COMPUTER-READABLE RECORDING MEDIUM STORING PROGRAM, DATA PROCESSING APPARATUS, AND DATA PROCESSING METHOD

Abstract

An apparatus performs processing including: specifying, for each constraint term in an evaluation function of a combinatorial optimization problem, a first auxiliary variable representing an input amount subjected to a constraint of each constraint term; generating a second auxiliary variable obtained by quantizing the first auxiliary variable; for a first constraint term, with which the second auxiliary variable changes in association with a change in a first state variable of the evaluation function, calculating a difference of a contribution amount of the first constraint term, to a first change amount in the evaluation function associated with the change in the first state variable; and based on the difference, calculating the first change amount or updating a local field to be used to specify a second change amount in the evaluation function associated with the change in each state variable.

Description

CROSS-REFERENCE TO RELATED APPLICATION

This application is based upon and claims the benefit of priority of the prior Japanese Patent Application No. 2023-84203, filed on May 22, 2023, the entire contents of which are incorporated herein by reference.

FIELD

The embodiments discussed herein are related to a non-transitory computer-readable recording medium storing a program, a data processing apparatus, and a data processing method.

BACKGROUND

Some combinatorial optimization problems have a constraint condition supposed to be satisfied by a solution, and an approach for performing a search in consideration of the constraint condition has been proposed. Examples of the constraint condition include an inequality constraint, an equality constraint, an absolute value constraint, and the like. An evaluation function reflecting a constraint condition can be represented by following Formula (1).

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}1\right]& \\ H\left(x\right)=E\left(x\right)+P_{\mathrm{tot}}\left(x\right)& \left(1\right)\end{array}$

The first term on the right side of Formula (1) is a cost term, and the second term on the right side is a total value of a plurality of constraint terms. The total value of the plurality of constraint terms can be represented by following Formula (2).

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}2\right]& \\ P_{\mathrm{tot}}\left(x\right)=\sum _{k=1}^m{\lambda }_k{P}_k\left(x\right)& \left(2\right)\end{array}$

In Formula (2), P_k(x) denotes a penalty function of a constraint term relating to a constraint having k as the identification number. P_k(x) is represented by a nonlinear function. The summed number of constraint terms is represented by m. In addition, λ_k denotes a predetermined coefficient (hereinafter referred to as a constraint coefficient) for the constraint having k as the identification number. Hereinafter, λ_kP_k(x) will be referred to as a constraint term.

In a case where the constraint condition is an inequality constraint, P_k(x) in Formula (2) can be represented by following Formula (3).

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}3\right]& \\ P_k\left(x\right)=\max \left(0,P_{\mathrm{in}}^{\left(k\right)}\left(x\right)\right)& \left(3\right)\end{array}$

In a case where the constraint condition is an equality constraint, P_k(x) in Formula (2) can be represented by following Formula (4).

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}4\right]& \\ P_k\left(x\right)=❘P_{\mathrm{in}}^{\left(k\right)}\left(x\right)❘& \left(4\right)\end{array}$

In P_k(x) as indicated in Formulas (3) and (4), P_in^(k)(x) denotes an input amount (also referred to as a resource amount or a used resource amount) subjected to the constraint having k as the identification number. The input amount is represented by following Formula (5).

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}5\right]& \\ P_{\mathrm{in}}^{\left(k\right)}\left(x\right)=\sum _{i=1}^n{C}_{\mathrm{ki}}{x}_i+b^{\left(k\right)}& \left(5\right)\end{array}$

A coefficient (weight value) indicating a weight for x_i in the constraint having k as the identification number is denoted by C_ki. A coefficient representing an upper limit or the like of the input amount in the constraint having k as the identification number is denoted by b^(k).

Meanwhile, the cost term is represented by, for example, a function of a quadratic form as in following Formula (6).

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}6\right]& \\ E\left(x\right)=-\frac{1}{2}\sum _{i=1}^n\sum _{j=1}^n{W}_{\mathrm{ij}}{x}_i{x}_j-\sum _{i=1}^n{b}_i{x}_i& \left(6\right)\end{array}$

The first term on the right side is for integrating products of values (0 or 1) of two state variables and a weight value (representing strength of correlation between the two state variables) for all combinations of n state variables with neither omission nor duplication. A state variable having i as the identification number is denoted by x_i, a state variable having j as the identification number is denoted by x_j, and a weight value indicating magnitude of correlation between the state variables having i and j as the identification numbers is denoted by W_ij. The second term on the right side is for summing up products of a bias coefficient and a state variable for each identification number. A bias coefficient for i as the identification number is indicated by b_i.

In addition, a change amount (ΔE_i) of the cost term associated with the change in the value of x_i is represented by following Formula (7).

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}7\right]& \\ \Delta E_i=-\Delta x_i\left(\sum _j{W}_{\mathrm{ij}}{x}_j+b_i\right)=-\Delta x_i{h}_i& \left(7\right)\end{array}$

In Formula (7), when x_i changes to 0 from 1, Δx_i has −1, and when x_i changes to 1 from 0, Δx_i has 1. Note that h_i is called a local field, and ΔE_i is obtained by multiplying h_i by a sign (+1 or −1) according to Δx_i. For this reason, h_i can also be said to be a variable used to specify the change amount of the value of the evaluation function. By holding and appropriately updating such h_i, ΔE_i may be calculated without performing a product-sum operation.

In a search for a solution to a combinatorial optimization problem, a Markov-Chain Monte Carlo (MCMC) method is used. As a kind of the MCMC method, there is a simulated annealing method and a replica exchange method. In such a search by the MCMC method, when the change amount (ΔH_i) of the value of the evaluation function associated with the change in the value of x_i is smaller than a threshold value (sometimes called a noise value) obtained based on a random number and the value of a temperature parameter, a process of inverting the value of x_i is repeated. In this manner, a search for a state (a combination of the values of the state variables) in which the value of the evaluation function locally becomes the minimum is performed. The state in which the minimum value of local minimum values of the evaluation function is reached is regarded as an optimal solution.

When the evaluation function including a constraint term is used, ΔH_i denotes the total value of ΔE_i and the change amount of P_tot(x) associated with the change in the value of x_i.

Examples of the related art include: Japanese Laid-open Patent Publication No. 2020-204928, U.S. Patent Application Publication No. 2008/0065573, and International Publication Pamphlet No. WO 2020/196862.

SUMMARY

According to an aspect of the embodiments, there is provided a non-transitory computer-readable recording medium storing a program for causing a computer to execute processing including: acquiring, from a storage unit, evaluation function information on an evaluation function of a combinatorial optimization problem that includes a cost term and a plurality of constraint terms; specifying, for each of the plurality of constraint terms, values of a first auxiliary variable that represents an input amount subjected to a constraint that relates to each of the plurality of constraint terms, based on the evaluation function information; generating values of a second auxiliary variable obtained by quantizing the first auxiliary variable, based on the values of the first auxiliary variable; for a first constraint term, among the plurality of constraint terms, with which the values of the second auxiliary variable change in association with a change in values of a first state variable among a plurality of state variables of the evaluation function, calculating a first difference value of a contribution amount of the first constraint term associated with the change in the values of the first state variable, to a first change amount in values of the evaluation function associated with the change in the values of the first state variable; and calculating the first change amount by using the first difference value, or updating a local field to be used to specify a second change amount in the values of the evaluation function associated with the change in each of the plurality of state variables, by using the first difference value.

The object and advantages of the invention will be realized and attained by means of the elements and combinations particularly pointed out in the claims.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory and are not restrictive of the invention.

BRIEF DESCRIPTION OF DRAWINGS

FIG. 1 is a diagram illustrating an exemplary data processing device according to a first embodiment;

FIGS. 2A and 2B are diagrams illustrating an example of a relationship between P_k and y_k and a relationship between g_i^(k) and y_k in a case where a constraint condition is an equality constraint;

FIG. 3 is a diagram illustrating a hardware example of the data processing device;

FIG. 4 is a block diagram illustrating exemplary functions of the data processing device;

FIG. 5 is a flowchart illustrating an example of a processing procedure of a data processing device according to a second embodiment;

FIG. 6 is a diagram illustrating a search unit of a comparative example;

FIG. 7 is a flowchart illustrating a processing procedure of a data processing device including the search unit of the comparative example;

FIG. 8 is a diagram illustrating a hardware example of the search unit;

FIG. 9 is a diagram illustrating an example of an h-update circuit and a y_k update circuit;

FIG. 10 is a diagram illustrating an example of a g-calculation circuit that is an example of a g/Δg calculation circuit;

FIG. 11 is a diagram illustrating an example of a Δg calculation circuit that is an example of the g/Δg calculation circuit;

FIG. 12 is a timing chart illustrating an example of operation timings of the data processing device;

FIGS. 13A and 13B are diagrams illustrating an example of a relationship between P_k and y_k and a relationship between g_i^(k) and y_k in a case where the constraint condition is an inequality constraint and C_ki has a ternary value;

FIGS. 14A and 14B are diagrams illustrating an example of a relationship between P_k and y_k and a relationship between g_i^(k) and y_k in a case where the constraint condition is an equality constraint and C_ki has a ternary value; and

FIG. 15 is a diagram illustrating a Δg calculation circuit according to a modification.

DESCRIPTION OF EMBODIMENTS

The constraint term is represented using, for example, a nonlinear penalty function as indicated in Formula (3) or Formula (4). Therefore, it is difficult to calculate the change amount of the constraint term, using the local field (h_i) as in the cost term. For this reason, every time the change amount of the constraint term associated with the change in the value of x_i is computed, the input amount as indicated by, for example, Formula (5) will have to be computed with respect to each constraint term, also using the values of the state variables other than x_i. Consequently, the amount of computation is expanded in some cases.

In one aspect, an object of the embodiments is to reduce the amount of computation of a combinatorial optimization problem having a constraint condition.

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

A data processing device according to each embodiment illustrated below is designed to search for a solution to a combinatorial optimization problem having a constraint condition. As an example of the combinatorial optimization problem having a constraint condition, there is a knapsack problem.

First Embodiment

FIG. 1 is a diagram illustrating an exemplary data processing device according to a first embodiment.

A data processing device 10 according to the first embodiment includes a storage unit 11 and a processing unit 12.

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

The storage unit 11 stores evaluation function information 11a on an evaluation function of a combinatorial optimization problem including a cost term and a plurality of constraint terms. The evaluation function information 11a includes information on E(x) and P_tot(x) as indicated in Formula (1). In a case where E(x) is represented as Formula (6), the information on E(x) includes W_ij and b_i. The information on P_tot(x) includes, for example, λ_k and m in Formula (2) and C_ki and b^(k) in Formula (5).

Note that the storage unit 11 may further store a computational condition for the combinatorial optimization problem. For example, in a case where a search is performed by the replica exchange method, the computational condition includes the number of replicas, the values of a temperature parameter to be set in each replica, and the like.

The processing unit 12 can be implemented by an electronic circuit such as an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA), for example. 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), for example. The processor executes, for example, a program stored in a memory such as a RAM (which may be the storage unit 11). A set of processors may 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.

FIG. 1 illustrates an exemplary procedure of a process performed by the processing unit 12.

Step S1: The processing unit 12 acquires the evaluation function information 11a from the storage unit 11.

Step S2: The processing unit 12 specifies, for each of the plurality of constraint terms, the values (initial values) of a first auxiliary variable representing the input amount subjected to a constraint relating to each of the plurality of constraint terms, based on the evaluation function information 11a. The input amount subjected to the constraint having k as the identification number is represented by above-described Formula (5). Hereinafter, the input amount is represented by the first auxiliary variable (y_k). That is, y_k is represented by following Formula (8).

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}8\right]& \\ y_k=P_{\mathrm{in}}^{\left(k\right)}\left(x\right)=\sum _{i=1}^n{C}_{\mathrm{ki}}{x}_i+b_k& \left(8\right)\end{array}$

Note that b_k in Formula (8) and b^(k) in Formula (5) are the same. The value of y_k is computed from C_ki, b_k, and the number (n) of state variables included in the evaluation function information 11a and the initial values of x₁ to x_n. The initial values of x₁ to x_n may be stored in the storage unit 11, or may be input from the outside of the data processing device 10, for example, by a user working on an input device. The initial values of x₁ to x_n may be, for example, all zero or all one. In addition, the initial values of x₁ to x_n may be random numbers.

FIG. 1 illustrates an example of a relationship between a penalty function (P_k) of an inequality constraint and y_k. The horizontal axis represents the value of y_k, and the vertical axis represents the magnitude of P_k.

P_k is represented by a function as indicated by Formula (3). Therefore, when y_k representing the input amount is 0 or less, P_k=0 is met, and when y_k is greater than 0, P_k=y_k is met.

Step S3: The processing unit 12 generates a value of a second auxiliary variable (z_k) obtained by quantizing y_k, based on the value of y_k. For example, a ternary value indicating whether y_k is positive, negative, or zero is denoted by z_k. That is, z_k can be represented by following Formula (9).

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}9\right]& \\ z_k=\{\begin{array}{cc}0& \mathrm{if}{y}_k<0\\ 1& \mathrm{if}{y}_k=0\\ 2& \mathrm{if}{y}_k>0\end{array}& \left(9\right)\end{array}$

Note that z_k is not limited to a ternary value of 0, 1, and 2 and may be a ternary value of 11, 00, and 01. The reason for using such a ternary z_k will be described later.

Step S4: The processing unit 12 calculates a contribution amount of each of the plurality of constraint terms to a change amount (ΔH_i) of the value of the evaluation function associated with the change in the value of x_i, based on the value of y_k. The contribution amount can also be said to be a value representing the magnitude of contribution to the local field (h_i) used to specify ΔH_i. ΔH_i can be represented by following Formula (10).

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}10\right]& \\ \Delta H_i=-h_i\Delta x_i& \left(10\right)\end{array}$

ΔH_i is obtained by multiplying h_i by a sign (+1 or −1) according to Δx_i. Following Formula (11) can represent h_i.

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}11\right]& \\ h_i=\sum _{j=1}^n{W}_{\mathrm{ij}}{x}_j+b_i+\sum _{k=1}^m{g}_i^{\left(k\right)}& \left(11\right)\end{array}$

The contribution amount of the constraint term relating to the constraint with the identification number=k to ΔH_i(or h_i) is denoted by g_i^(k). In a case where the constraint condition is an inequality constraint, g_i^(k) can be represented by following Formula (12).

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}12\right]& \\ g_i^{\left(k\right)}=-{\lambda }_k\left[\max \left(0,y_k{❘}_{x_i\to x_i+\Delta x_i}\right)-\max \left(0,y_k\right)\right]/\Delta x_i& \left(12\right)\end{array}$

By multiplying the slope of the graph of P_k−y_k as illustrated in FIG. 1 by −λ_k, g_i^(k) is obtained. In a case where y_k has 0, in consideration of the fact that the slope of the graph of P_k−y_k differs depending on the direction in which x_i changes, g_i^(k) can be represented as following Formula (13).

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}13\right]& \\ g_i^{\left(k\right)}=\{\begin{array}{cc}0& \mathrm{if}{y}_k<0\\ -{\lambda }_k{C}_{\mathrm{ki}}\left(1-x_i\right)& \mathrm{if}{y}_k=0\\ -{\lambda }_k{C}_{\mathrm{ki}}& \mathrm{if}{y}_k>0\end{array}& \left(13\right)\end{array}$

A weight value indicating a weight for x_i in the constraint having k as the identification number is denoted by C_ki. Hereinafter, C_ki will be assumed to be a value representing whether or not x_i affects P_k. In a case where x_i affects P_k, C_ki=1 is met, and in a case where x_i does not affect P_k, C_ki=0 is met.

FIG. 1 illustrates a diagram illustrating an example of a relationship between g_i^(k) and y_k in a case where the constraint condition is an inequality constraint. The horizontal axis represents the value of y_k, and the vertical axis represents the value of g_i^(k). When y_k=0 is met, according to Formula (13), g_i^(k)=0 is met in a case of x_i=1, and g_i^(k)=−λ_kC_ki is met in a case of x_i=0.

The above-mentioned reason for using the ternary z_k is that g_i^(k) can be distinguished into three types of values, depending on whether y_k is positive, negative, or zero as described above.

Note that, in a case where Δg_i^(k) to be described later is calculated from Δz_k, the processing unit 12 does not have to compute g_i^(k) itself.

The processes in steps S1 to S4 described above are an initialization process, and thereafter the following search process is performed. Note that the values of y_k and z_k and g_i^(k) calculated in the processes in steps S2 to S4 are held in a holding unit (such as a register) (not illustrated) of the processing unit 12. The value of h_i represented by Formula (11) may also be held in the holding unit. In addition, the storage unit 11 may store the values of y_k, z_k, g_i^(k), or h_i.

Step S5: The processing unit 12 selects a transition candidate x_i from among x₁ to x_n. The processing unit 12 may randomly select x_i from x₁ to x_n, or may select one x_i in accordance with a predetermined rule. In addition, in the process in step S5, the processing unit 12 may perform, for example, the following processes in steps S6 to S8 in parallel for x₁ to x_n without selecting x_i.

Step S6: The processing unit 12 determines whether or not the value of z_k relating to any one constraint term of m constraint terms changes in association with a change in the value of x_i among x₁ to x_n. The value of y_k in a case where the value of x_i has changed is obtained by adding C_kiΔx_i to the original value of y_k. Then, depending on whether y_k in a case where the value of x_i has changed is positive, negative, or zero, the value of z_k in a case where the value of x_i has changed is obtained by Formula (9).

The processing unit 12 performs the process in step S7 when determining that the value of z_k changes and performs the process in step S8 when determining that the value of z_k does not change.

Step S7: The processing unit 12 calculates a difference value (Δg_i^(k)) of g_i^(k) associated with the change in the value of x_i. As can be seen from Formulas (9) and (13), the value of g_i^(k) changes when the value of z_k changes, and in other cases, the value does not change. The processing unit 12 may calculate a difference value between the held current g_i^(k) and g_i^(k) calculated based on the value of x_i after the change, in accordance with Formula (13). Alternatively, the processing unit 12 may calculate Δg_i^(k) according to a difference value (Δz_k) of z_k in accordance with following Formula (14).

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}14\right]& \\ \Delta g_i^{\left(k\right)}=\{\begin{array}{cc}-{\lambda }_k{C}_{\mathrm{ki}}\left(1-x_i\right)\Delta z_k& \mathrm{if}{z}_k:0\leftrightarrow 1\\ -{\lambda }_k{C}_{\mathrm{ki}}{x}_i\Delta z_k& \mathrm{if}{z}_k:1\leftrightarrow 2\end{array}& \left(14\right)\end{array}$

Two types of values of Δg_i^(k) in a case where z_k has changed between 0 and 1 and a case where z_k has changed between 1 and 2 are obtained. In a case where Δg (k) is calculated in accordance with Formula (14), the processing unit 12 does not have to compute g_i^(k) itself in the process in step S4.

Step S8: The processing unit 12 determines whether or not to permit the change in the value of x_i (transition acceptability). The transition acceptability is confirmed based on an added value (ΔH_i) of the change amount (ΔE_i) of the cost term associated with the change in the value of x_i and the change amount (ΔP_k) of P_k calculated based on Δg_i^(k). ΔE_i can be computed in accordance with Formula (7). ΔP_k can be calculated by multiplying Δg_i^(k) by −Δx_i. Note that, in a case where h_i represented by Formula (11) is held, the processing unit 12 may calculate ΔH_i by multiplying h_i by −Δx_i.

Then, for example, in a case where ΔH_i is smaller than a threshold value obtained based on a random number and the value of the temperature parameter, the processing unit 12 permits the change in the value of the transition candidate state variable. For example, log (rand)×T is used as the threshold value. A random number value is denoted by rand, and a value of the temperature parameter is denoted by T.

In a case where the processes in steps S6 to S8 are performed for x₁ to x_n in parallel, ΔH_i(i=1 to n) corresponding to changes in x₁ to x_n is calculated. Then, a change in the value of any of x₁ to x_n is permitted based on ΔH_i.

Step S9: The processing unit 12 updates the value of x_i when determining to permit the change in the value of x_i. In addition, the processing unit 12 updates the values of y_k and z_k according to the change in the value of x_i. In a case where the value of g_i^(k) or h_i is held, the processing unit 12 updates these values, using Δg_i^(k). The processing unit 12 may update the value of the evaluation function.

After the process in step S9, the processing unit 12 repeats the processes from step S5.

In a case where the simulated annealing method is performed, the processing unit 12 decreases the value of the above-mentioned temperature parameter in accordance with a predetermined temperature parameter switching schedule while the processes in steps S5 to S9 are repeated. Then, the processing unit 12 outputs, for example, the values of x₁ to x_n when a predetermined search end condition is satisfied, as a search result for a solution to the combinatorial optimization problem (for example, displays the values on a display device (not illustrated)). Note that the processing unit 12 may update, for example, the value of the evaluation function every time the transition is accepted and hold the minimum value until then and the values of x₁ to x_n when that minimum value was obtained. In that case, the processing unit 12 may output the values of x₁ to x_n corresponding to the minimum value stored when the search end condition is satisfied, as a search result.

In a case where the processing unit 12 performs the replica exchange method, the processing unit 12 performs the processes in steps S5 to S9 illustrated in FIG. 1 for each of a plurality of replicas set with different values of the temperature parameter from each other. Then, the processing unit 12 performs replica exchange each time the processes in steps S5 to S9 are repeated a predetermined number of times. For example, the processing unit 12 randomly selects two of the plurality of replicas. Then, the processing unit 12 exchanges the values of the temperature parameter or the values of x₁ to x_n between the two selected replicas with a predetermined exchange probability based on a difference in the values of the evaluation function or a difference in the values of the temperature parameter between the replicas. For example, the processing unit 12 updates the value of the evaluation function every time the transition is accepted in each replica and holds the minimum value until then and the values of x₁ to x_n when that minimum value was obtained. Then, for example, when a predetermined search end condition is satisfied, the processing unit 12 outputs, as a search result, the values of x₁ to x_n corresponding to a smallest minimum value in all the replicas among the minimum values stored in each replica.

As described above, the processing unit 12 of the data processing device 10 according to the first embodiment acquires, from the storage unit 11, the evaluation function information 11a on the evaluation function of the combinatorial optimization problem including the cost term and the plurality of constraint terms. Then, the processing unit 12 specifies, for each of the plurality of constraint terms, the values of the first auxiliary variable (y_k) representing the input amount subjected to a constraint relating to each of the plurality of constraint terms, based on the evaluation function information 11a. In addition, the processing unit 12 generates a value of the second auxiliary variable (z_k) obtained by quantizing y_k, based on the value of y_k. Then, for the constraint term among the plurality of constraint terms, with which the value of z_k changes in association with the change in the value of x_i among x₁ to x_n, the processing unit 12 calculates the difference value (Δg_i^(k)) of the contribution amount of the constraint term associated with the change in the value of x_i to the change amount of the value of the evaluation function associated with the change in the value of x_i. The processing unit 12 calculates ΔH_i, using Δg_i^(k), or updates h_i (i=1 to n), using Δg_i^(k).

This ensures that, only for the constraint term with which the value of z_k changes in association with the change in the value of x_i, the change amount of the constraint term is calculated or Δg_i^(k) used to update h_i is calculated. Therefore, the processing unit 12 no longer has to calculate ΔH_i in association with a change in the value of x_i or perform computation using the values of all the state variables for all the constraint terms every time h_i (i=1 to n) is updated. Accordingly, the amount of computation may be reduced.

For example, a case where the number of state variables is n=100, the summed number of constraint terms is m=10, and each constraint term is related to one state variable (a case where other than one of 100 instances of C_ki for each constraint term has 0) will be examined. In this case, the amount of computation may be reduced to 1/(10×100) as compared with a case of performing computation using the values of all the state variables for all the constraint terms.

In addition, since the amount of computation may be reduced, an increase in the amount of hardware materials to perform computation may be suppressed.

In addition, since the ternary z_k or Δz_k is used to calculate Δg_i^(k), accumulation of errors may be suppressed as compared with a case of computing the change amount of the constraint term using the input amount that is a real variable as indicated by Formula (5).

Such a data processing device 10 may be expected to be useful as a technique for suppressing the amount of computation and the amount of hardware materials and obtaining an accurate solution when solving various problems in modern society that can be transformed into a combinatorial optimization problem having a constraint condition.

Note that, in the above example, description has been given by assuming that the constraint condition is an inequality constraint, but the constraint condition is not limited to this. The constraint condition may be an equality constraint.

FIGS. 2A and 2B are diagrams illustrating an example of a relationship between P_k and y_k and a relationship between g_i^(k) and y_k in a case where the constraint condition is an equality constraint.

In FIG. 2A illustrating a relationship between P_k and y_k, the horizontal axis represents the value of y_k, and the vertical axis represents the value of P_k. In a case where the constraint condition is an equality constraint, P_k is represented by a function as indicated by Formula (4). Therefore, P_k=|y_k| is met.

In a case where the constraint condition is an equality constraint, g_i^(k) and Δg_i^(k) can be represented as following Formulas (15) and (16).

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}15\right]& \\ g_i^{\left(k\right)}=\{\begin{array}{cc}{\lambda }_k^{-}{C}_{\mathrm{ki}}& \mathrm{if}{z}_k=0\\ -{\lambda }_k^{-}{C}_{\mathrm{ki}}{x}_i-{\lambda }_k^{+}{C}_{\mathrm{ki}}\left(1-x_i\right)& \mathrm{if}{z}_k=1\\ -{\lambda }_k^{+}{C}_{\mathrm{ki}}& \mathrm{if}{z}_k=2\end{array}& \left(15\right)\end{array}$ $\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}16\right]& \\ \Delta g_i^{\left(k\right)}=\{\begin{array}{cc}-\left({\lambda }_k^{-}+{\lambda }_k^{+}\right)C_{\mathrm{ki}}\left(1-x_i\right)\Delta z_k& \mathrm{if}{z}_k:0\leftrightarrow 1\\ -\left({\lambda }_k^{-}+{\lambda }_k^{+}\right)C_{\mathrm{ki}}{x}_i\Delta z_k& \mathrm{if}{z}_k:1\leftrightarrow 2\end{array}& \left(16\right)\end{array}$

A constraint coefficient in a case of y_k=P_in^(k)<0, that is, in a case of z_k=0 is denoted by λ_k⁻. A constraint coefficient in a case of y_k=P_in^(k)>0, that is, in a case of z_k=2 is denoted by λ_k⁺.

In FIG. 2B illustrating a relationship between g_i^(k) and y_k, the horizontal axis represents the value of y_k, and the vertical axis represents the value of g_i^(k).

When y_k=0 is met, according to Formula (15), g_i^(k)=+λ_k C_ki is met in a case of x_i=1, and g_i^(k)=−λ_k⁺C_ki is met in a case of x_i=0.

In a case where the constraint condition is an equality constraint, by using such g_i^(k) and Δg_i^(k), an effect similar to the effect in a case where the constraint condition is an inequality constraint may be obtained.

Second Embodiment

Next, a data processing device according to a second embodiment will be described.

FIG. 3 is a diagram illustrating a hardware example of the data processing device.

A data processing device 20 is, for example, a computer and includes a CPU 21, a RAM 22, an HDD 23, a GPU 24, an input interface 25, a medium reader 26, and a communication interface 27. The above units are coupled to a bus.

The CPU 21 is a processor including an arithmetic circuit that executes program instructions. The CPU 21 loads at least a part of a program and data stored in the HDD 23 into the RAM 22 to execute the program. Note that the CPU 21 may include, for example, a plurality of processor cores in order to execute processes for a plurality of replicas in parallel. In addition, the data processing device 20 may include a plurality of processors. Note that a set of the plurality of processors (multiprocessor) may be called a “processor”.

The RAM 22 is a volatile semiconductor memory that temporarily stores a program executed by the CPU 21 and data used by the CPU 21 for arithmetic operations. Note that the data processing device 20 may include a memory of a type other than the RAM 22 or may include a plurality of memories.

The HDD 23 is a nonvolatile storage device that stores software programs such as an operating system (OS), middleware, and application software, and data. The programs include, for example, a program for causing the data processing device 20 to perform a process of searching for a solution to a combinatorial optimization problem as described above. Note that the data processing device 20 may include another type of storage device such as a flash memory or a solid state drive (SSD) or may include a plurality of nonvolatile storage devices.

The GPU 24 outputs an image (for example, an image representing a computation result for the combinatorial optimization problem, or the like) to a display 24a coupled to the data processing device 20 in accordance with an instruction from the CPU 21. As the display 24a, a cathode ray tube (CRT) display, a liquid crystal display (LCD), a plasma display panel (PDP), an organic electro-luminescence (OEL) display, or the like can be used.

The input interface 25 acquires an input signal from an input device 25a coupled to the data processing device 20 and outputs the acquired input signal to the CPU 21. As the input device 25a, a pointing device such as a mouse, a touch panel, a touch pad, or a trackball, a keyboard, a remote controller, a button switch, or the like can be used. In addition, a plurality of types of input devices may be coupled to the data processing device 20.

The medium reader 26 is a reading device that reads a program or data recorded on a recording medium 26a. As the recording medium 26a, for example, a magnetic disk, an optical disc, a magneto-optical disk (MO), a semiconductor memory, or the like can be used. Examples of the magnetic disk include a flexible disk (FD) and an HDD. Examples of the optical disc include a compact disc (CD) and a digital versatile disc (DVD).

The medium reader 26 copies, for example, a program or data read from the recording medium 26a to another recording medium such as the RAM 22 or the HDD 23. The read program is executed by, for example, the CPU 21. Note that the recording medium 26a may be a portable recording medium and is sometimes used for distribution of a program or data. In addition, the recording medium 26a and the HDD 23 will be sometimes referred to as computer-readable recording media.

The communication interface 27 is an interface that is coupled to a network 27a and communicates with another information processing device via the network 27a. The communication interface 27 may be a wired communication interface coupled by a cable to a communication device such as a switch or may be a wireless communication interface coupled by a wireless link to a base station.

Next, functions of the data processing device 20 will be described.

FIG. 4 is a block diagram illustrating exemplary functions of the data processing device.

The data processing device 20 includes an input unit 31, a storage unit 32, an initialization processing unit 33, a search unit 34, and an output unit 35. The storage unit 32 is equipped using, for example, the RAM 22 or the HDD 23. The input unit 31, the initialization processing unit 33, the search unit 34, and the output unit 35 are equipped using, for example, the CPU 21 and a program.

The storage unit 32 is a functional block as an example of the storage unit 11 illustrated in FIG. 1, and other functional blocks are functional blocks as an example of the processing unit 12 illustrated in FIG. 1.

The input unit 31 acquires evaluation function information on an evaluation function of a combinatorial optimization problem including a cost term and a plurality of constraint terms. The evaluation function information includes information on E(x) and P_tot(x) as indicated in Formula (1). In a case where E(x) is represented as Formula (6), the information on E(x) includes W_ij and b_i. The information on P_tot(x) includes, for example, λ_k and m in Formula (2) and C_ki and b^(k) in Formula (5).

The input unit 31 further acquires a computational condition for the combinatorial optimization problem. For example, in a case where a search is performed by the replica exchange method, the computational condition includes the number of replicas, the values of the temperature parameter to be set in each replica, and the like.

In addition, the input unit 31 may acquire the initial values of x₁ to x_n input by a user working on the input device 25a. The initial values of x₁ to x_n may be, for example, all zero or all one. In addition, the initial values of x₁ to x_n may be random numbers.

The storage unit 32 stores the evaluation function information and the like acquired by the input unit 31.

The initialization processing unit 33 acquires the evaluation function information and the initial values of x₁ to x_n stored in the storage unit 32 and performs an initialization process for the search unit 34, based on these acquired evaluation function information and initial values. The initialization process includes, for example, the following processes.

The initialization processing unit 33 calculates an initial value of y_k represented by above-described Formula (8) for each of the plurality of constraint terms, based on the evaluation function information and the initial values of x₁ to x_n. In addition, the initialization processing unit 33 calculates an initial value of z_k represented by above-described Formula (9), based on the initial value of y_k. The initialization processing unit 33 may calculate the initial value of y_k, based on the evaluation function information, and may also calculate an initial value of g_i^(k) in accordance with Formula (13).

The search unit 34 includes a transition candidate selection unit 34a, a ΔE calculation unit 34b, a Δg calculation unit 34c, a ΔH calculation unit 34d, a transition acceptability determination unit 34e, a state update unit 34f, and an H-update unit 34g.

The transition candidate selection unit 34a selects a transition candidate x_i from among x₁ to x_n. The transition candidate selection unit 34a may randomly select x_i from x₁ to x_n, or may select one x_i in accordance with a predetermined rule. In addition, the transition candidate selection unit 34a outputs Δx_i=1−2x_i that is a difference value between the value of the selected x_i and the value of x_i when the value of x_i is changed.

The ΔE calculation unit 34b receives Δx_i and calculates ΔE_i, based on Formula (7).

The Δg calculation unit 34c receives Δx_i and finds a value of y_k in a case where the value of x_i has changed. The value of y_k in a case where the value of x_i has changed is obtained by adding C_kiΔx_i to the original value of y_k. Then, the Δg calculation unit 34c finds the value of z_k in accordance with Formula (9) depending on whether the value of y_k in a case where the value of x_i has changed is positive, negative, or zero. The Δg calculation unit 34c calculates Δg_i^(k) in a case where the value of z_k has changed from the original value of z_k.

The Δg calculation unit 34c may calculate a difference value between the held current g_i^(k) and g_i^(k) calculated based on the value of x_i after the change, in accordance with Formula (13). Alternatively, the Δg calculation unit 34c may calculate Δg_i^(k) in accordance with above-described Formula (14) according to the difference value (Δz_k) of z_k. The Δg calculation unit 34c outputs ΔP_k calculated by multiplying Δg_i^(k) by −Δx_i.

The ΔH calculation unit 34d adds ΔE_i and ΔP_k to calculate the change amount (ΔH_i) of the evaluation function associated with the change in the value of x_i.

The transition acceptability determination unit 34e determines whether or not to permit the change in the value of x_i. For example, in a case where ΔH_i is smaller than a threshold value obtained based on a random number and the value of the temperature parameter, the transition acceptability determination unit 34e permits the change in the value of the transition candidate state variable.

The state update unit 34f updates the value of x_i when it is determined to permit the change in the value of x_i. Note that, in a case where the change in the value of x_i is permitted, the values of y_k and z_k are also updated to values calculated as described above. When holding the value of g_i^(k) or h_i, the processing unit 12 also updates these values according to the change in the value of x_i.

In a case where the change in the value of x_i is permitted, the H-update unit 34g updates H by adding ΔH_i to the original value (H) of the evaluation function.

The output unit 35 outputs a search result (for example, the values of x₁ to x_n corresponding to the minimum value of the evaluation function obtained until then) when a search end condition is satisfied. For example, the output unit 35 may output the search result to the display 24a to display the output search result on the display 24a, transmit the search result to another information processing device via the network 27a, or store the search result in an external storage device.

FIG. 5 is a flowchart illustrating an example of a processing procedure of the data processing device according to the second embodiment.

Step S10: The input unit 31 acquires the evaluation function information on an evaluation function of a combinatorial optimization problem including a cost term and a plurality of constraint terms.

Step S11: The initialization processing unit 33 performs the initialization process as described above, based on the evaluation function information and the initial values of x₁ to x_n stored in the storage unit 32.

Step S12: The transition candidate selection unit 34a selects a transition candidate x_i from among x₁ to x_n.

Step S13: The ΔE calculation unit 34b calculates ΔE_i, based on Formula (7).

Step S14: The Δg calculation unit 34c finds a value of y_k in a case where the value of x_i has changed and finds a value of z_k in accordance with Formula (9). Then, the Δg calculation unit 34c determines whether or not the value of z_k has changed from the original value of z_k. When determining that the value of z_k has changed from the original value of z_k, the Δg calculation unit 34c performs the process in step S15. When determining that the value of z_k has not changed from the original value of z_k, the Δg calculation unit 34c performs the process in step S16.

Step S15: The Δg calculation unit 34c calculates a difference value between the held current g_i^(k) and g_i^(k) calculated based on the value of x_i after the change, in accordance with Formula (13). Alternatively, the Δg calculation unit 34c calculates Δg_i^(k) in accordance with above-described Formula (14) according to the difference value (Δz_k) of z_k.

Step S16: The ΔH calculation unit 34d calculates ΔH_i by adding ΔE_i and ΔP_k calculated by multiplying Δg_i^(k) by −Δx_i.

Step S17: The transition acceptability determination unit 34e determines whether or not to permit the change in the value of x_i. In a case where the transition acceptability determination unit 34e determines to permit the change in the value of x_i, the process in step S18 is performed. In a case where the transition acceptability determination unit 34e determines not to permit the change in the value of x_i, the process in step S19 is performed.

Step S18: The above-described update process is performed by the state update unit 34f and the H-update unit 34g.

Step S19: A determination unit (not illustrated) of the search unit 34 determines whether or not a search end condition is satisfied. For example, in a case where the process in step S12 has been repeated a predetermined number of times, or in a case where a value of the evaluation function smaller than a predetermined value has been obtained, it is determined that the search end condition is satisfied. In a case where it is determined that the search end condition is satisfied, the process in step S20 is performed. In a case where it is determined that the search end condition is not satisfied, the processes from step S12 are repeated.

In a case where the simulated annealing method is applied, while the processes in steps S12 to S19 are repeated, the value of the temperature parameter for generating a threshold value to be used by the transition acceptability determination unit 34e is decreased in accordance with a predetermined temperature parameter switching schedule.

In a case where the replica exchange method is applied, the processes in steps S12 to S19 are performed for each of the plurality of replicas set with different values of the temperature parameter from each other. Then, replica exchange is performed each time the processes in steps S12 to S19 are repeated a predetermined number of times. For example, two of the plurality of replicas are randomly selected, and the values of the temperature parameter or the values of x₁ to x_n are exchanged between the two selected replicas with a predetermined exchange probability based on a difference in the values of the evaluation function or a difference in the values of the temperature parameter between the replicas.

Step S20: The output unit 35 outputs a search result. This ends the process by the data processing device 20.

Note that the above order of processes is an example, and the order of processes may be appropriately altered.

COMPARATIVE EXAMPLE

FIG. 6 is a diagram illustrating a search unit of a comparative example. In FIG. 6, the same elements as the elements of the search unit 34 of the data processing device 20 illustrated in FIG. 4 are given the same reference signs.

In a search unit 40 of the comparative example, a transition candidate selection unit 41 selects a transition candidate x_i from among x₁ to x_n, similarly to the transition candidate selection unit 34a. The transition candidate selection unit 41 outputs the values (x′) of x₁ to x_n when the value of the selected x_i is changed and additionally, outputs Δx_i that is a difference value when the value of x_i is changed.

Unlike the search unit 34, the search unit 40 of the comparative example includes a P_tot calculation unit 42, a P_tot0 update unit 43, and a ΔP calculation unit 44.

The P_tot calculation unit 42 calculates the total value of the m constraint terms represented by Formula (2), using the values (x′) of x₁ to x_n when the value of the selected x_i is changed.

The P_tot0 update unit 43 updates and outputs P_tot0 that is the total value of the m constraint terms before the value of x_i changes. The value of P_tot0 is updated to the value of P_tot output by the P_tot calculation unit 42 when a transition acceptability determination unit 34e permits the change in the value of x_i.

The ΔP calculation unit 44 calculates and outputs ΔP_tot that is a difference between P_tot and P_tot0.

In the search unit 40 of the comparative example, a ΔH calculation unit 45 calculates and outputs ΔH_i by adding ΔE_i and ΔP_tot.

FIG. 7 is a flowchart illustrating a processing procedure of a data processing device including the search unit of the comparative example.

The processes in steps S30 to S33 are the same as the processes in steps S10 to S13 illustrated in FIG. 5.

Step S34: The P_tot calculation unit 42 calculates the total value (P_tot) of the m constraint terms represented by Formula (2), using the values (x′) of x₁ to x_n when the value of the selected x_i is changed.

Step S35: The ΔP calculation unit 44 calculates ΔP_tot that is a difference between P_tot and P_tot0.

Step S36: the ΔH calculation unit 45 calculates ΔH_i by adding ΔE_i and ΔP_tot.

The process in step S37 is the same as the process in step S17 illustrated in FIG. 5.

Step S38: An update process for P_tot0 by the P_tot0 update unit 43, as well as the update process by a state update unit 34f and an H-update unit 34g, are performed.

The subsequent processes in steps S39 and S40 are the same as the processes in steps S19 and S20 illustrated in FIG. 5.

In such a search unit 40 of the comparative example, in a case where the transition candidate x_i is selected, since P_tot is calculated also using the values of the state variables other than x_i, for example, the input amount as indicated by Formula (5) will be computed for all the constraint terms.

On the other hand, in the data processing device 20 of the second embodiment, the search unit 34 computes Δg_i^(k) to be used to calculate the change amount of the constraint term only for the constraint term with which the value of z_k changes in association with the change in the value of x_i. Therefore, the processing unit 12 no longer has to perform computation using the values of all the state variables for all the constraint terms every time ΔH_i is calculated in association with the change in the value of x_i. Accordingly, the amount of computation may be reduced. In addition, since the amount of computation may be reduced, an increase in the amount of hardware materials to perform computation may be suppressed.

In addition, since the ternary z_k or Δz_k is used to compute Δg_i^(k), accumulation of errors may be suppressed as compared with a case of computing the change amount of the constraint term using the input amount that is a real variable as indicated by Formula (5).

(Example of Equipping Search Unit 34 by Hardware)

The search unit 34 can also be equipped by, for example, hardware as follows. Note that the search unit 34 illustrated below holds h_i as indicated by Formula (11) and uses the held h_i to calculate ΔH_i.

FIG. 8 is a diagram illustrating a hardware example of the search unit.

The search unit 34 includes a state variable update circuit 51, an h-update circuit 52, a y_k update circuit 53, a ΔH calculation circuit 54, a selection circuit 55, and a controller 56. Note that FIG. 8 illustrates an example of storing the evaluation function information in the storage unit 32. The storage unit 32 stores n×n instances of W_ij to be used to update h_i. Furthermore, the storage unit 32 stores m×n coupling coefficients (C_ik) from z_k to x_i and n×m coupling coefficients (C_kj) from x_j to y_k.

The state variable update circuit 51 holds the values of x₁ to x_n. In addition, the state variable update circuit 51 changes the value of the state variable with the identification number output by the selection circuit 55 under the control of the controller 56.

The h-update circuit 52 holds the value of h_i (i=1 to n) indicated by Formula (11). In addition, the h-update circuit 52 updates the value of h_i under the control of the controller 56.

The y_k update circuit 53 holds the value of y_k indicated by Formula (8). In addition, the y_k update circuit 53 updates the value of y_k under the control of the controller 56.

More specific examples of the h-update circuit 52 and the y_k update circuit 53 will be described later (refer to FIG. 9).

The ΔH calculation circuit 54 calculates ΔH_i(i=1 to n) by multiplying h_i by −Δx_i.

The selection circuit 55 selects one state variable that permits the change in the value from among x₁ to x_n, based on ΔH_i associated with the change in the value of each of x₁ to x_n and a threshold value supplied from the controller 56.

For example, the selection circuit 55 computes max (0, ΔH_i) for each of ΔH₁ to ΔH_n. A function that maintains any of ΔH_i to ΔH_n having 0 or more as it is and transforms any of ΔH₁ to ΔH_n being negative into 0 is denoted by max (0, ΔH_i). The selection circuit 55 adds a threshold value (which may have n different values or may have the same value) to each of n instances of max (0, ΔH_i). The selection circuit 55 outputs the identification number of the state variable, among x₁ to x_n, that gives the minimum addition result when the value has changed.

However, the selection circuit 55 is not limited to one that performs the selection operation as described above. The selection circuit 55 may permit the change in the value of x_i in a case where ΔH_i is smaller than a threshold value. In a case where the change in the values of a plurality of instances of x_i is permitted, the selection circuit 55 may select one state variable randomly or in accordance with a predetermined rule and output the identification number of the selected state variable.

The controller 56 controls an operation timing of each circuit of the search unit 34 by sending a control signal or a clock signal. In addition, the controller 56 generates a threshold value and supplies the generated threshold value to the selection circuit 55. For example, log (rand)×T is used as the threshold value.

FIG. 9 is a diagram illustrating an example of the h-update circuit and the y_k update circuit.

The h-update circuit 52 includes a g/Δg calculation circuit 52a, a difference addition circuit 52b, and an h-holding circuit 52c.

The y_k update circuit 53 includes a difference addition circuit 53a and a y_k holding circuit 53b.

The g/Δg calculation circuit 52a calculates g_i^(k) or Δg_i^(k). An example of the g/Δg calculation circuit 52a will be described later (refer to FIG. 10).

In a case where the change in the value of x_j is permitted, the difference addition circuit 52b receives the control signal from the controller 56 and updates h_i by adding W_ijΔx_j to the original h_i. In a case where Δg_i^(k) occurs when the change in the value of x_j is permitted, the difference addition circuit 52b further adds Δg (k) to h_i to update h_i.

The h-holding circuit 52c holds n instances of h_i. In addition, the h-holding circuit 52c outputs h_i in synchronization with the clock signal.

In a case where the change in the value of x_j is permitted, the difference addition circuit 53a receives the control signal from the controller 56 and updates y_k by adding C_kjΔx_j to the original y_k.

The y_k holding circuit 53b holds m instances of y_k. In addition, the y_k holding circuit 53b outputs y_k in synchronization with the clock signal.

FIG. 10 is a diagram illustrating an example of a g-calculation circuit that is an example of the g/Δg calculation circuit.

The g-calculation circuit 52a1 includes a multiplier 60, a selection circuit 61, and a multiplier 62.

The multiplier 60 outputs a product of 1 and 1−x_i.

The selection circuit 61 receives y_k and outputs 0 if y_k<0 is met, outputs 1−x_i if y_k=0 is met, and outputs 1 if y_k >0 is met.

The multiplier 62 outputs a product of the output of the selection circuit 61 and −λ_kC_ki, as g_i^(k) indicated by Formula (13).

FIG. 11 is a diagram illustrating an example of a Δg calculation circuit that is an example of the g/Δg calculation circuit.

The Δg calculation circuit 52a2 includes a z_k generation circuit 65 and a Δg generation circuit 66.

The z_k generation circuit 65 receives y_k and outputs z_k=0 if y_k<0 is met, outputs z_k=1 if y_k=0 is met, and outputs z_k=2 if y_k >0 is met.

The Δg generation circuit 66 holds z_k. In a case where the value of z_k generated by the z_k generation circuit 65 and the held value of z_k have changed between 0 and 1, the Δg generation circuit 66 selects 1-x_i and outputs −λ_kC_ki(1−x_i)Δz_k, as Δg_i^(k). In a case where the value of z_k generated by the z_k generation circuit 65 and the held value of z_k have changed between 1 and 2, the Δg generation circuit 66 selects x_i and outputs −Δ_kC_kix_iΔz_k, as Δg_i^(k).

FIG. 12 is a timing chart illustrating an example of operation timings of the data processing device. FIG. 12 illustrates an example of operation timings in a case where the search unit 34 having the hardware configuration illustrated in FIG. 8 is used. FIG. 12 illustrates a processing example in a case where 15 replicas are used. The search unit 34 performs processes in order from the replica with the replica number=0 while shifting the timing by a pipeline process.

From a timing t1, a process in which the ΔH calculation circuit 54 calculates ΔH_i(i=1 to n) from h_i is performed.

From a timing t2, a process in which the selection circuit 55 selects the state variable (x_j) that permits the change in the value, from among x₁ to x_n, is performed.

From a timing t3, a process in which the state variable update circuit 51 changes the value (inverts 0 to 1 or 1 to 0) of x_j selected by the selection circuit 55 is performed.

From a timing t4, a process in which the h-update circuit 52 and the y_k update circuit 53 update h_i and y_k is performed.

From a timing t5, a process in which the h-update circuit 52 calculates z_k from y_k is performed.

From a timing t6, a process in which the h-update circuit 52 calculates Δg_i^(k) from z_k and x_i is performed.

From a timing t7, a process in which the h-update circuit 52 further updates h_i, using Δg_i^(k), is performed.

Thereafter, similar processes are repeated.

Each of the above processes can be started at the time point when the previous process for the replica with the replica number=0 has ended even if the immediately preceding process is not completed for all 15 replicas.

By such a hardware process, Δg_i^(k) to be used to update h_i is calculated only for the constraint term with which the value of z_k changes in association with the change in the value of x_j. Therefore, the processing unit 12 no longer has to perform computation using the values of all the state variables for all the constraint terms every time h_i (i=1 to n) is updated. Accordingly, the amount of computation may be reduced.

Note that the hardware process as described above can also be performed by the CPU 21 executing a program.

(Modifications)

In the above description, C_ki used to calculate y_k in Formula (8) is assumed to have 1 when x_i affects the constraint with the identification number=k and have 0 when x_i does not affect the constraint with the identification number=k.

However, C_ki may be a ternary value of −1, 0, or 1. In the following, it is assumed that C_ki=1 is met in a case where x_i gives a positive contribution to the constraint with the identification number=k, C_ki=−1 is met in a case where x_i gives a negative contribution to the constraint with the identification number=k, and C_ki=0 is met in a case where x_i does not affect the constraint with the identification number=k.

FIGS. 13A and 13B are diagrams illustrating an example of a relationship between P_k and y_k and a relationship between g_i^(k) and y_k in a case where the constraint condition is an inequality constraint and C_ki has a ternary value.

In FIG. 13A illustrating a relationship between P_k and y_k, the horizontal axis represents the value of y_k, and the vertical axis represents the value of P_k. In a case where the constraint condition is an inequality constraint, P_k is represented by a function as indicated by Formula (3). Therefore, when y_k representing the input amount is 0 or less, P_k=0 is met, and when y_k is greater than 0, P_k=y_k is met.

In FIG. 13B illustrating a relationship between g_i^(k) and y_k, the horizontal axis represents the value of y_k, and the vertical axis represents the value of g_i^(k).

As in FIG. 13B, in a case of C_kiΔx_i>0, g_i^(k)=−λ_kC_ki is met when y_k=0. In a case of C_kiΔx_i<0, g_i^(k)=0 is met when y_k=0.

FIGS. 14A and 14B are diagrams illustrating an example of a relationship between P_k and y_k and a relationship between g_i^(k) and y_k in a case where the constraint condition is an equality constraint and C_ki has a ternary value.

In FIG. 14A illustrating a relationship between P_k and y_k, the horizontal axis represents the value of y_k, and the vertical axis represents the value of P_k. In a case where the constraint condition is an equality constraint, P_k is represented by a function as indicated by Formula (4). Therefore, P_k=|y_k| is met.

In FIG. 14B illustrating a relationship between g_i^(k) and y_k, the horizontal axis represents the value of y_k, and the vertical axis represents the value of g_i^(k).

As in FIG. 14B, in a case of C_kiΔx_i>0, g_i^(k)=−Δ_k+C_ki is met when y_k=0. In a case of C_kiΔx_i<0, g_i^(k)=+λ_k C_ki is met when y_k=0.

When the ternary C_ki as described above is used, since the change in y_k due to Δx_i is C_kiΔx_i, z_k generated from y_k can be represented by following Formula (17).

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}17\right]& \\ z_k=\left(z_k^{+},z_k^{-}\right)=\{\begin{array}{cc}\left(0,0\right)& \mathrm{if}{y}_k<0\\ \left(1,0\right)& \mathrm{if}{y}_k=0\\ \left(1,1\right)& \mathrm{if}{y}_k>0\end{array}& \left(17\right)\end{array}$

Assuming that z_k corresponding to Δx_i is z_i^(k), z_i^(k) can be represented by following Formula (18).

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}18\right]& \\ z_i^{\left(k\right)}=\{\begin{array}{cc}{z}_k^{+}& \mathrm{if}{C}_{\mathrm{ki}}\Delta x_i>0\\ z_k^{-}& \mathrm{if}{C}_{\mathrm{ki}}\Delta x_i<0\end{array}& \left(18\right)\end{array}$

By using such z_i^(k), Δg_i^(k) can be represented by following Formula (19).

$\begin{array}{cc}\left[\mathrm{Mathematical}\mathrm{Formula}19\right]& \\ \Delta g_i^{\left(k\right)}=-{\lambda }_k{C}_{\mathrm{ki}}\Delta z_i^{\left(k\right)}& \left(19\right)\end{array}$

FIG. 15 is a diagram illustrating a Δg calculation circuit according to a modification.

The Δg calculation circuit 52a3 includes a z_k generation circuit 70, a Δz_i^(k) generation circuit 71, and a multiplier 72.

The z_k generation circuit 70 receives y_k and, assuming z_k=(z_k⁺, z_k⁻), outputs (0, 0) if y_k<0 is met, outputs (1, 0) if y_k=0 is met, outputs (1, 1) if y_k >0 is met.

The Δz_i^(k) generation circuit 71 receives C_kiΔx_i. Then, if C_kiΔx_i>0 is met, the Δz_i^(k) generation circuit 71 outputs a difference value between z_k⁺ and the original z_i^(k), as Δz_i^(k). If C_kiΔx_i<0 is met, the Δz_i^(k) generation circuit 71 outputs a difference value between z_k⁻ and the original z_i^(k), as Δz_i^(k).

The multiplier 72 outputs a product of Δz_i^(k) and −Δ_kC_ki, as Δg_i^(k).

In a case where the ternary C_ki as described above is used, such a Δg calculation circuit 52a3 can be used instead of the Δg calculation circuit 52a2 illustrated in FIG. 11.

Note that, as described above, the processing contents of the data processing devices 10 and 20 of the above respective embodiments can be implemented by causing the data processing devices 10 and 20 to execute a program.

The program can be recorded beforehand on a computer-readable recording medium (for example, the recording medium 26a). As the recording medium, for example, a magnetic disk, an optical disc, a magneto-optical disk, a semiconductor memory, or the like can be used. Examples of the magnetic disk include an FD and an HDD. Examples of the optical disc include a CD, a CD-recordable (R)/rewritable (RW), a DVD, and a DVD-R/RW. The program is sometimes recorded on a portable recording medium and distributed. In that case, the program may be copied from the portable recording medium to another recording medium (for example, the HDD 23) and then executed.

While one form of the program, the data processing device, and the data processing method according to the embodiments has been described based on the embodiments, these embodiments are merely examples and are not limited to the above description.

For example, in the above description, the state variable is assumed to have a value of 0 or 1, but a variable having a value of −1 or +1 can also be used as the state variable by appropriately transforming the formulas.

All examples and conditional language provided herein are intended for the pedagogical purposes of aiding the reader in understanding the invention and the concepts contributed by the inventor to further the art, and are not to be construed as limitations to such specifically recited examples and conditions, nor does the organization of such examples in the specification relate to a showing of the superiority and inferiority of the invention. Although one or more embodiments of the present invention have been described in detail, it should be understood that the various changes, substitutions, and alterations could be made hereto without departing from the spirit and scope of the invention.

Claims

1. A non-transitory computer-readable recording medium storing a program for causing a computer to execute processing comprising: acquiring, from a storage unit, evaluation function information on an evaluation function of a combinatorial optimization problem that includes a cost term and a plurality of constraint terms;specifying, for each of the plurality of constraint terms, values of a first auxiliary variable that represents an input amount subjected to a constraint that relates to each of the plurality of constraint terms, based on the evaluation function information;generating values of a second auxiliary variable obtained by quantizing the first auxiliary variable, based on the values of the first auxiliary variable;for a first constraint term, among the plurality of constraint terms, with which the values of the second auxiliary variable change in association with a change in values of a first state variable among a plurality of state variables of the evaluation function, calculating a first difference value of a contribution amount of the first constraint term associated with the change in the values of the first state variable, to a first change amount in values of the evaluation function associated with the change in the values of the first state variable; andcalculating the first change amount by using the first difference value, or updating a local field to be used to specify a second change amount in the values of the evaluation function associated with the change in each of the plurality of state variables, by using the first difference value.

2. The non-transitory computer-readable recording medium according to claim 1, the processing further comprising: calculating the first change amount by adding a third change amount of the cost term associated with the change in the values of the first state variable and a fourth change amount of the first constraint term obtained from the first difference value;determining whether or not to permit the change in the values of the first state variable, based on a comparison result between the first change amount and a threshold value; andupdating the values of the first state variable when it is determined to permit the change in the values of the first state variable.

3. The non-transitory computer-readable recording medium according to claim 1, the processing further comprising: holding the local field; andupdating the local field by adding the first difference value to the local field for each of the plurality of state variables when the change in the values of the first state variable is permitted.

4. The non-transitory computer-readable recording medium according to claim 1, wherein the second auxiliary variable is a ternary value that indicates whether the values of the first auxiliary variable are positive, negative, or zero.

5. The non-transitory computer-readable recording medium according to claim 1, the processing further comprising: calculating a second difference value of the values of the second auxiliary variable associated with the change in the values of the first state variable; andcalculating the first difference value, based on the second difference value, the values of the first state variable, and a weight value between the first state variable and the first constraint term.

6. A data processing apparatus comprising: a storage unit configured to store evaluation function information on an evaluation function of a combinatorial optimization problem that includes a cost term and a plurality of constraint terms;a processing unit configured to perform processing including:specifying, for each of the plurality of constraint terms, values of a first auxiliary variable that represents an input amount subjected to a constraint that relates to each of the plurality of constraint terms, based on the evaluation function information;generating values of a second auxiliary variable obtained by quantizing the first auxiliary variable, based on the values of the first auxiliary variable;for a first constraint term, among the plurality of constraint terms, with which the values of the second auxiliary variable change in association with a change in values of a first state variable among a plurality of state variables of the evaluation function, calculating a first difference value of a contribution amount of the first constraint term associated with the change in the values of the first state variable, to a first change amount in values of the evaluation function associated with the change in the values of the first state variable; andcalculating the first change amount by using the first difference value, or updating a local field to be used to specify a second change amount in the values of the evaluation function associated with the change in each of the plurality of state variables, by using the first difference value.

7. A data processing method implemented by a computer, the data processing method comprising: acquiring, from a storage unit, evaluation function information on an evaluation function of a combinatorial optimization problem that includes a cost term and a plurality of constraint terms;specifying, for each of the plurality of constraint terms, values of a first auxiliary variable that represents an input amount subjected to a constraint that relates to each of the plurality of constraint terms, based on the evaluation function information;generating values of a second auxiliary variable obtained by quantizing the first auxiliary variable, based on the values of the first auxiliary variable;for a first constraint term, among the plurality of constraint terms, with which the values of the second auxiliary variable change in association with a change in values of a first state variable among a plurality of state variables of the evaluation function, calculating a first difference value of a contribution amount of the first constraint term associated with the change in the values of the first state variable, to a first change amount in values of the evaluation function associated with the change in the values of the first state variable; andcalculating the first change amount by using the first difference value, or updating a local field to be used to specify a second change amount in the values of the evaluation function associated with the change in each of the plurality of state variables, by using the first difference value.