As one of such combinatorial optimization problems, there is an assignment problem such as a quadratic assignment problem (QAP). The assignment problem is a non-deterministic polynomial time (NP)-hard combinatorial optimization problem that has various real-world applications such as vehicle assignment routing and block placement of a field programmable gate array (FPGA). The assignment problem is represented using a flow matrix and a distance matrix.

[0040] In the QAP for allocating n elements to n allocation destinations, the flow matrix and the distance matrix are represented by the following formulas (3) and (4).

[0041] [Mathematical formula 3]

[0042]

[0043] [Mathematical formula 4]

[0044]

[0045] The flow matrix (F) is a matrix with n rows and n columns. The matrix element at the i-th row and j-th column is represented by f ij It means that f ijIndicates the flow amount (eg, cost, such as the amount of supply transportation between facilities) between elements (eg, facilities) having identification numbers = i and j.

[0046] The distance matrix (D) is a matrix with n rows and n columns. The distance between the kth row and the lth column is d kl Indicates that d kl Indicates the distance between allocation destinations (for example, places where facilities are arranged) having identification numbers = k and l.

[0047] The calculation of the QAP is performed by searching for an assignment that minimizes the value of an evaluation function represented by the following formula (5).

[0048] [Mathematical formula 5]

[0049]

[0050] In formula (5), φ i (i=1 to n) is a multi-valued variable indicating the allocation destination of the element with identification number=i. i φ j The item obtained by appending to d represents the distance between the allocation destination of the element with identification number = i and the allocation destination of the element with identification number = j. i The term obtained by appending to b represents the bias coefficient when the element with identification number=i is assigned to φi. Note that when the sign of the evaluation function is reversed, the calculation of QAP is performed by searching for a state in which the value of the evaluation function is a local maximum.

[0051] (First Embodiment)

[0052] Figure 1 is a diagram showing an exemplary data processing device according to the first embodiment.

[0053] The data processing device 10 according to the first embodiment includes a storage unit 11 and a processing unit 12 .

[0054] The storage unit 11 may include a volatile semiconductor memory such as a random access memory (RAM), or may include a nonvolatile storage device such as a hard disk drive (HDD) or a flash memory. In addition, the storage unit 11 may include both a volatile semiconductor memory and a nonvolatile storage device.

[0055] The storage unit 11 stores evaluation function information 11a about an evaluation function of a combinatorial optimization problem represented by a plurality of (hereinafter assumed to be n) multi-valued variables. The evaluation function information 11a includes information such as a transition range of the value of each of the n multi-valued variables and a coefficient included in the evaluation function. Each of the n multi-valued variables may be an integer variable or a continuous variable. In addition, the n multi-valued variables may have different transition ranges from each other.

[0056] Figure 1 It shows that φ 1 To φ n As an example of n multi-valued variables. Subscripts 1 to n represent the identification numbers of the multi-valued variables. Figure 1 In the example, φ i is a multi-valued variable with identification number = i, φ i is an integer variable, and φ i The transformable range of has m values ​​of 1, 2, ..., and m.

[0057] For example, as coefficients included in the evaluation function, there are matrix elements (flow rate and distance) of the flow matrix and the distance matrix as indicated by formulas (3) and (4).

[0058] Note that the storage unit 11 may also store calculation conditions for the combinatorial optimization problem. For example, in the case of performing a search by the replica exchange method, the calculation conditions include the number of replicas, the value of a temperature parameter to be set in each replica, and the like.

[0059] For example, the processing unit 12 can be implemented by an electronic circuit, such as an application specific integrated circuit (ASIC) or an FPGA. However, the processing unit 12 can also be implemented by a processor such as a central processing unit (CPU), a graphics processing unit (GPU) or a digital signal processor (DSP). The processor executes a program stored in a memory such as a RAM (which can be the storage unit 11). A collection of processors can be referred to as a "multiprocessor" or simply a "processor". In addition, the processing unit 12 may include a processor and an electronic circuit, such as an ASIC or an FPGA.

[0060] Figure 1 An example of the procedure of processing performed by the processing unit 12 is shown.

[0061] The processing unit 12 generates a transition destination candidate that is a part of a value transitionable range for each of the n multi-valued variables based on the evaluation function information 11 a (step S1 ).

[0062] For example, the processing unit 12 randomly generates a transition destination candidate from within the transition range for each of the n multi-valued variables. The processing unit 12 may randomly generate two or more transition destination candidates from within the transition range for each of the n multi-valued variables (refer to the fourth embodiment described later).

[0063] When the combinatorial optimization problem has constraints, the transition destination candidates are generated so as to satisfy the constraints. For example, the transition destination candidates may be randomly generated for the multi-valued variables for which transition destination candidates that do not satisfy the constraints have been generated until the constraints are satisfied.

[0064] Note that the processing unit 12 may generate transition destination candidates not randomly but according to a predetermined rule. Figure 1 In the example, m is generated as φ i The transition destination candidate φ' i .

[0065] Figure 2 15 is a diagram showing an example of transition destination candidates for n multi-valued variables. The horizontal axis represents the identification number of the multi-valued variable, and the vertical axis represents the value of the multi-valued variable. The graph group 15a represents the current value or initial value of the n multi-valued variables. The graph group 15b represents the value of the transition destination candidate for the initial value or current value of the n multi-valued variables.

[0066] exist Figure 2 In the example of , each multi-valued variable has two possible values, and the transition destination for the current value of each multi-valued variable is narrowed down to one.

[0067] After the process of step S1 , the processing unit 12 calculates the amount of change in the value of the evaluation function associated with transition to the transition destination candidate for each of the multi-valued variables based on the evaluation function information 11 a (step S2 ).

[0068] In the case where the evaluation function is expressed by formula (5), i Transformed into φ' i The amount of change (ΔE(φ i →φ' i )) can be expressed by the following formula (6).

[0069] [Mathematical formula 6]

[0070]

[0071] In formula (6), the subscript φ' i φ jThe d represents the distance between the current allocation destination of the element with identification number = j and the allocation destination candidate of the element with identification number = i. The current allocation destination of the element with identification number = j can be represented by φ j , and the allocation destination candidate of the element with identification number = i can be represented by φ' i The distance between the allocation destination of the element with identification number = j and the current allocation destination of the element with identification number = i is represented by a subscript φ i φ j The allocation destination of the element with identification number = i can be represented by φ i When the element with identification number = i is assigned to φ' i The bias coefficient is represented by the subscript iφ' i When the element with identification number = i is assigned to φ i The bias coefficient is represented by the subscript iφ i The b represents.

[0072] After the processing in step S2, the processing unit 12 specifies a multi-value variable that accepts transition to a transition destination candidate based on the calculated change amount (step S3). For example, the processing unit 12 determines whether to accept the transition from φ based on the comparison result between the change amount and a predetermined value. i to φ' i The predetermined value is, for example, a noise value obtained based on a random number and a value of a temperature parameter. i When the amount of change associated with the transition is less than log(rand)×T, the processing unit 12 accepts the transition, which is an example of a noise value obtained based on a uniform random number (rand) equal to or greater than zero but equal to or less than one and a temperature parameter (T).

[0073] Note that the processing unit 12 may accept a transition of the multi-valued variable that achieves a transition that gives a minimum value of the evaluation function based on the calculated amount of change.

[0074] The processing unit 12 causes the value of the multi-value variable subjected to transition to be transitioned to the value of the transition destination candidate (step S4 ). Figure 1 It shows that i The value of is converted into φ' as a candidate for the transition destination i Examples of values ​​of . i = m, φ i The value of changes to m. Note that at this time, φ i The original value of is adopted as φ' as the candidate for the transition destination i The value of .

[0075] In the case where there are a plurality of multi-valued variables that have been subjected to the transition, the processing unit 12 may select one of the multi-valued variables. In this case, for example, the processing unit 12 selects a multi-valued variable with which the value of the evaluation function becomes minimum due to the amount of change associated with the transition from among the plurality of multi-valued variables that have been subjected to the transition. The processing unit 12 may randomly select a multi-valued variable to be transformed from among the plurality of multi-valued variables that have been subjected to the transition.

[0076] After the processing of step S4, the processing from step S2 is repeated. At this time, in the processing in step S2, the processing unit 12 can calculate the difference in the amount of change of the evaluation function associated with the transition through the processing in step S4, and update the amount of change using the calculated difference. In the case of j≠i, when φ j Transformed into φ' j When, with the transformation φ i →φ' i The difference in the associated change amounts is expressed by the following formula (7).

[0077] [Mathematical formula 7]

[0078]

[0079] The processing unit 12 may read the matrix elements of the flow matrix and the distance matrix included in the formula (7) from the storage unit 11 to calculate δΔE(φ i →φ' i ).

[0080] Note that in the case of j = i, with the transition φ i →φ' i The difference in the associated change amounts is expressed by the following formula (8).

[0081] [Mathematical formula 8]

[0082] δΔE(φ′ i →φ i )=-2ΔE(φ i →φ′ i ) (8)

[0083] The processing unit 12 may use the difference indicated by the formula (7) to update the change amount in the case where j≠i according to the following formula (9).

[0084] [Mathematical formula 9]

[0085]

[0086] The processing unit 12 may use the difference indicated by the formula (8) to update the change amount in the case of j=i according to the following formula (10).

[0087] [Mathematical formula 10]

[0088] ΔE(φ′ i →φ i )←-ΔE(φ j →φ′ j ) (10)

[0089] Note that formula (9) can also be expressed as the following formula (11).

[0090] [Mathematical formula 11]

[0091]

[0092] In formula (11), when assuming k = φ j or k = φv j When , the Δd in the brackets of the second term on the right side is expressed by the following formula (12).

[0093] [Mathematical formula 12]

[0094]

[0095] In the second embodiment, a method for efficiently calculating δΔE(φ i →φ′ i ) configuration example.

[0096] For n multi-valued variables, the processing in steps S2 to S4 can be performed in parallel. That is, the processing in steps S2 to S4 can be performed by a machine (in Figure 1 However, the processing in steps S2 to S4 may be performed continuously on n multi-valued variables.

[0097] Note that whenever the processing in step S3 is repeated a predetermined number of times, the processing in step S1 as described above may be performed again, and the transition destination candidate for each multi-valued variable may be regenerated. Alternatively, even if the processing in step S3 is repeated a predetermined number of times, but the minimum value of the evaluation function is not updated, the transition destination candidate for each multi-valued variable may be regenerated.

[0098] In the case of executing the simulated annealing method, the processing unit 12 reduces the value of the temperature parameter (T) described above according to a predetermined temperature parameter switching schedule while repeating the processing in steps S2 to S4. Then, for example, when a predetermined search end condition is met, the processing unit 12 outputs the values ​​of the n multi-valued variables as the search result of the solution of the combinatorial optimization problem (for example, the values ​​of the n multi-valued variables are displayed on a display device (not shown)). Note that the processing unit 12 can update the value of the evaluation function represented by formula (5) whenever a transition is accepted, and save the minimum value until then and the values ​​of the n multi-valued variables when the minimum value is obtained. In this case, the processing unit 12 can output the values ​​of the n multi-valued variables corresponding to the minimum value stored when the search end condition is met as the search result.

[0099] In the case where the processing unit 12 executes the replica exchange method, the processing unit 12 executes, for each of the plurality of replicas set with different values ​​of the temperature parameter, Figure 1 The processing in steps S2 to S4 indicated in . Then, the processing unit 12 performs replica exchange each time the processing in steps S2 to S4 is repeated a predetermined number of times. For example, the processing unit 12 randomly selects two replicas from the plurality of replicas. Then, the processing unit 12 exchanges the value of the temperature parameter or the value of the n multi-valued variables between the selected two replicas with a predetermined exchange probability based on the difference in the value of the evaluation function between the replicas or the difference in the value of the temperature parameter. For example, whenever a transition is accepted in each replica, the processing unit 12 updates the value of the evaluation function represented by formula (5), and saves the minimum value until then and the values ​​of the n multi-valued variables when the minimum value is obtained. Then, for example, when a predetermined search end condition is met, the processing unit 12 outputs the values ​​of the n multi-valued variables corresponding to the smallest minimum value among all replicas among the minimum values ​​stored in each replica as the search result.

[0100] As described above, the processing unit 12 of the data processing device 10 according to the first embodiment generates a transition destination candidate as a part of the value transitionable range for each of the n multi-valued variables based on the evaluation function information 11a. Then, the processing unit 12 calculates the amount of change in the value of the evaluation function associated with the transition to the transition destination candidate for each of the n multi-valued variables based on the evaluation function information 11a, and specifies the multi-valued variable to be subjected to the transition based on the amount of change. In addition, the processing unit 12 causes the value of the specified multi-valued variable to be transitioned to the value of the transition destination candidate.

[0101] Compared with the case where each of the n multi-valued variables is converted into a large number of binary state variables of 0 or 1, this can avoid the complexity of moving the data used to calculate the variation of the evaluation function, and thus facilitate parallel calculation. Therefore, the combinatorial optimization problem represented by the multi-valued variables can be calculated by parallel experiments. Since parallel experiments can also be used for combinatorial optimization problems represented by multi-valued variables, the computational efficiency of such combinatorial optimization problems is improved.

[0102] Furthermore, in the case of calculating an allocation problem for allocating n elements to m allocation destinations using a state variable of 0 or 1, the number of state variables representing the allocation state is n×m, and the number of weighting factors between the state variables is n. 2 ×m 2 In contrast, by using a multi-valued variable representing the allocation state, the factors to be stored are only n 2 flow and m 2 distance. In the case of updating the variation using the difference as described above, the variation should be saved. However, since the amount of data expected to be saved is on the order of n×m, the memory capacity can be made smaller than the case of using binary state variables even in consideration of the flow rate and distance.

[0103] (Second Embodiment)

[0104] The data processing apparatus according to the second embodiment searches for a solution to an assignment problem expressed by using a flow matrix and a distance matrix.

[0105] Figure 3 is a diagram showing an exemplary data processing device according to the second embodiment.

[0106] The data processing device 20 according to the second embodiment includes a storage unit 21 and a processing unit 22 .

[0107] The storage unit 21 includes an F memory 21a and D memories 21b and 21c. For example, the F memory 21a and the D memories 21b and 21c may be volatile semiconductor memories such as RAM, or may be nonvolatile storage devices such as flash memory.

[0108] The F memory 21a stores the flow matrix represented by the formula (3).

[0109] The D memory 21b stores a matrix which is a distance matrix represented by the formula (4) and includes matrix elements arranged based on the current values ​​(representing the current state) of the n multi-valued variables.

[0110] The D memory 21c stores a matrix which is a distance matrix represented by Formula (4) and includes matrix elements arranged based on the values ​​of the corresponding transition destination candidates of n multi-valued variables (indicating the states of the transition destination candidates).

[0111] In the case of performing a solution search using multiple replicas, such as in the case of performing replica exchange, two types of distance matrices as described above are stored for each replica. The flow matrix can be shared between multiple replicas.

[0112] The processing unit 22 includes a δΔE calculation circuit 22 a , a ΔE update circuit 22 b , a ΔE preservation circuit 22 c , a transition acceptance determination circuit 22 d , a column interchange circuit 22 e , a state preservation-update circuit 22 f , and a control circuit 22 g .

[0113] The δΔE calculation circuit 22a calculates φ in parallel for n terms. j Transformed into φ' j The difference in the amount of change in the value of the evaluation function when ΔE is denoted by the formula (7) described above. In the following, the amount of change is denoted by ΔE, and the difference is denoted by δΔE.

[0114] Figure 3 A circuit example is shown as an example of the δΔE calculation circuit 22a. The δΔE calculation circuit 22a includes registers 22a1, 22a2, 22a3, 22a4, and 22a5, adders 22a6, 22a7, and 22a8, and a multiplier 22a9.

[0115] The register 22a1 stores the flow rate (f ij ).

[0116] The register 22a2 holds the allocation destination (φ) of the element with identification number = j which has been read from the D memory 21b. j ) and the current allocation destination (φ) of the element with identification number = i i ) between the two.

[0117] The register 22a3 holds the allocation destination candidate (φ′) of the element with identification number=j which has been read from the D memory 21b. j ) and the allocation destination (φ) of the element with identification number = i i ) between the two.

[0118] The register 22a4 holds the allocation destination (φ) of the element with identification number = j which has been read from the D memory 21c. j ) and the allocation destination candidate (φ') of the element with identification number = i i ) between the two.

[0119] The register 22a5 holds the allocation destination candidate (φ′) of the element with identification number=j which has been read from the D memory 21c. j ) and the allocation destination candidate (φ') of the element with identification number = i i ) between the two.

[0120] The adder 22a6 outputs a value obtained by subtracting the distance stored in the register 22a2 from the distance stored in the register 22a3.

[0121] The adder 22a7 outputs a value obtained by subtracting the distance stored in the register 22a4 from the distance stored in the register 22a5.

[0122] The adder 22a8 outputs a value obtained by subtracting the output value of the adder 22a6 from the output value of the adder 22a7.

[0123] The multiplier 22a9 outputs the product of the flow rate stored in the register 22a1 and the adder 22a8. The output of the multiplier 22a9 is δΔE.

[0124] By the circuit configuration as described above, with respect to n multi-valued variables φ 1 To φ n , the calculation of δΔE can be performed in parallel.

[0125] In the case of i≠j, the ΔE updating circuit 22b updates ΔE by adding δΔE calculated by the δΔE calculating circuit 22a to the original ΔE. In the case of i=j, the ΔE updating circuit 22b updates ΔE by inverting the sign of ΔE. 1 To φ n , the ΔE updating circuit 22 b can perform the updating process for ΔE in parallel.

[0126] The ΔE saving circuit 22c saves the above-described ΔE with respect to n multi-valued variables. In the case of performing a solution search using a plurality of replicas, for example, in the case of performing replica exchange, the ΔE saving circuit 22c is provided for each replica.

[0127] Based on the calculated ΔE, the transition acceptance determination circuit 22d specifies a multi-valued variable that accepts the transition to become a candidate for the transition destination. For example, the transition acceptance determination circuit 22d determines whether to accept the transition of the value of the multi-valued variable based on the comparison result between ΔE and the predetermined value supplied from the control circuit 22g. The predetermined value is, for example, a noise value obtained based on the value of a random number and a temperature parameter. For example, in the case where the ΔE associated with the transition is less than log(rand)×T, the transition acceptance determination circuit 22d accepts the transition, and log(rand)×T is an example of a noise value obtained based on a uniform random number (rand) equal to or greater than zero but equal to or less than one and a temperature parameter (T). The transition acceptance determination circuit 22d outputs the identification number of the multi-valued variable that has accepted the transition.

[0128] The column exchange circuit 22e exchanges the columns of the distance matrix in the D memory 21b with the columns of the distance matrix in the D memory 21c, the columns being specified by the identification number of the multi-valued variable to be transformed. Note that in the case where the reading from the distance matrix for calculating δΔE is performed for each column, the column exchange circuit 22e may exchange rows instead of columns.

[0129] The reason for swapping the columns or rows of the distance matrix in this way is as follows.

[0130] Although not shown, each of the ΔΔE calculation circuit 22a, the ΔE update circuit 22b, the ΔE holding circuit 22c, and the transition acceptance determination circuit 22d includes a circuit with respect to φ i (i=1 to n) n circuit blocks that perform processing in parallel. The n circuit blocks are arranged in the order of identification number=i. The F memory 21a stores f ij The part is arranged near the circuit block corresponding to i so that the f read ij n instances of φ are easily propagated in parallel to register 22a1 comprising n circuit blocks. In order to use similar parallel propagation to read the matrix elements of the distance matrix, it is desirable to i With φ j The distance between is saved in such a way that i is the first identifier (e.g., column number), and φ j This is why the columns or rows are interchanged in the distance matrix.

[0131] Note that an example of interchange will be described later (see Figure 6 ).

[0132] The state save-update circuit 22f saves the values ​​of the n multi-valued variables and, in addition, causes the value of the multi-valued variable corresponding to the identification number output by the transition acceptance determination circuit 22d to be transitioned to the value of the transition destination candidate. Furthermore, the state save-update circuit 22f specifies the matrix element having the identification number of the multi-valued variable to be accepted for transition to be read from the F memory 21a for the δΔE calculation circuit 22a.

[0133] In addition, the state save-update circuit 22f specifies the matrix elements having the values ​​before and after the transition of the multi-valued variable to be subjected to the transition to be read from the D memories 21b and 21c for the δΔE calculation circuit 22a. In the case of performing a solution search using a plurality of replicas, such as in the case of performing replica exchange, the state save-update circuit 22f saves the values ​​of n multi-valued variables for each replica.

[0134] The control circuit 22g controls each unit of the data processing device 20. In addition, the control circuit 22g generates and regenerates the transition destination candidate value of each multi-valued variable, and rearranges the matrix elements of the distance matrix stored in the D memory 21c so as to correspond to the generated transition destination candidates. In addition, the control circuit 22g generates, for example, a predetermined value for comparison with ΔE, and supplies the generated predetermined value to the transition acceptance determination circuit 22d. In addition, the control circuit 22g performs control of pipeline processing and the like described later.

[0135] Figure 4 is a diagram showing an example of a current state and states of transition destination candidates. Figure 4 An example of a current state and states of transition destination candidates in an assignment problem for assigning eight facilities to eight locations is shown.

[0136] The current state indicates the locations where the eight facilities are currently arranged (current facility locations), and is represented by φ as a multi-valued variable. 1 To φ 8 The facility number is 1 to 8. For example, Figure 4 As shown, the facility with facility number = 1 is arranged in φ 1 = 2, and the facility with facility number = 6 is arranged at φ 6 =4.

[0137] The states of the transition destination candidates indicate the positions of the movement destination candidates of the eight facilities and are represented by φ' which is a multi-valued variable. 1 to φ' 8 For example, Figure 4 As shown, the position of the destination candidate with facility number = 1 is φ' 1 =4, and the position of the mobile destination candidate with facility number =6 is φ'6 =2.

[0138] The following instructions Figure 4 The shown accepts φ in the current state 6 to φ' 6 An example of the matrix elements to be read for the calculation of δΔE in the case of a transition.

[0139] Figure 5 is a diagram showing an example of matrix elements to be read from the F memory.

[0140] When accepting φ 6 to φ' 6 When the transition is made, the matrix element (f i6 ) in order to calculate δΔE.

[0141] Figure 6 is a diagram of an example of matrix elements to be read from D memory.

[0142] The matrix elements (distances) of the distance matrices stored in the D memories 21b and 21c are arranged in the order of identification numbers (1 to 8) of the positions in the column direction of the matrix, and in the order of facility numbers in the row direction. In the distance matrix stored in the D memory 21b, the matrix elements are arranged based on φ 1 To φ 8 In the distance matrix stored in the D memory 21c, the matrix elements are based on φ 1 To φ 8 The transition destination candidate φ' 1 to φ' 8 The values ​​are arranged.

[0143] In the distance matrix stored in the D memories 21 b and 21 c , a row is specified by the value of the multi-valued variable that undergoes transition or the transition value of the multi-valued variable, and matrix elements included in the specified row are read.

[0144] When accepting φ 6 to φ' 6 When the transition is made, the current location (φ) of the facility with the facility number = 6 is read from the D memory 21b. 6 ) and the current location (φ) of the facility with facility number = i (i = 1 to 8) i ) in order to calculate δΔE. Figure 4 In the example shown, since φ 6 =4, so the distance matrix φ is read from D memory 21b j = the matrix element in the 4th row. In addition, the position (φ′) of the movement destination candidate of the facility with the facility number = 6 is read from the D memory 21b.6 ) and the current location (φ i ). Figure 4 In the example shown, since φ' 6 =2, so the distance matrix φ is read from D memory 21b j = the matrix element in row 2.

[0145] At the same time, the current location (φ) of the facility with facility number = 6 is read from the D memory 21c. 6 ) and the position (φ') of the candidate destination of the facility with facility number = i i ). Figure 4 In the example shown, since φ 6 =4, so the φth distance matrix is ​​read from the D memory 21c. j = 4 rows. In addition, the position (φ′) of the destination candidate of the facility with the facility number = 6 is read from the D memory 21c. 6 ) and the position (φ') of the candidate destination of the facility with facility number = i i ). Figure 4 In the example shown, since φ' 6 =2, so the φth distance matrix is ​​read from the D memory 21c. j =Matrix elements in row 2.

[0146] Read the φth distance matrix from the D memories 21b and 21c j =4 rows can be read in parallel at the first read timing. The φth row of the distance matrix is ​​read from the D memories 21b and 21c. j = 2 matrix elements in the row can be executed in parallel at the second reading timing.

[0147] After calculating the difference indicated by formula (7), the matrix elements in the sixth column of the distance matrix stored in D memories 21b and 21c are interchanged with each other by column exchange circuit 22e. This ensures that the distance matrix reflects the updated status in D memories 21b and 21c.

[0148] Next, an operation example of the data processing device 20 according to the second embodiment will be described.

[0149] Figure 7 2 is a timing chart showing an example of the operation of the data processing device according to the second embodiment. Figure 7 In the operation example in , an example of pipeline processing using four replicas is shown. The processing includes a trial phase and an update phase.

[0150] The test phase includes a process in which the transition acceptance determination circuit 22 d acquires ΔE from the ΔE holding circuit 22 c , and a process in which the transition acceptance determination circuit 22 d performs transition acceptance determination.

[0151] The update phase includes the process of reading the matrix elements of the flow matrix from the F memory 21a, the process of reading the matrix elements of the distance matrix from the D memories 21b and 21c, and the process of swapping the columns of the distance matrix in the D memories 21b and 21c. The D memories 21b and 21c are provided for each replica. Figure 7 In FIG. 1 , the D memories 21 b and 21 c of the respective replicas 0 to 3 are represented as “D0 memory”, “D1 memory”, “D2 memory”, and “D3 memory”. In addition, the update phase includes a process in which the δΔE calculation circuit 22 a is caused to store the matrix elements of the flow matrix or the matrix elements of the distance matrix (in Figure 7 , a process in which the δΔE calculation circuit 22a calculates δΔE, and a process in which ΔE is updated.

[0152] Between timing t1 and timing t4, the process of acquiring ΔE is performed in the order from replica 0 to replicas 1, 2, and 3. Between timing t2 and timing t5, acceptance determination is performed in the above order starting from replica 0. In each replica, the identification number (expressed as "Idk") of the multi-valued variable that has accepted the transition is output from the transition acceptance determination circuit 22d.

[0153] Between timing t3 and timing t9, the process of reading the matrix elements of the flow matrix from the F memory 21a and the process of reading the matrix elements of the distance matrix from the D memories 21b and 21c are performed. For each copy, the reading of the matrix elements of the distance matrix is ​​performed in two cycles. For example, Figure 6 As shown, this is because the matrix elements of two rows are read twice in parallel from the D memories 21b and 21c.

[0154] From timing t4 to timing t10, the matrix elements read from the F memory 21a and the D memories 21b and 21c are stored in the registers 22a1 to 22a5 of the δΔE calculation circuit 22a. The matrix elements to be used for calculating δΔE in the replica 0 are stored between timing t4 and timing t7, the arithmetic operation of δΔE is performed between timing t5 and timing t7, and the arithmetic result is confirmed between timing t7 and timing t8. At timing t7 and subsequent timings, processing similar to that described above regarding the replica 0 is performed in the order of replicas 1, 2, and 3. In addition, at timing t8 and subsequent timings, each time the arithmetic result of δΔE for each replica is confirmed, ΔE of each replica is updated.

[0155] At timing t11 and subsequent timings, Figure 6 In the distance matrices stored in the D memories 21b and 21c shown, the columns are interchanged in parallel for each copy, each column including n matrix elements. Since the n matrix elements included in each column are interchanged one by one between the two columns, the interchange is completed within n cycles.

[0156] Figure 8 : is a flowchart showing an example of a processing procedure of the data processing device according to the second embodiment.

[0157] First, under the control of the control circuit 22g, initialization of each unit of the data processing device 20 (step S10) and data setting and operation setting (step S11) are performed.

[0158] In the processing of steps S10 and S11, the flow matrix is ​​written into the F memory 21a, and the distance matrix including the matrix elements arranged based on the initial values ​​of the n multi-valued variables is written into the D memory 21b. In addition, for each multi-valued variable, for example, Figure 4 As shown, the transition destination candidates are generated. Then, a distance matrix including matrix elements arranged based on the values ​​of the transition destination candidates is written into the D memory 21c.

[0159] In addition, based on the initial values ​​of the n multi-valued variables and the evaluation function information (flow matrix, distance matrix, bias coefficient, etc.), the initial value of the evaluation function represented by formula (5) is calculated and saved. In addition, for each multi-valued variable, the initial value of ΔE represented by formula (6) is calculated and saved in the ΔE saving circuit 22c. In addition, in the case of executing the replica exchange method, the initial value of the temperature parameter (T) is set in each replica. For each replica, the value of the temperature parameter is set to a different value. In addition, the search end condition, etc. are also set.

[0160] The control circuit 22g determines whether to switch the transition destination candidate (step S12). For example, each time the acceptance determination processing in step S16 described later is repeated a predetermined number of times, the control circuit 22g determines whether to switch the transition destination candidate. In the case where the minimum value of the evaluation function is not updated even if the processing in step S16 is repeated a predetermined number of times, the control circuit 22g may determine to switch the transition destination candidate. In the case of determining to switch the transition destination candidate, the processing in step S13 is performed; and in the case of determining not to switch the transition destination candidate, the processing in step S15 is performed.

[0161] In the processing in step S13, the control circuit 22g switches the transition destination candidate and rewrites the distance matrix stored in the D memory 21c to a distance matrix based on the value of the switched transition destination candidate. In addition, the control circuit 22g calculates ΔE based on the current value of the multi-valued variable and the value of the switched transition destination candidate, and saves the calculated ΔE in the ΔE saving circuit 22c (step S14).

[0162] In the process of step S15, the transition acceptance determination circuit 22d acquires n ΔEs from the ΔE storage circuit 22c. Then, the transition acceptance determination circuit 22d determines whether to accept transition to a transition destination candidate based on the n ΔEs for each of the n multi-valued variables (step S16). In the case where there is a multi-valued variable that accepts transition to a transition destination candidate, the transition acceptance determination circuit 22d selects the multi-valued variable to be transitioned so as to satisfy the constraint condition (step S17).

[0163] The control circuit 22g determines whether there is a multi-valued variable that has been subjected to the transition (and the multi-valued variable has been selected in the processing in step S17) (step S18). In the case where it is determined that there is a multi-valued variable that has been subjected to the transition, the processing in step S19 is performed. In the case where it is determined that there is no multi-valued variable that has been subjected to the transition, the processing in step S23 is performed.

[0164] In the process of step S19, a process of reading the matrix elements of the flow matrix from the F memory 21a and a process of reading the matrix elements of the distance matrix from the D memories 21b and 21c are performed.

[0165] The δΔE calculation circuit 22a calculates δΔE using each of the read matrix elements (step S20). Then, the ΔE update circuit 22b updates the ΔEs of the n multi-valued variables using the calculated δΔE (step S21).

[0166] For example, Figure 6 As shown, the column interchanging circuit 22e performs a column interchanging process of interchanging the columns of the distance matrices stored in the D memories 21b and 21c according to the accepted transition of the values ​​of the multi-valued variables (step S22).

[0167] The control circuit 22g determines whether to end the search (step S23). For example, when the search time reaches the timeout limit, it is determined that the search has ended, and the search ends. In the case where it is determined not to end the search, the processing starting from step S12 is repeated.

[0168] In the case where it is determined to end the search, the control circuit 22g outputs the search result (step S24) and ends the processing.

[0169] In the case of executing the simulated annealing method, the control circuit 22g reduces the value of the temperature parameter (T) described above according to the predetermined temperature parameter switching schedule, while repeating the processing in steps S12 to S23. Then, for example, when the predetermined search end condition is met, the control circuit 22g outputs the values ​​of the n multi-valued variables as the search result of the solution of the allocation problem (for example, the values ​​of the n multi-valued variables are displayed on a display device (not shown)). Note that the control circuit 22g can update the value of the evaluation function represented by formula (5) whenever a transition is accepted, and save the minimum value until then and the values ​​of the n multi-valued variables when the minimum value is obtained. In this case, the control circuit 22g can output the values ​​of the n multi-valued variables corresponding to the minimum value stored when the search end condition is met as the search result.

[0170] In the case where the control circuit 22g performs the replica exchange method, the control circuit 22g performs the processing in steps S12 to S23 for each of the multiple replicas set with the values ​​of the temperature parameter different from each other. Then, the control circuit 22g performs the replica exchange each time the processing in steps S12 to S23 is repeated a predetermined number of times. For example, the control circuit 22g randomly selects two replicas from the multiple replicas, and based on the difference in the value of the evaluation function between the replicas or the difference in the value of the temperature parameter, exchanges the value of the temperature parameter or the value of the n multi-valued variables between the selected two replicas with a predetermined exchange probability. For example, whenever a transition is accepted in each replica, the control circuit 22g updates the value of the evaluation function represented by formula (5), and saves the minimum value until then and the values ​​of the n multi-valued variables when the minimum value is obtained. Then, for example, when a predetermined search end condition is satisfied, the control circuit 22g outputs the values ​​of the n multi-valued variables corresponding to the smallest minimum value among all the replicas among the minimum values ​​stored in each replica as the search result.

[0171] According to the data processing device 20 as described above, compared with the case where each of the n multi-valued variables is converted into a large number of binary state variables of 0 or 1, the complexity of moving the data used to calculate the variation of the evaluation function can be avoided, and thus parallel calculation is facilitated. Therefore, the combinatorial optimization problem represented by the multi-valued variables can be calculated by parallel experiments. Since parallel experiments can also be used for the combinatorial optimization problem represented by the multi-valued variables, the calculation efficiency of such combinatorial optimization problem is improved.

[0172] Furthermore, the data processing device 20 saves the ΔE associated with the transition of each of the n multi-valued variables, and updates ΔE using the calculated δΔE. This ensures that ΔE itself no longer needs to be recalculated.

[0173] In addition, the data processing device 20 includes an F memory 21a that stores a flow matrix. The data processing device 20 also includes: a D memory 21b that stores a distance matrix including matrix elements arranged based on the current values ​​of n multi-valued variables; and a D memory 21c that stores a distance matrix including matrix elements arranged based on the values ​​of the corresponding transition destination candidates of the n multi-valued variables. Then, when calculating δΔE, the matrix elements of the flow matrix specified by the identification number of the multi-valued variable that undergoes the transition are read from the F memory 21a. In addition, the matrix elements of the distance matrix specified by the values ​​before and after the transition of the multi-valued variable are read from the D memories 21b and 21c. This can enable δΔE to be efficiently calculated in parallel with respect to n multi-valued variables.

[0174] (Third Embodiment)

[0175] The data processing device of the third embodiment is obtained by using a common circuit for the data processing device 20 of the second embodiment.

[0176] Fig. 9 FIG. 4 is a diagram showing an exemplary data processing device according to the third embodiment. Fig. 9 In, for Figure 3 Elements shown in FIG. 1 are given the same reference numerals. Fig. 9 middle, Figure 3 The column interchange circuit 22e and the state save-update circuit 22f shown in FIG. 2 are not shown.

[0177] The storage unit 31 of the data processing device 30 of the third embodiment includes a D memory 31 a and a storage control circuit 31 b , as well as an F memory 21 a and D memories 21 b and 21 c .

[0178] The D memory 31a stores a distance matrix common to each copy. The matrix elements of the distance matrix stored in the D memory 31a are arranged in the order of position identification numbers (1 to 8) in both the row direction and the column direction. In the case of interchanging the columns of the distance matrices stored in the D memories 21b and 21c, two columns corresponding to the two columns to be interchanged are read from the distance matrix stored in the D memory 31a. Then, the interchange is performed by writing one column into the position of the column to be interchanged in the D memory 21b, and writing another column into the column to be interchanged in the D memory 21c.

[0179] As described above Figure 3 In the data processing device 20, such a D memory 31a can also be used to interchange columns.

[0180] The storage control circuit 31b controls the write and read operations with respect to the F memory 21a and the D memories 21b, 21c and 31a under the control of the control circuit 33. In order to calculate δΔE, the storage control circuit 31b uses a common read path so that the five matrix elements described above included in the formula (7) are read from the F memory 21a and the D memories 21b and 21c at different read timings from each other.

[0181] The δΔE calculation circuit 32 includes registers 32 a , 32 b , and 32 d , and an arithmetic circuit 32 c .

[0182] The register 32a stores the matrix elements (flow rate and distance) read from the F memory 21a and the D memories 21b and 21c. The register 32b stores the matrix elements read from the D memories 21b and 21c.

[0183] The five kinds of matrix elements described above held in the registers 32a and 32b are sequentially input to the arithmetic circuit 32c, and an intermediate arithmetic result or δΔE of the formula (7) is output based on these input matrix elements and the intermediate arithmetic result held in the register 32d.

[0184] For example, the arithmetic circuit 32c sequentially calculates the φ stored in the registers 32a and 32b. j With φ i The distance between j With φ i The distance between j With φ' i The distance between j With φ i The distance between is added or subtracted, thereby performing the arithmetic operation in the brackets on the right side of formula (7). The result of the arithmetic operation is stored in register 32d as an intermediate arithmetic result. Then, arithmetic circuit 32c outputs the flow rate (f ij ) is multiplied by the intermediate arithmetic result stored in register 32d to obtain δΔE.

[0185] The register 32d holds the intermediate arithmetic result output by the arithmetic circuit 32c or δΔE as the arithmetic result.

[0186] The control circuit 33 has Figure 3 The control circuit 33 has functions similar to those of the control circuit 22g of the processing unit 22 of the data processing device 20 shown in FIG. 3. Furthermore, the control circuit 33 instructs the storage control circuit 31b at the data reading timing, etc., so as to cause the δΔE calculation circuit 32 to calculate δΔE as described above.

[0187] Note that the data processing device 30 may store whether the current value of each multi-valued variable is the original value or the value of the generated transition destination candidate. In this case, the matrix elements of the distance matrix used to calculate δΔE can be appropriately read without interchanging the columns between the distance matrices stored in the D memories 21b and 21c. However, when regenerating the transition destination candidate, the control circuit 33 rearranges the matrix elements of the distance matrix stored in the D memory 21b according to the current value of the multi-valued variable.

[0188] Fig.10 2 is a timing chart showing an example of the operation of the data processing device according to the third embodiment. Fig.10 In the operation example in Figure 7 The timing diagram shown in is similar to that shown in , showing an example of pipeline processing using four replicas.

[0189] Between timing t20 and timing t23, the process of acquiring ΔE is performed in the order from replica 0 to replicas 1, 2, and 3. Between timing t21 and timing t24, acceptance determination is performed in the above order starting from replica 0. In each replica, the identification number of the multi-valued variable that has accepted the transition is output from the transition acceptance determination circuit 22d.

[0190] Between timing t22 and timing t25, the process of reading the matrix elements of the flow matrix from the F memory 21a and the process of reading the matrix elements of the distance matrix from the D memories 21b and 21c are performed. In the data processing device 30 of the third embodiment, the reading of the matrix elements of the distance matrix is ​​performed in four cycles for each copy.

[0191] From timing t23 to timing t29, the matrix elements read from the F memory 21a and the D memories 21b and 21c are stored in the registers 32a and 32b of the δΔE calculation circuit 32. The matrix elements to be used for calculating δΔE in the replica 0 are stored between timing t23 and timing t26, the arithmetic operation of δΔE is performed between timing t24 and timing t26, and the arithmetic result is confirmed between timing t26 and timing t27. At timing t26 and subsequent timings, processing similar to that described above with respect to the replica 0 is performed in the order of replicas 1, 2, and 3. Furthermore, at timing t27 and subsequent timings, each time the arithmetic result of δΔE for each replica is confirmed, ΔE of each replica is updated.

[0192] At timing t30 and subsequent timings, a write process of exchanging columns is performed using the matrix elements of the distance matrix read from the D memory 31a, each column including n matrix elements in the distance matrix stored in the D memories 21b and 21c. The write process is performed in parallel for each copy. Since the n matrix elements included in each column are exchanged one by one between the two columns, the exchange is completed within n cycles. Note that, as described above, in the case where the data processing device 30 stores whether the current value of each multi-valued variable is the original value or the value of the generated transition destination candidate, there is no need to exchange columns between the distance matrices stored in the D memories 21b and 21c.

[0193] According to the data processing device 30 of the third embodiment as described above, the δΔE calculation circuit 32 processes five kinds of matrix elements using the common arithmetic circuit 32c. Figure 3 This can enable the circuit scale to be reduced by comparing with the δΔE calculation circuit 22a shown in FIG.

[0194] (Fourth Embodiment)

[0195] In the second and third embodiments, one transition destination candidate for each value of the multi-valued variable has been assumed and described, but a plurality of transition destination candidates may exist for each value of the multi-valued variable.

[0196] Fig.11 is a diagram showing an example of current states of a plurality of transition destination candidates and states of the transition destination candidates. Fig.11 The current state and the states of two transition destination candidates (positions of the transfer destination candidates) in the allocation problem for allocating eight facilities to eight locations are shown. Fig.11 In the example, multiple facilities are allowed to be arranged at one location.

[0197] The state of the first transition destination candidate is represented by φ' which is a multi-valued variable 1 to φ' 8 For example, Fig.11 As shown, the position of the first mobile destination candidate of the facility with facility number = 1 is φ' 1 =4, and the position of the second mobile destination candidate is φ" 1 = 6. The position of the first destination candidate of the facility with facility number = 6 is φ' 6 =6, and the position of the second mobile destination candidate is φ" 6 =1.

[0198] The following instructions Fig.11 The shown accepts φ in the current state 6 to φ" 6An example of the matrix elements to be read for the calculation of δΔE in the case of a transition.

[0199] For matrix elements to be read from F memory, Figure 5 The acceptance φ shown in 6 to φ' 6 The case of the transformation is similar to that of reading the matrix element (f i6 ).

[0200] Fig.12 is a diagram showing an example of matrix elements to be read from the D memory.

[0201] When there are the first transition destination candidate state and the second transition destination candidate state, the D memory 21d corresponding to the second transition destination candidate state is used in addition to the D memory 21b corresponding to the current state and the D memory 21c corresponding to the first transition destination candidate state.

[0202] The matrix elements (distances) of the distance matrices stored in the D memories 21b, 21c, and 21d are arranged in the order of the identification numbers (1 to 8) of the locations in the column direction of the matrix and in the order of the facility numbers in the row direction. In the distance matrix stored in the D memory 21b, the matrix elements are arranged based on φ 1 To φ 8 In the distance matrix stored in the D memory 21c, the matrix elements are based on φ 1 To φ 8 The first transition destination candidate φ' 1 to φ' 8 In the distance matrix stored in the D memory 21d, the matrix elements are arranged based on φ 1 To φ 8 The second transition destination candidate φ" 1 to φ" 8 The values ​​are arranged.

[0203] In the case of j≠i, when the multi-valued variable φ j Transformed into φ" j When , the difference in the amount of change associated with the value of the multi-valued variable with identification number = i transitioning to the first transition destination candidate is expressed by the following formula (13).

[0204] [Mathematical formula 13]

[0205]

[0206] Note that in the case of j=i, the difference value of the change amount associated with the transition of the value of the multi-valued variable having identification number=i to the first transition destination candidate is expressed by the following formula (14).

[0207] [Mathematical formula 14]

[0208] δΔE(φ i →φ′ i )=-ΔE(φ i →φ" i )+ΔE(φ i →φ′ i ) (14)

[0209] In addition, in the case of j≠i, when the multi-valued variable φ j Transformed into φ" j When , the difference in the amount of change associated with the transition of the value of the multi-valued variable with identification number = i to the second transition destination candidate is expressed by the following formula (15).

[0210] [Mathematical formula 15]

[0211]

[0212] Note that in the case of j=i, the difference value of the change amount associated with the transition of the value of the multi-valued variable having identification number=i to the second transition destination candidate is expressed by the following formula (16).

[0213] [Mathematical formula 16]

[0214] δΔE(φ″Δ→φ i )=-2ΔE(φ i →φ″ i ) (16)

[0215] When accepting φ 6 to φ" 6 When the transition occurs (in the case of j=6), the current position (φ6) of the facility with the facility number=6 and the current position (φ) of the facility with the facility number=i (i=1 to 8) are read from the D memory 21b. i ) to calculate δΔE in formula (13). Fig.11 In the example shown, since φ 6 =5, so the φth distance matrix is ​​read from D memory 21b. j = 5 rows. In addition, the position (φ"6) of the second mobile destination candidate of the facility with facility number = 6 and the current position (φ i ). Fig.11 In the example shown, since φ"6 =1, so the φth distance matrix is ​​read from D memory 21b j =Matrix elements in row 1.

[0216] At the same time, the current position (φ6) of the facility with the facility number = 6 and the position (φ′) of the first candidate destination of the facility with the facility number = i are read from the D memory 21c. i ). Fig.11 In the example shown, since φ 6 =5, so the φth distance matrix is ​​read from the D memory 21c. j = 5 rows. In addition, the position (φ′) of the first mobile destination candidate of the facility with the facility number = 6 is read from the D memory 21c. 6 ) and the position (φ') of the first candidate destination of the facility with facility number = i i ). Fig.11 In the example shown, since φ' 6 =1, so the φth distance matrix is ​​read from the D memory 21c. j =Matrix elements in row 1.

[0217] The distance read from the D memory 21b to calculate δΔE in the formula (15) is the same as the distance in the case of calculating δΔE in the formula (13). At the same time, the current position (φ) of the facility with the facility number = 6 is read from the D memory 21d to calculate δΔE in the formula (15). 6 ) and the distance between the position (φ"i) of the second mobile destination candidate of the facility with facility number = i. Fig.11 In the example shown, since φ 6 =5, so the φth distance matrix is ​​read from the D memory 21d. j = 5 rows. In addition, the position (φ" of the second mobile destination candidate of the facility with facility number = 6 is read from the D memory 21d. 6 ) and the position (φ" of the second mobile destination candidate of the facility with facility number = i i ). Fig.11 In the example shown, since φ" 6 =1, so the φth distance matrix is ​​read from the D memory 21d. j =Matrix elements in row 1.

[0218] Although the configuration of the data processing device of the fourth embodiment is not shown, it can be Figure 3 or Fig. 9 The configuration shown in is implemented to use a configuration of multiple transition destination candidates. Fig.12The D memory 21d shown in FIG. Figure 3 or Fig. 9 In addition, by changing Figure 3 or Fig. 9 The δΔE calculation circuit 22a or 32 shown in FIG. 2 calculates the δΔE in formulas (13) to (16). In addition, the ΔE storage circuit 22c is i Transformed into φ' i In the case of saving ΔE(φ i →φ' i ), and in φ i Transformed into φ" i In the case of saving ΔE(φ i →φ" i ). The ΔE updating circuit 22b updates ΔE (φ) using δΔE calculated based on equations (13) and (14). i →φ' i ), and update ΔE(φ using δΔE calculated based on equations (15) and (16) i →φ" i ). Based on ΔE(φ i →φ' i ) and ΔE(φ i →φ" i ), the transition acceptance determination circuit 22d specifies a multi-valued variable that accepts transition to the first transition destination candidate or the second transition destination candidate.

[0219] Note that whether to accept the transition to the first transition destination candidate and the second transition destination candidate may not be determined at the same time. In the case where the D memory 21c is used to determine whether to accept the transition to the first transition destination candidate, it is not necessary to use the D memory 21d to determine whether to accept the transition to the second transition destination candidate. The same applies to the opposite case. The transition destination candidate can be appropriately transferred between the first transition destination candidate and the second transition destination candidate.

[0220] As described above, the data processing apparatus of the fourth embodiment includes the D memories 21c and 21d which store distance matrices including matrix elements arranged based on the values ​​of the corresponding transition destination candidates. This can make it possible to reduce the processing overhead caused by regenerating the transition destination candidates. In addition, for example, while performing the solution search using the D memory 21c, the distance matrix in the D memory 21d can also be rewritten as a distance matrix corresponding to the new transition destination candidate.

[0221] (Fifth Embodiment)

[0222] Fig.13 FIG. 4 is a diagram showing an exemplary data processing device according to the fifth embodiment. Fig.13 In, with Fig. 9 The same elements as those shown in FIG. 1 are given the same reference numerals.

[0223] In formula (7), δΔE can be modified as in the following formula (17).

[0224] [Mathematical formula 17]

[0225]

[0226] The right side of formula (17) includes φ' j With φ' i The distance between j With φ i The difference in distance between j With φ' i The distance between j With φ i The difference in distance between .

[0227] In the data processing device 40 of the fifth embodiment, the storage unit 41 includes a ΔD memory 41a, which stores a matrix including matrix elements represented by two differences as described above. The differences are arranged in the matrix so that they are readable by specifying the row number or column number. In the following, the above matrix based on the difference will be referred to as the difference distance matrix.

[0228] Fig.14 is a diagram of an example of reading matrix elements from a ΔD memory.

[0229] The difference distance matrix φ stored in the ΔD memory 41a j The matrix elements in the rows indicate φ j With φ' i The distance between j With φ i The difference in distance between . The difference distance matrix φ' j The matrix elements in the row indicate φ' j With φ' i The distance between j With φ i The difference in distance between .

[0230] When accepting φ 6 to φ' 6 When the transition occurs, the difference distance matrix φ is read from the ΔD memory 41a. 6 The matrix elements in the row are used to calculate δΔE. Fig.14 In the example shown, since φ 6 =5, so read the φth 6= 5 rows of matrix elements. In addition, φ' of the difference distance matrix is ​​read from the ΔD memory 41a. 6 The matrix elements in the row. Fig.14 In the example shown, since φ' 6 =6, so read the φ'th 6 =Matrix elements in 6 rows.

[0231] Such a difference distance matrix is ​​generated by difference calculation (Δd calculation) using the distance matrix stored in the D memory 31a. For example, the difference calculation is performed by the control circuit 33. In the case of performing a solution search using a plurality of replicas, a difference distance matrix is ​​generated for each replica.

[0232] Note that the data processing device 40 may store whether the current value of each multi-valued variable is the original value or the value of the generated transition destination candidate. In this case, while the transition destination candidate is not changed, the matrix elements for Δd calculation can be appropriately read without interchanging the columns in the distance matrix stored in the D memory 31a. However, when the transition destination candidate is regenerated, the control circuit 33 rearranges the matrix elements of the distance matrix according to the current value of the multi-valued variable and the value of the transition destination candidate. Thereafter, the control circuit 33 performs Δd calculation to initialize the difference distance matrix stored in the ΔD memory 41a.

[0233] The δΔE calculation circuit 42 of the data processing device 40 includes registers 42a and 42b. The register 42a stores the matrix elements (differences (Δd) between flow rate and distance) read from the F memory 21a and the ΔD memory 41a. The register 42b stores the matrix elements read from the ΔD memory 41a.

[0234] The arithmetic circuit 32c outputs an intermediate arithmetic result or δΔE of the formula (17) based on the matrix elements or intermediate arithmetic results held in the registers 42a, 42b, and 32d.

[0235] For example, the arithmetic circuit 32c performs the arithmetic operation in the brackets on the right side of the formula (17) by subtracting the two differences of Δd stored in the registers 42a and 42b. The result of the arithmetic operation is stored in the register 32d as an intermediate arithmetic result. Then, the arithmetic circuit 32c outputs the flow rate (f ij ) is multiplied by the intermediate arithmetic result stored in register 32d to obtain δΔE.

[0236] The other elements of the data processing device 40 are Fig. 9 The elements shown in are the same.

[0237] Fig.15 1 is a timing chart showing an example of the operation of the data processing device according to the fifth embodiment. Fig.15 In the operation example in Figure 7 The timing diagram shown in is similar to that shown in , showing an example of pipeline processing using four replicas.

[0238] Between timing t40 and timing t43, the process of acquiring ΔE is performed in the order from replica 0 to replicas 1, 2, and 3. Between timing t41 and timing t44, acceptance determination is performed in the above order starting from replica 0. In each replica, the identification number of the multi-valued variable that has accepted the transition is output from the transition acceptance determination circuit 22d.

[0239] Between timing t42 and timing t48, the process of reading the matrix elements of the flow matrix from the F memory 21a and the process of reading the matrix elements of the difference distance matrix from the ΔD memory 41a are performed. In the data processing device 40 of the fifth embodiment, the matrix elements of the difference distance matrix are read in two cycles for each replica.

[0240] From timing t43 to timing t49, the matrix elements read from the F memory 21a and the ΔD memory 41a are stored in the registers 42a and 42b of the δΔE calculation circuit 42. The matrix elements to be used for calculating δΔE in the replica 0 are stored between timing t43 and timing t46, the arithmetic operation of δΔE is performed between timing t44 and timing t46, and the arithmetic result is confirmed between timing t46 and timing t47. At timing t46 and subsequent timings, processing similar to that described above regarding the replica 0 is performed in the order of replicas 1, 2, and 3. In addition, at timing t47 and subsequent timings, each time the arithmetic result of δΔE for each replica is confirmed, ΔE of each replica is updated.

[0241] When regenerating the transition destination candidate, at timing t50 and subsequent timings, the control circuit 33 rearranges the matrix elements of the distance matrix in the D memory 31a according to the current value of the multi-valued variable and the value of the transition destination candidate. Thereafter, the control circuit 33 performs a write process of performing Δd calculation to initialize the difference distance matrix stored in the ΔD memory 41a. The write process is performed in parallel for each copy.

[0242] According to the data processing device 40 of the fifth embodiment as described above, by storing the difference in distance used to calculate δΔE in advance in the ΔD memory 41a, the circuit scale of the δΔE calculation circuit 42 can be reduced. In addition, the time taken to calculate δΔE can be shortened.

[0243] Note that the processing content of each of the above embodiments (for example, Figure 1 , Figure 8 etc.) can be realized by software by causing a data processing device to execute a program.

[0244] The program may be recorded in a computer-readable recording medium. For example, a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. may be used as a recording medium. Examples of magnetic disks include a floppy disk (FD) and an HDD. Examples of optical disks include a compact disk (CD), a recordable (R) / rewritable (RW) CD, a digital versatile disk (DVD), and a DVD-R / RW. Sometimes, the program may be recorded in a portable recording medium and distributed. In this case, the program may be copied from the portable recording medium to another recording medium for execution.

[0245] Fig.16 is a diagram showing an example of hardware of a computer as an example of a data processing device.

[0246] The computer 50 includes a CPU 51, a RAM 52, a HDD 53, a GPU 54, an input interface 55, a medium reader 56, and a communication interface 57. The above-mentioned units are coupled to a bus.

[0247] The CPU 51 is a processor including an arithmetic circuit that executes program commands. The CPU 51 loads at least a portion of the program and data stored in the HDD 53 into the RAM 52, and executes the program. Note that the CPU 51 may include, for example, a plurality of processor cores so as to perform processing of a plurality of copies in parallel. In addition, the computer 50 may include a plurality of processors. Note that a collection of a plurality of processors (a multiprocessor) may be referred to as a "processor".

[0248] The RAM 52 is a volatile semiconductor memory that temporarily stores programs to be executed by the CPU 51 and data to be used for arithmetic operations by the CPU 51. Note that the computer 50 may include a type of memory other than the RAM 52, or may include a plurality of memories.

[0249] The HDD 53 is a non-volatile storage device that stores software programs such as an operating system (OS), middleware, and application software, and data. The program includes, for example, a program for causing the computer 50 to perform a process of searching for a solution to a combinatorial optimization problem as described above. Note that the computer 50 may include another type of storage device, such as a flash memory or a solid-state drive (SSD), or may include a plurality of non-volatile storage devices.

[0250] The GPU 54 outputs an image (e.g., an image representing a calculation result of a combinatorial optimization problem, etc.) to a display 54a coupled to the computer 50 according to a command from the CPU 51. As the display 54a, a cathode ray tube (CRT) display, a liquid crystal display (LCD), a plasma display panel (PDP), an organic electroluminescence (OEL) display, or the like can be used.

[0251] The input interface 55 acquires an input signal from an input device 55a coupled to the computer 50, and outputs the acquired input signal to the CPU 51. As the input device 55a, a pointing device (such as a mouse, a touch panel, a touch pad, or a trackball), a keyboard, a remote controller, a button switch, etc. can be used. In addition, various types of input devices can be coupled to the computer 50.

[0252] The medium reader 56 is a reading device that reads the program and data recorded on the recording medium 56a. For example, a magnetic disk, an optical disk, a magneto-optical disk (MO), a semiconductor memory, etc. can be used. Examples of magnetic disks include FD and HDD. Examples of optical disks include CD and DVD.

[0253] The medium reader 56 copies, for example, a program or data read from the recording medium 56a to another recording medium such as the RAM 52 or the HDD 53. The read program is executed by the CPU 51, for example. Note that the recording medium 56a may be a portable recording medium and is sometimes used to distribute the program or data. In addition, the recording medium 56a and the HDD 53 are sometimes referred to as computer-readable recording media.

[0254] The communication interface 57 is an interface coupled to the network 57a and communicates with another information processing device via the network 57a. The communication interface 57 may be a wired communication interface coupled to a communication device such as a switch through a cable, or may be a wireless communication interface coupled to a base station through a wireless link.

[0255] (Revise)

[0256] In the data processing device 20 in the second embodiment and the like as described above, the columns or rows of the distance matrix are interchanged according to the current value of the multi-valued variable and the value of the transition destination candidate. i Such an exchange can be applied when solving the problem of assigning destinations.

[0257] However, in the case where the combinatorial optimization problem is formulated as a method for assigning a distribution destination having a first identification number = i to a distribution destination having a second identification number = ψ i In the case of the problem of elements of , the roles of the flow matrix and the distance matrix are opposite to those in the case described above. In this case, the columns or rows of the flow matrix are interchanged according to the current value of the multi-valued variable or the value of the transition destination candidate.

[0258] For example, in the case where the columns or rows of the flow matrix are interchanged, processing based on the following formulas (18) and (19) is performed instead of formulas (11) and (12).

[0259] [Mathematical formula 18]

[0260]

[0261] [Mathematical formula 19]

[0262]

[0263] In formula (19), k = ψ j Or k = ψ' j .

[0264] In order to i (i=1 to n) performs the processing based on formulas (18) and (19) in parallel, and converts the difference Δf in formula (19) into ik (Alternatively, with identification number = ψ i or ψ' i It is sufficient that the flow between elements of and elements with identification number = k) is stored in a memory having i as the first identification number and k as the second identification number.

[0265] Furthermore, with respect to the embodiments other than the second embodiment, if the roles of the flow matrix and the distance matrix are reversed, and ψ i Replace φ i , the implementation can be converted into an implementation that supports the interchange of columns or rows of the flow matrix.

[0266] Although one aspect of the data processing apparatus, the program, and the data processing method according to the embodiment has been described based on the embodiment, this is merely an example and is not limited to the above description.

Claims

1. A data processing device, comprising: a storage unit configured to store evaluation function information indicating an evaluation function of a combinatorial optimization problem expressed by using a plurality of multi-valued variables; as well as A processing unit configured to perform a process comprising: generating transition destination candidates that are part of respective transitionable ranges of values ​​of the plurality of multi-valued variables based on the evaluation function information; calculating, for each of the plurality of multi-valued variables, an amount of change in the value of the evaluation function associated with transition to the transition destination candidate based on the evaluation function information; specifying a multi-valued variable to be subjected to the transition from among the plurality of multi-valued variables based on the change amount; and The value of the designated multi-valued variable is converted into the value of the conversion destination candidate.

2. The data processing device according to claim 1, wherein: The processing unit comprises: a storage circuit that stores the change amount associated with the transition of each of the plurality of multi-valued variables; a calculation circuit that calculates, for each of the plurality of multi-valued variables, a difference in the amount of change associated with a change in the value of the multi-valued variable that undergoes the transition; and An updating circuit updates the variation by using the difference value.

3. The data processing device according to claim 2, wherein: The combinatorial optimization problem is an assignment problem represented by using a flow matrix and a distance matrix. The storage unit comprises: a first memory storing a first matrix that is one of the flow matrix and the distance matrix; a second memory storing a second matrix which is the other of the flow matrix and the distance matrix and includes matrix elements arranged based on the values ​​of the plurality of multi-valued variables; and a third memory storing a third matrix which is the other of the flow matrix and the distance matrix and includes the matrix elements arranged based on the values ​​of the transition destination candidates of the plurality of multi-valued variables, and The calculation circuit calculates the difference based on: the first matrix element of the first matrix specified by the identification number of the multi-valued variable that undergoes the transformation; the second matrix element of the second matrix and the third matrix element of the third matrix specified by the value before the transformation of the multi-valued variable that undergoes the transformation; and the fourth matrix element of the second matrix and the fifth matrix element of the third matrix specified by the value after the transformation of the multi-valued variable.

4. The data processing device according to claim 3, wherein: The rows or columns of the second matrix and the third matrix are specified by the values ​​of the multi-valued variable that undergo the transformation or the values ​​of the multi-valued variable after the transformation, and The second matrix element, the third matrix element, the fourth matrix element, or the fifth matrix element included in the designated row or column is read from the second memory or the third memory.

5. The data processing device according to claim 3, wherein: reading the second matrix element and the third matrix element in parallel from the second memory and the third memory at a first read timing, and The fourth matrix element and the fifth matrix element are read in parallel from the second memory and the third memory at a second read timing.

6. The data processing device according to claim 3, wherein: reading the first matrix element to the fifth matrix element from the first memory, the second memory, or the third memory at read timings different from each other by using a common read path, and The calculation circuit inputs the first to fifth matrix elements to a common arithmetic circuit to calculate the difference value.

7. The data processing device according to claim 3, wherein: The processing unit interchanges between the second matrix and the third matrix the columns or rows of the second matrix and the third matrix specified by the identification number of the multi-valued variable subjected to the conversion.

8. The data processing device according to claim 3, wherein: the transition destination candidates include a first transition destination candidate and a second transition destination candidate for each of the plurality of multi-valued variables, The third matrix is ​​arranged based on the values ​​of the first transition destination candidates, and The storage unit further includes a fourth memory storing a fourth matrix which is the other of the flow matrix and the distance matrix and includes the matrix elements arranged based on the value of the second transition destination candidate.

9. The data processing device according to claim 2, wherein: The combinatorial optimization problem is an assignment problem represented by using a flow matrix and a distance matrix. The storage unit comprises: a first memory storing a first matrix which is one of the flow matrix and the distance matrix; and a second memory storing a fourth matrix including matrix elements represented by differences between: values ​​of matrix elements of a second matrix that is the other of the flow matrix and the distance matrix and is arranged based on the values ​​of the plurality of multi-valued variables; and values ​​of a third matrix that is the other of the flow matrix and the distance matrix and is arranged based on the values ​​of the transition destination candidates of the plurality of multi-valued variables, and The calculation circuit calculates the difference based on: a first matrix element of the first matrix specified by an identification number of the multi-valued variable that undergoes the transformation; a second matrix element of the fourth matrix specified by a value before the transformation of the multi-valued variable that undergoes the transformation; and a third matrix element of the fourth matrix specified by a value after the transformation of the multi-valued variable.

10. A non-transitory computer-readable recording medium storing a program for causing a computer to execute a process, the process comprising: obtaining evaluation function information stored in a storage unit of the computer, the evaluation function information indicating an evaluation function of a combinatorial optimization problem expressed by using a plurality of multi-valued variables; as well as generating transition destination candidates that are part of respective transitionable ranges of values ​​of the plurality of multi-valued variables based on the evaluation function information; calculating, for each of the plurality of multi-valued variables, an amount of change in the value of the evaluation function associated with transition to the transition destination candidate based on the evaluation function information; specifying a multi-valued variable to be subjected to the transition from among the plurality of multi-valued variables based on the change amount; and The value of the designated multi-valued variable is converted into the value of the conversion destination candidate.

11. A data processing method implemented by a computer, the data processing method comprising: obtaining evaluation function information stored in a storage unit of the computer, the evaluation function information indicating an evaluation function of a combinatorial optimization problem expressed by using a plurality of multi-valued variables; as well as generating transition destination candidates that are part of respective transitionable ranges of values ​​of the plurality of multi-valued variables based on the evaluation function information; calculating, for each of the plurality of multi-valued variables, an amount of change in the value of the evaluation function associated with transition to the transition destination candidate based on the evaluation function information; specifying a multi-valued variable to be subjected to the transition from among the plurality of multi-valued variables based on the change amount; and The value of the designated multi-valued variable is converted into the value of the conversion destination candidate.