Variable assignment device and variable assignment method

The variable assignment device efficiently allocates variables to parallel processing units, addressing the challenge of maintaining constraint satisfaction in large-scale combinatorial optimization problems, enhancing processing efficiency and solution quality.

JP7582496B2Active Publication Date: 2024-11-13NEC CORP
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
JP2023549197
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Filing Date
2021-09-21
Publication Date
2024-11-13
Estimated Expiration
2041-09-21

Smart Images

  • Figure 0007582496000009
    Figure 0007582496000009
  • Figure 0007582496000010
    Figure 0007582496000010
  • Figure 0007582496000011
    Figure 0007582496000011
Patent Text Reader

Abstract

This variable allocation device allocates variables of a combinatorial optimization problem to a plurality of parallel processing means to which variables of a combinatorial optimization problem are allocated and which perform processing in parallel to obtain the values of the allocated variables, the plurality of parallel processing means having a defined order. A variable allocation means 71 allocates, for each set of variables having a defined constraint, the variables belonging to the set to any one of the parallel processing means, and after having allocated all of the variables to any parallel processing means, transforms an index of the variables allocated to the parallel processing means in accordance with the order of the parallel processing means. A matrix transformation means 72 transforms a matrix used by an evaluation function of the combinatorial optimization problem in accordance with the post-variable-transformation index.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a variable assignment device, a variable assignment method, and a variable assignment program that assign variables for a combinatorial optimization problem to a plurality of devices performing processing in parallel. [Background technology]

[0002] One of the statistical mechanical models of magnetic materials is the Ising model, which is formulated by a variable that takes two values, 1 or -1, and a Hamiltonian given by its quadratic form. The Ising model describes the behavior of spin in magnetic materials. The Ising model can be applied as an evaluation function for many combinatorial optimization problems. For example, by applying the simulated annealing method to the Ising model, it is possible to solve general-purpose combinatorial optimization problems.

[0003] For example, Patent Literature 1 describes a general-purpose approach to solving combinatorial optimization problems using an Ising model. Also, Non-Patent Literature 1 describes a specific method for formulating various types of combinatorial optimization problems, including number partitioning problems, graph partitioning problems, satisfiability problems, traveling salesman problems, and the like, into the Ising model or a model equivalent thereto.

[0004] The probability distribution of the Ising model at a certain temperature T can be calculated by the Markov chain Monte Carlo method (MCMC). In the Markov chain Monte Carlo method, one spin is selected, and the change in the Hamiltonian when changing the spin state is calculated. Furthermore, the spin state is sampled with a probability distribution according to the temperature T and the Hamiltonian by a dynamical system that accepts or rejects state updates according to the probability according to the Boltzmann weight corresponding to the temperature.

[0005] Hereinafter, a description will be given of an example in which a general Markov chain Monte Carlo method is applied to an Ising model. First, notation and terminology will be explained.

[0006]

number

[0007] In formula (1), s i is the i-th variable. A number is assigned to each variable. This number is called an index. The subscript i represents the index. If there are N variables, then 1≦i≦N. In the Ising model, the value of a variable is either +1 or -1. In the fields of statistical mechanics and combinatorial optimization problems, each variable is also called a spin. s is a vector with all N variables as elements. Also, J ij is a constant that represents the bond between the i-th variable and the j-th variable. ij The smaller the value of is, the more likely it is that the two connected variables will have the same value. ij can be said to be elements of the matrix used in the evaluation function.

[0008] In the field of optimization, the Hamiltonian is called an evaluation function or cost function. The combination of variables that minimizes the evaluation function, as expressed in equation (2), is called the optimal solution or ground state. Non-Patent Document 1 shows Hamiltonians for many optimization problems. For example, the method shown in Non-Patent Document 1 can be used to find J ij Then, we obtain the optimal solution as shown in equation (2).

[0009]

number

[0010] The objective of the Markov chain Monte Carlo method is to sample state s with a probability proportional to the probability distribution (Boltzmann distribution) expressed by equation (3) for temperature T.

[0011]

number

[0012] The sampling operation includes the following five processes.

[0013] Step 1: Randomly determine the initial state s.

[0014] Step 2: Randomly choose one of the N variables.

[0015] Step 3: The state in which the selected variable is multiplied by -1 is defined as s', and the change in the evaluation function is calculated using equation (4).

[0016]

number

[0017] Process 4: Determine whether or not to accept the state transition from s to s' with the probability given by the following formula (5), and if it is determined to be accepted, update state s to state s'. Also, if it is determined to reject the state transition, keep the state as s.

number

[0018] Process 5. Repeat processes 2 to 4.

[0019] During the above operation, in the limit of performing the repeated process an infinite number of times, it is guaranteed that the occurrence probability of state s will converge to the desired probability distribution (equation (3)).

[0020] Also, a model in which variables take on the binary values ​​of 1 or 0 is often used as a model equivalent to equation (1). To distinguish between these models, the symbol s is used for variables that take on the binary values ​​of 1 or -1, and the symbol x is used for variables that take on the binary values ​​of 1 or 0, and symbols are used differently. Furthermore, the symbol J is used for the matrix used in the evaluation function of the former model, and the symbol Q is used for the matrix used in the evaluation function of the latter model, and symbols are used differently. Both models are equivalent, and the following will take the former model (Ising model) as an example.

[0021] Processes 1 to 5 are one embodiment of the Metropolis method of the Markov chain Monte Carlo method. There is arbitrariness in the method of selecting the next state from the current state (corresponding to process 2) and the method of selecting the probability of accepting the update that appears in process 4.

[0022] However, in order for the converged probability distribution to match the desired probability distribution, it is necessary to determine the method for selecting the next state and the acceptance probability so as to satisfy the two conditions of balance and ergodicity. In general, it is not easy to construct a dynamical system that satisfies these two conditions. Well-known methods include the Metropolis method and the heat bath method.

[0023] When a solution using the Markov chain Monte Carlo method is used, the processing time increases as the size of the target model increases. Therefore, a solution using the Markov chain Monte Carlo method requires some ingenuity to shorten the processing time. Processing of dynamical systems for large-scale models is sometimes performed using large-scale computers. Even when a large-scale computer is used, it is important to parallelize the processing with high efficiency in order to shorten the processing time.

[0024] Patent Documents 2 and 3, for example, are documents that describe techniques for parallelizing state updates of dynamical systems. The technique described in Patent Document 2 selects multiple variables in a process equivalent to the above process 2, and performs processes 3 and 4 independently for each variable. This technique selects one of the variables accepted in process 4, and changes the state of only that variable. Although processes 3 and 4 are executed multiple times in one iterative process, they are independent and therefore can be parallelized.

[0025] The technology described in Patent Document 3 treats variables as real values ​​instead of discrete values, and performs state updates based on Hamiltonian mechanics. Unlike the Markov chain Monte Carlo method, Hamiltonian mechanics is a deterministic dynamical system that follows differential equations. In principle, this technology can update each variable in parallel.

[0026] In addition, a solution to a combinatorial optimization problem can be obtained by using a dynamical system based on the Markov chain Monte Carlo method. In addition to the basic operations of the Markov chain Monte Carlo method shown in steps 1 to 5, a process of gradually lowering the temperature T from a high temperature can be performed, making it possible to obtain a solution to equation (2) with a high probability when updating the dynamical system for a long period of time.

[0027] A technique for improving the efficiency of searching for solutions to combinatorial optimization problems is described in, for example, Patent Document 4.

[0028] In the technology described in Patent Document 4, in addition to the Hamiltonian, a constraint is given to the set of variables. The constraint given to the set of variables is a condition that is set separately from the condition shown in formula (2). One specific example of a constraint is a constraint that "among the variables belonging to the set, only one variable is 1, and the other variables are -1." This constraint is sometimes called a One-Hot constraint. In combinatorial optimization problems, a setting is often used in which several such constraints are given and a solution to formula (2) is found. The One-Hot constraint is one example of a constraint.

[0029] In the technology described in Patent Document 4, when a variable (spin) is selected in a process corresponding to the above process 2, if updating the value of the variable changes the constraint defined in the set to which the variable belongs from a satisfied state to an unsatisfied state, in addition to the variable, a variable that is a candidate for updating is further selected in order to maintain the state in which the constraint is satisfied. That is, multiple variables are selected simultaneously. As a result, it is possible to omit a solution search that does not satisfy the constraint condition. For example, assume that there is a set of variables for which the above One-Hot constraint is defined, and that the variables belonging to the set satisfy the One-Hot constraint. If only one variable is selected from the variables and the state transition is accepted, the One-Hot constraint will always be unsatisfied. To avoid this, one variable is selected, and another variable is selected to keep the One-Hot constraint satisfied. Then, a judgment is made as to whether the state transition of the two variables is acceptable, and if the state transition is accepted, the states of the two variables are updated simultaneously. As a result, the One-Hot constraint is satisfied even after the state transition. That is, in process 3, the amount of change in the evaluation function when all of the selected multiple variables are updated is calculated, and in process 4, it is determined whether or not to accept the updates of all of the multiple variables. Therefore, the technology described in Patent Document 4 can transition the state while satisfying the given constraints as much as possible. And it can avoid unnecessary solution searches that do not satisfy the constraints as much as possible, and realize efficient solution searches. [Prior art documents] [Patent documents]

[0030] [Patent Document 1] International Publication No. 2019 / 078354 [Patent Document 2] JP 2018-41351 A [Patent Document 3] JP 2019-145010 A [Patent Document 4] International Publication No. 2021 / 059338 [Non-patent literature]

[0031] [Non-Patent Document 1] A. Lucas, "Ising formulations of many NP problems", Frontiers in Physics, 2014, Volume 2, Article 5 Summary of the Invention [Problem to be solved by the invention]

[0032] The inventor of the present invention has found a technology for parallelizing the state update of a dynamical system. Hereinafter, this technology will be referred to as "Technology T" for convenience. In Technology T, a distributed parallel environment is assumed, and processes are distributed. In other words, devices that execute processes are distributed, and each device executes each process. Then, in each distributed process, process 2 and process 3 are executed independently. That is, variables are selected independently, and the change amount Δ of the evaluation function is also calculated independently. In the subsequent process 4, the acceptance judgment of the state transition is performed in the order of the process numbers. At this time, in the subsequent process, Δ is calculated without considering the state transition accepted in the previous process. Therefore, the value of the change amount Δ obtained by formula (4) will be different from the original value. However, the subsequent process communicates the number (index) of the variable whose state has been updated in the preceding process, correcting the value of the change amount Δ, and then performs the acceptance judgment of the state transition. The calculation of equation (4) in process 3 is a bottleneck process, but the technology T discovered by the inventors makes it possible to parallelize this calculation while configuring a dynamical system without losing the sequentiality of state updates.

[0033] Each process is assigned a variable whose value can be updated, and each process can update the value of the assigned variable. However, each process cannot update the value of a variable that is not assigned to it. Also, the element J of the matrix used in the evaluation function (Equation (1)) ij are distributed across the memory of each process. Each process calculates the element J for the set of assigned variables with index i.ij (1≦j≦N) is held. This set of indexes is denoted as I. In addition, processes (devices that execute each process) are connected in a circular manner so that they can communicate with each other, and each process can communicate with the two processes before and after it.

[0034] In the above technology T, the processing proceeds as follows in each process.

[0035] Process 1': A common initial state is given to each process.

[0036] Process 2': Randomly select p variables from set I. p is the number of variables that one process processes, and is a parameter for efficient process execution.

[0037] Process 3'-1. The state at this point is defined as s. Then, the state in which one of the p selected variables is multiplied by -1 is defined as s'. Since the number of selected variables is p, there are also p s'. Then, for each s', the calculation of equation (4) is performed to calculate the amount of change Δ from the value of the evaluation function in state s to the value of the evaluation function in state s'. Therefore, the calculation of Δ is also performed p times.

[0038] Process 3'-2. Receive a set (let's call it K) of indexes of updated variables from the previous process. Let s'' be the state obtained by multiplying the variables indicated by the indexes included in set K from state s by -1. Then, calculate the difference Δ' between the value of the evaluation function in state s'' and the value of the evaluation function in state s'. The calculation of Δ' can be simplified by using Δ that has already been calculated. In other words, the calculation can be simplified by calculating Δ' using the following equation (6).

[0039]

number

[0040] Δ' is the original value of Δ that reflects the results of updating the values ​​of the variables executed so far. In other words, in equation (6), Δ is corrected to its original value.

[0041] Step 4: Calculate the acceptance probability of the state transition by using Δ' instead of Δ in equation (5) which expresses the acceptance probability.

[0042] Process 4'-1: Determine whether the state transition is acceptable. If it is determined that the state transition is acceptable, update the value of the variable with index i. In this case, add index i to set K.

[0043] Process 4'-2: Processes 3'-2 to 4'-1 are repeated for p variables.

[0044] Process 4'-3: Send set K to the next process on the communication path.

[0045] Process 5': Repeat processes 2' to 4'-3.

[0046] The technology described in Patent Document 4 cannot be simply applied to such technology T. In Technology T, variables are assigned to each process. At this time, it is considered that variables belonging to a set of variables with defined constraints are often distributed and assigned to multiple processes. This makes it impossible to realize the technology described in Patent Document 4. That is, since there is no guarantee that variables belonging to a set of variables with defined constraints are assigned collectively to one process, it often happens that one process cannot simultaneously select multiple variables belonging to a set of variables with defined constraints, as in the technology described in Patent Document 4. As a result, the solution search proceeds without satisfying many of the given constraints.

[0047] Therefore, an object of the present invention is to provide a variable assignment device, a variable assignment method, and a variable assignment program that can appropriately assign variables of a combinatorial optimization problem to a plurality of devices performing processing in parallel. [Means for solving the problem]

[0048] A variable assignment device according to the present invention is a variable assignment device which assigns variables of a combinatorial optimization problem to a plurality of parallel processing means which are assigned with processes for determining values ​​of the assigned variables in parallel, the variable assignment device respectively assigning variables of the combinatorial optimization problem to a plurality of parallel processing means in a predetermined order, the variable assignment device comprising: variable assignment means which assigns, for each pair of variables for which constraints are defined, the variables belonging to the pair to one of the parallel processing means, and after all variables have been assigned to one of the parallel processing means, converts the indices of the variables assigned to the parallel processing means in accordance with the order of the parallel processing means; and matrix conversion means which converts a matrix used in an evaluation function of the combinatorial optimization problem in accordance with the converted indexes of the variables.

[0049] A variable assignment method according to the present invention is a variable assignment method in which variables of a combinatorial optimization problem are assigned to a plurality of parallel processing means which perform processing in parallel to find values ​​of the assigned variables, the method assigning variables of the combinatorial optimization problem to each of a plurality of parallel processing means in a predetermined order, the method being characterized in that, for each pair of variables for which constraints are defined, a computer assigns the variables belonging to the pair to one of the parallel processing means, and after all variables have been assigned to one of the parallel processing means, converts the indices of the variables assigned to the parallel processing means in accordance with the order of the parallel processing means, and converts a matrix used in an evaluation function of the combinatorial optimization problem in accordance with the converted indexes of the variables.

[0050] A variable assignment program according to the present invention is a variable assignment program for causing a computer to execute a process of assigning variables of a combinatorial optimization problem to each of a plurality of parallel processing means having a set order, the plurality of parallel processing means being assigned variables and performing in parallel a process of determining values ​​of the assigned variables, the program causing the computer to execute a variable assignment process of assigning variables belonging to each set of variables for which constraints are defined to one of the parallel processing means, and after allocating all variables to one of the parallel processing means, converting indices of the variables assigned to the parallel processing means in accordance with the order of the parallel processing means, and a matrix conversion process of converting a matrix used in an evaluation function of the combinatorial optimization problem in accordance with the converted indices of the variables. The present invention may also be a computer-readable recording medium having the above variable assignment program recorded thereon. Effect of the Invention

[0051] According to the present invention, variables for a combinatorial optimization problem can be appropriately assigned to a plurality of devices performing processing in parallel. [Brief description of the drawings]

[0052] [Figure 1] FIG. 2 is a schematic diagram showing a variable assignment device according to the present invention and a number of elements executing distributed processes in technology T. [Diagram 2] 1 is a block diagram showing an example of the configuration of a variable assignment device; [Diagram 3] 4 is a flowchart showing an example of a processing progress of an embodiment of the present invention. [Figure 4] 4 is a flowchart showing an example of a processing progress of an embodiment of the present invention. [Diagram 5] FIG. 2 is a schematic diagram showing a set of variables sorted in order of priority. [Figure 6] FIG. 13 is a schematic diagram showing variables assigned to a dynamics system update unit; [Figure 7] FIG. 13 is a schematic diagram showing a converted index. [Figure 8]FIG. 2 is a block diagram showing an example of the configuration of one dynamical system updating unit. [Figure 9] 13 is a flowchart showing an example of the process progress of a dynamics system update unit. [Figure 10] 13 is a flowchart showing an example of the process progress of a dynamics system update unit. [Figure 11] FIG. 2 is a schematic block diagram showing an example of the configuration of a computer related to a variable assignment device according to an embodiment of the present invention. [Figure 12] 1 is a block diagram showing an example of an overview of a variable assignment device according to the present invention; DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0053] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0054] FIG. 1 is a schematic diagram showing a variable assignment device according to the present invention and a number of elements that execute distributed processes in the above-mentioned technology T.

[0055] 1 are elements that execute a distributed process in technology T. However, the dynamical system updating units 201-203 execute the distributed process in technology T by applying the technique described in Patent Document 4. The variable assignment device 10 of the embodiment of the present invention assigns variables of a combinatorial optimization problem to each of the dynamical system updating units 201-203, so that the dynamical system updating units 201-203 can execute the distributed process in technology T by applying the technique described in Patent Document 4.

[0056] When there is no need to distinguish between the dynamical system updating units 201 to 203, the reference number 20 without a subscript will be used.

[0057] Each dynamical system update unit 20 may be referred to as a parallel processing means.

[0058] Although three dynamic system updating units 20 are shown in FIG. 1, the number of dynamic system updating units 20 may be any number as long as it is more than one, and is not limited to three.

[0059] The dynamic system updating units 201 to 203 are connected in a circular manner so as to be able to communicate with each other. The multiple dynamic system updating units 20 are ordered. In this embodiment, the subscript of the reference numeral 20 indicates the order of the dynamic system updating units 20. Each dynamic system updating unit 20 is capable of communicating with the dynamic system updating unit 20 immediately preceding it and the dynamic system updating unit 20 immediately following it.

[0060] Furthermore, each dynamical system updating unit 20 has a set upper limit on the number of variables that can be assigned.

[0061] In this embodiment, an example is taken of a case where each dynamic system updating unit 20 is realized by an information processing device (computer), but each dynamic system updating unit 20 may be realized by, for example, a processor or the like.

[0062] In addition, in this embodiment, a case will be described in which the variable assignment device 10 is realized by an information processing device (computer) separate from each dynamical system updating unit 20.

[0063] The variable assignment device 10 is provided in advance with an evaluation function for the combinatorial optimization problem, each variable of the combinatorial optimization problem, a set of variables for which a constraint is defined, and the contents of the constraint. The number of sets of variables for which a constraint is defined is not limited to one. Each set of variables and the constraints that each set must satisfy are defined, for example, by a user who uses the variable assignment device 10 of this embodiment.

[0064] The variable assignment device 10 is communicatively connected to each of the dynamical system updating units 201 to 203. The variable assignment device 10 assigns variables to each of the dynamical system updating units 201 to 203, and transmits information on the assigned variables to the dynamical system updating unit 20 to which the variables are assigned.

[0065] Here, the types of constraints that are set on the set of variables for a combinatorial optimization problem will be described.

[0066] (1) One type of constraint for a set of variables is a constraint (One-Hot constraint) that sets only one of the multiple variables in the set to "1" and sets all other variables in the set to "-1." Hereinafter, this constraint will be referred to as the "first constraint."

[0067] (2) Another type of constraint on a set of variables is the constraint that in a set of three variables x, y, and z, x=y*z. In this case, x is called an auxiliary spin (auxiliary variable). Hereinafter, this constraint will be referred to as the "second constraint." The number of variables in the set for which the second constraint is defined is three.

[0068] (3) A constraint other than the first and second constraints is referred to as a third constraint. Examples of the third constraint include a constraint that at least one variable among multiple variables belonging to a set is set to "1" and a constraint that at least one variable among multiple variables belonging to a set is set to "-1."

[0069] A variable may belong to multiple sets with constraints.

[0070] In this embodiment, the priority of the constraints is determined according to the type of the constraint. In this embodiment, the "first constraint" has the highest priority, the "second constraint" has the second highest priority, and the "third constraint" has the third highest priority.

[0071] A priority is set according to the type of constraint as described above. In addition, if the types of constraints are the same, the higher the number of variables that belong to the set for which the constraint is set, the higher the priority. For example, among multiple sets for which a "first constraint" is set, the higher the number of variables that belong to the set, the higher the priority. The same applies to the third constraint. In addition, if the number of variables that belong to multiple sets for which the same type of constraint is set is the same, the priorities of those sets may be set arbitrarily.

[0072] Next, a description will be given of the variable assignment device 10. Fig. 2 is a block diagram showing an example of the configuration of the variable assignment device 10. The variable assignment device 10 includes a variable assignment unit 11, a matrix conversion unit 12, a communication unit 13, and an index inverse conversion unit 14.

[0073] The variable assignment unit 11 rearranges the sets of variables for which constraints are defined according to the priority of the constraints. The variable assignment unit 11 selects sets of variables for which constraints are defined in descending order of priority, and for each selected set, assigns the variables belonging to the set to one of the dynamical system update units 20. At this time, the variable assignment unit 11 assigns the variables belonging to the selected set to the dynamical system update unit 20 having the largest difference between the upper limit of the number of variables that can be assigned to the dynamical system update unit 20 and the number of variables already assigned to the dynamical system update unit 20. However, there are cases where the variable assignment unit 11 allows variables belonging to one set to be assigned to multiple dynamical system update units 20.

[0074] After all the variables are assigned to any one of the dynamic system updating units 20, the variable assignment unit 11 converts the indexes of the variables assigned to the dynamic system updating units 20 according to the order of the dynamic system updating units 20. The variable assignment unit 11 converts the indexes so that the variable indexes are in ascending order from 1 according to the order of the dynamic system updating units 20. For example, assume that three variables with indexes "10", "16", and "20" are assigned to the dynamic system updating unit 201. In this case, the variable assignment unit 11 converts the indexes "10", "16", and "20" to indexes "1", "2", and "3", respectively. Then, the variable assignment unit 11 converts the indexes of the variables assigned to the second dynamic system updating unit 202 to "4", "5", and so on. The variable assignment unit 11 holds the correspondence between the index before conversion and the index after conversion of each variable.

[0075] The matrix transformation unit 12 transforms a matrix used in an evaluation function for a combinatorial optimization problem according to the transformed index of the variable.

[0076] This transformation will be explained in more detail. The matrix used in the evaluation function for combinatorial optimization problems is J, which corresponds to two indexes i and j. ij Here, if the number of variables is N, then 1≦i≦N, 1≦j≦N. The elements of the transformed matrix are expressed by adding the index as a subscript to J'. The transformed indices of indexes i and j are respectively σ(i) and σ(j). In this case, the transformed J' σ(i)σ(j) But before conversion, ij For example, assume that indexes "10" and "16" are converted to indexes "1" and "2". In this case, the converted J' 1,2 But before conversion, 10,16 In this manner, the matrix transformation unit 12 transforms each element of the matrix.

[0077] The communication unit 13 transmits and receives information to and from each dynamical system update unit 20 .

[0078] The communication unit 13 transmits to each dynamical system updating unit 20 the converted index of the assigned variable, the converted matrix elements corresponding to the assigned variable, and the constraints defined for the set of assigned variables. The transmitted matrix elements will be explained in more detail. For example, if one of the converted indexes of the variables assigned to the dynamical system updating unit 20 is "5", the communication unit 13 transmits J' 5,1 ~J' 5,N The communication unit 13 transmits the transformed matrix elements to the dynamical system updating unit 20 according to the other transformed indexes. That is, if the transformed index of the variable assigned to the dynamical system updating unit 20 is σ(i), the communication unit 13 transmits J′ σ(i),1 ~J' σ(i),N is transmitted to the dynamical system update unit 20. Here, there are multiple types of values ​​for σ(i).

[0079] Furthermore, the communication unit 13 receives, from each dynamic system updating unit 20, the value of each variable (a value of 1 or −1) determined in each dynamic system updating unit 20, together with the index of each variable.

[0080] The index inverse conversion unit 14 converts the index of each variable received from each dynamic system update unit 20 together with the value of each variable back to the index before conversion. That is, the index inverse conversion unit 14 converts the index of each variable back to the index before conversion.

[0081] The communication unit 13 is realized, for example, by a CPU (Central Processing Unit) of a computer that operates according to a variable assignment program, and a communication interface of the computer. For example, the CPU may read the variable assignment program from a program recording medium such as a program storage device of the computer, and operate as the communication unit 13 using the communication interface according to the variable assignment program. This communication interface is an interface used for communication with each dynamical system update unit 20. In addition, the variable assignment unit 11, the matrix conversion unit 12, and the index inverse conversion unit 14 are realized, for example, by a CPU of a computer that operates according to the variable assignment program. For example, the CPU may read the variable assignment program from a program recording medium as described above, and operate as the variable assignment unit 11, the matrix conversion unit 12, and the index inverse conversion unit 14 according to the variable assignment program.

[0082] Next, the process will be described. Figures 3 and 4 are flow charts showing an example of the process of the embodiment of the present invention. Detailed description of the items that have already been described will be omitted as appropriate.

[0083] First, the variable allocation unit 11 rearranges the sets of variables for which constraints are defined in order of priority (step S1). FIG. 5 is a schematic diagram showing sets of variables rearranged in order of priority. As shown in FIG. 5, the priority of a set for which a first constraint is defined is high, and among sets for which a first constraint is defined, the priority is higher as the number of variables belonging to the set increases. Next, the priority of a set for which a second constraint is defined is high. Since the number of variables belonging to sets for which a second constraint is defined is the same, the priority of a plurality of sets for which a second constraint is defined is arbitrarily determined. Next, the priority of a set for which a third constraint is defined is high, and among sets for which a third constraint is defined, the priority is higher as the number of variables belonging to the set increases.

[0084] Next, the variable assignment unit 11 selects one unselected set (a set of variables for which a constraint is defined) in order of decreasing priority (step S2).

[0085] Then, the variable assignment unit 11 determines (step S3) whether or not some of the variables belonging to the selected set have already been assigned to any of the dynamical system update units 20. If all of the variables belonging to the selected set have not been assigned to any of the dynamical system update units 20 (No in step S3), the process proceeds to step S7.

[0086] In step S7, the variable assignment unit 11 assigns the variables belonging to the selected set to one dynamical system updating unit 20. At this time, the variable assignment unit 11 assigns the variables belonging to the selected set to the dynamical system updating unit 20 having the largest difference between the upper limit of the number of variables that can be assigned to the dynamical system updating unit 20 and the number of variables already assigned to the dynamical system updating unit 20.

[0087] If some of the variables belonging to the selected set have already been assigned to any of the dynamical system updating units 20 (Yes in step S3), the process proceeds to step S4.

[0088] In step S4, the variable assignment unit 11 determines whether or not at least two variables among the variables belonging to the selected set have already been assigned to different dynamical system updating units 20. If at least two variables have been assigned to different dynamical system updating units 20 (Yes in step S4), the processing from step S2 onwards is repeated. In this case, among the variables belonging to the selected set, variables that have not yet been assigned to a dynamical system updating unit 20 are assigned to the dynamical system updating unit 20 in step S9, which will be described later. This means that the variable assignment unit 11 allows the variables belonging to the selected set to be assigned to different dynamical system updating units 20.

[0089] If the above-mentioned some variables are assigned to only one dynamical system updating unit 20 (No in step S4), the process proceeds to step S5. In step S5, the variable assignment unit 11 determines whether or not the difference between the upper limit of the number of variables that can be assigned in the dynamical system updating unit 20 to which the above-mentioned some variables are assigned and the number of variables already assigned to that dynamical system updating unit 20 is equal to or greater than the number of variables other than the above-mentioned some variables (referred to as the remaining variables) among the variables belonging to the selected set.

[0090] If the difference is smaller than the number of remaining variables (No in step S5), the processes from step S2 onwards are repeated. In this case, among the variables belonging to the selected set, those not yet assigned to the dynamical system updating unit 20 are assigned to the dynamical system updating unit 20 in step S9 described below. This means that the variable assignment unit 11 allows the variables belonging to the selected set to be assigned to different dynamical system updating units 20.

[0091] If the difference is equal to or greater than the number of remaining variables (Yes in step S5), the variable assignment unit 11 assigns the remaining variables to the dynamical system update unit 20 to which the above-mentioned part of the variables has been assigned (step S6).

[0092] After step S6 and after step S7, the variable assignment unit 11 judges whether all pairs (pairs of variables for which constraints are defined) have been selected (step S8). If there are any pairs that have not been selected (No in step S8), the processes from step S2 onwards are repeated.

[0093] If all pairs have been selected (Yes in step S8), the process proceeds to step S9. The process from step S9 onwards is shown in FIG.

[0094] In step S9, the variable assignment unit 11 assigns variables that have not yet been assigned to the dynamical system updating unit 20 to the dynamical system updating unit 20. In step S9, the variable assignment unit 11 preferentially assigns variables to the dynamical system updating unit 20 having the largest difference between the upper limit of the number of variables that can be assigned to the dynamical system updating unit 20 and the number of variables already assigned to the dynamical system updating unit 20.

[0095] It is assumed that the upper limit of the number of variables that can be assigned to each of the dynamic system update units 201 to 203 is 10. It is also assumed that the nine pairs shown in FIG. 5 are given. In this case, in the process up to the above step S9, the variables belonging to each pair are assigned to the dynamic system update units 201 to 203, for example, as shown in FIG. 6. It is to be noted that in this example, the variables belonging to the pair with priority level "5" shown in FIG. 5 are assigned to different dynamic system update units 201 and 202 (see FIG. 5 and FIG. 6). It is also assumed that the variables indicated by indexes 18 and 19 belonging to the pair with priority level "8" shown in FIG. 5 are assigned to the dynamic system update unit 203 in step 9.

[0096] Next, the variable allocation unit 11 converts the indexes so that the indexes of the variables are in ascending order starting from 1 according to the order of the dynamical system update unit 20 (step S10). As shown in FIG. 6, it is assumed that the variables indicated by the indexes 3, 5, 10, 16, 20, 1, 7, 8, and 9 are assigned to the dynamical system update unit 201. It is also assumed that the variables indicated by the indexes 2, 4, 6, 12, and 14 are assigned to the dynamical system update unit 202. It is also assumed that the variables indicated by the indexes 11, 13, 15, 17, 18, and 19 are assigned to the dynamical system update unit 203. In this case, the variable allocation unit 11 converts the indexes 3, 5, 10, 16, 20, 1, 7, 8, and 9 to the indexes 1, 2, 3, 4, 5, 6, 7, 8, and 9. It is also assumed that the variable allocation unit 11 converts the indexes 2, 4, 6, 12, and 14 to the indexes 10, 11, 12, 13, and 14. Furthermore, the variable allocation unit 11 converts the indexes 11, 13, 15, 17, 18, and 19 into indexes 15, 16, 17, 18, 19, and 20. The state after the indexes are changed in this way is shown in Fig. 7. The variable allocation unit 11 holds the correspondence between the index before conversion and the index after conversion for each variable.

[0097] Next, the matrix conversion unit 12 converts the matrix used in the evaluation function (step S11). As described above, the converted indices of indexes i and j are σ(i) and σ(j), respectively. The matrix conversion unit 12 converts the converted J′ σ(i)σ(j) But before conversion, ij The matrix transformation unit 12 may transform each element of the matrix so that the matrix becomes equal to

[0098] Next, the communication unit 13 transmits to each dynamical system update unit 20 the converted index of the assigned variable, the converted matrix elements corresponding to the assigned variable, and the constraints defined for the set of assigned variables (step S12).

[0099] When the communication unit 13 assigns variables to each dynamic system updating unit 20, the communication unit 13 excludes from the transmission target constraints defined in pairs to which the variables belong that are assigned to different dynamic system updating units 20. For example, in the examples shown in Fig. 5 to Fig. 7, the constraints defined in the pairs with priority "5" and the pairs with priority "8" may be excluded from the transmission target.

[0100] Furthermore, when assigning variables to each dynamic system updating unit 20, the communication unit 13 may transmit, to each dynamic system updating unit 20, constraints defined for pairs whose variables are assigned to different dynamic system updating units 20. For example, in the example shown in Figs. 5 to 7, converted indexes 4, 19, and 20 belong to a pair with a priority level of "8", and the variable indicated by converted index 4 is assigned to the dynamic system updating unit 201, while the variables indicated by converted indexes 19 and 20 are assigned to the dynamic system updating unit 203. In this case, the communication unit 13 may transmit, to the dynamic system updating unit 201, information indicating that a third constraint is defined for a pair consisting of variables indicated by index 4, and may similarly transmit, to the dynamic system updating unit 203, information indicating that a third constraint is defined for a pair consisting of variables indicated by indexes 18 and 19. However, this technical matter does not apply to a pair to which a second constraint is defined.

[0101] Each of the dynamical system updating units 201 to 203 that received the information transmitted in step S12 executes in parallel the process of finding the values ​​of the assigned variables, and transmits the found variable values ​​and their indexes to the variable assignment device 10. The operation of this dynamical system updating unit 20 will be described later.

[0102] The variable assignment device 10 receives the index and value of the variable transmitted from each dynamical system update unit 20 (step S13).

[0103] Next, the index inverse conversion unit 14 inversely converts the index received together with the variable value from each dynamic system update unit 20 into the index before conversion (step S14).

[0104] As a result, the values ​​of the variables indicated by the pre-conversion indexes are obtained, which correspond to the solutions of the combinatorial optimization problem.

[0105] Here, a modified example of the above process will be described. In the above process, when at least two variables among the variables belonging to the selected set have already been assigned to different dynamic system update units 20 (Yes in step S4), the process immediately proceeds to step S2. In this case (Yes in step S4), the variable assignment unit 11 may assign, to one of the different dynamic system update units 20, a variable among the variables belonging to the selected set that has not yet been assigned to that dynamic system update unit 20.

[0106] For example, assume that the variable assignment unit 11 selects a pair with a priority level of "5" as shown in Fig. 5. At this time, the variables indicated by indexes 1 and 3 belonging to this pair are assigned to the dynamical system update unit 201, and the variable indicated by index 2 belonging to this pair is assigned to the dynamical system update unit 202. In this case, the variable assignment unit 11 may assign, to either the dynamical system update unit 201 or 202, a variable that belongs to the selected pair and has not yet been assigned to that dynamical system update unit 20. At this time, the variable assignment unit 11 may select, as the dynamical system update unit 20 to which the variables are assigned, the dynamical system update unit 20 having the largest difference between the upper limit value that can be assigned and the number of variables already assigned.

[0107] However, when the variable assignment unit 11 assigns variables to the selected dynamical system update unit 20, the following two conditions must be satisfied: The first condition is that the number of variables to be assigned is equal to or greater than the difference between the upper limit of variables that can be assigned and the number of variables that have already been assigned.

[0108] The second condition is that, when the variables are assigned, the number of overlapping variables assigned to the multiple dynamical system updating units 20 is equal to or less than the overlapping allowable number. For example, in the above example, it is assumed that the variable assignment unit 11 assigns variables with indexes 1 and 3 to the dynamical system updating unit 202. In this case, the variable indicated by index 1 and the variable indicated by index 3 are assigned to the dynamical system updating units 201 and 202, respectively. In this case, the overlapping number of variables assigned to the multiple dynamical system updating units 20 is 2. The overlapping allowable number is a value obtained by subtracting the number of variables in the combinatorial optimization problem from the sum of the upper limit values ​​of the assignable variables determined for each dynamical system updating unit 20. In this example, the sum is 30 and the number of variables is 20, so the overlapping allowable number is 30-20=10. If the number of overlapping variables assigned to multiple dynamical system update units 20 becomes greater than the allowable overlap number, variables will be generated that cannot be assigned to any of the dynamical system update units 20, and therefore the second condition must also be satisfied.

[0109] If both the first condition and the second condition are satisfied, the variable assignment unit 11 may assign the variables to the selected dynamical system update unit 20. After that, the operations from step S2 onwards are repeated. This concludes the description of the modified processing procedure.

[0110] Next, the dynamic system updating unit 20 will be described. Fig. 8 is a block diagram showing an example of the configuration of one dynamic system updating unit 20. Each dynamic system updating unit 20 has a common configuration. The dynamic system updating unit 20 includes a communication unit 21, a next state generating unit 22, an energy calculation unit 23, a transition determination unit 24, and a temperature control unit 25.

[0111] The communication unit 21 receives the information transmitted by the variable assignment device 10 in step S12 (see FIG. 4). That is, the communication unit 21 receives the indexes of the variables assigned to the dynamical system update unit 20, the elements of the transformed matrix corresponding to the assigned variables, and the constraints defined for the set of the assigned variables.

[0112] Furthermore, the communication unit 21 transmits and receives a set K of indices of updated variables between the previous dynamic system updating unit 20 and the next dynamic system updating unit 20.

[0113] The next state generating unit 22 generates p next states. In principle, the next state is a state in which the value of one variable is updated. However, if updating the value of that one variable causes a constraint defined for a set to which that variable belongs to change from a satisfied state to an unsatisfied state, the next state becomes a state in which, in addition to that one variable, the values ​​of other variables belonging to that set are also changed so that the constraint is maintained in a satisfied state. However, it is assumed that there is only one set in which the other variables are also updated so that the constraint is maintained in a satisfied state.

[0114] However, the variables whose values ​​can be updated are only those assigned to the dynamical system updating unit 20 by the variable assignment device 10.

[0115] The current state of each variable is denoted by s. Each next state is denoted by s'. The energy calculation unit 23 calculates the change amount Δ in the value of the evaluation function for each next state by the calculation of equation (4). That is, the energy calculation unit 23 calculates p changes Δ.

[0116] Furthermore, when the communication unit 21 receives the set K from the previous dynamical system update unit 20, the energy calculation unit 23 corrects the change amount Δ for each next state. The corrected change amount is set as Δ'. The energy calculation unit 23 calculates the corrected change amount Δ' by the calculation of the following formula (7).

[0117]

number

[0118] s'' is a state obtained by multiplying the variable indicated by the index included in the set K by -1 from state s.

[0119] The transition determination unit 24 determines, for each next state, whether to accept or reject a state transition to the next state. When the transition determination unit 24 determines to accept a state transition, it updates the state to the next state of interest. When the transition determination unit 24 determines to reject a state transition, it does not update to the next state of interest. When determining whether to accept or reject a state transition, the transition determination unit 24 uses the acceptance probability shown in equation (8).

[0120]

number

[0121] The temperature control unit 25 controls the temperature T. For example, the temperature control unit 25 reduces the temperature T by a predetermined value every time the state is updated a predetermined number of times.

[0122] The communication unit 21 is realized, for example, by a CPU of a computer that operates according to a program and a communication interface of the computer. For example, the CPU may read a program from a program recording medium such as a program storage device of the computer, and operate as the communication unit 21 using the communication interface according to the program. This communication interface is an interface used for communication with the variable assignment device 10 and the adjacent dynamical system update unit 20. In addition, the next state generation unit 22, the energy calculation unit 23, the transition determination unit 24, and the temperature control unit 25 are realized, for example, by a CPU of a computer that operates according to a program. For example, the CPU may read a program from a program recording medium as described above, and operate as the next state generation unit 22, the energy calculation unit 23, the transition determination unit 24, and the temperature control unit 25 according to the program.

[0123] Next, the process flow of the dynamic system update unit 20 will be described. Figures 9 and 10 are flowcharts showing an example of the process flow of the dynamic system update unit 20. Detailed description of matters that have already been explained will be omitted as appropriate. Note that the operation of the temperature control unit 25 is not shown in Figures 9 and 10. It is assumed that the temperature control unit 25 performs an operation of lowering the temperature T by a predetermined value every time it updates the state a predetermined number of times.

[0124] It is assumed that each dynamical system updating unit 20 has received the information transmitted by the variable assignment device 10 in step S12 (see FIG. 4). It is also assumed that a common initial state is provided to each dynamical system updating unit 20 as the initial state of each variable. There is no particular limitation on the method of providing a common initial state to each dynamical system updating unit 20.

[0125] First, the next state generating unit 22 generates p next states (step S21). However, the variables whose values ​​can be updated are only the variables assigned to the dynamical system updating unit 20 by the variable assignment device 10.

[0126] Next, the energy calculation unit 23 calculates the amount of change Δ in the value of the evaluation function for each next state by calculating equation (4) (step S22).

[0127] Next, the communication unit 21 receives a set K of indices of the updated variables from the previous dynamical system updating unit 20 (step S23).

[0128] Next, the energy calculation unit 23 selects one unselected next state from among the p next states (step S24).

[0129] Then, the energy calculation unit 23 calculates Δ' for the selected next state by the calculation of equation (7) (step S25). In other words, the energy calculation unit 23 corrects Δ to Δ' for the selected next state. Δ' is the original value of Δ that reflects the results of updating the values ​​of the variables that have been executed so far.

[0130] Next, the transition determination unit 24 performs an acceptance determination for the state transition to the next state selected in step S24 (step S26). In step S26, the transition determination unit 24 uses the acceptance probability shown in equation (8).

[0131] If the state transition to the next state is rejected, the operations from step S24 onwards are repeated without updating the state.

[0132] When the state transition to the next state is accepted, the transition determination unit 24 updates the current state to the next state selected in step S24 (step S27).

[0133] The transition determination unit 24 adds the index of the variable whose value has been updated in step S27 to the set K (step S28).

[0134] Next, the energy calculation unit 23 judges whether or not all of the p next states generated in step S21 have been selected (step S29). If there is still a next state that has not been selected (No in step S29), the operations in and after step S24 are repeated.

[0135] If all of the p next states have been selected (Yes in step S29), the communication unit 21 transmits the set K to the next dynamical system update unit 20 (step S30).

[0136] Next, the next state generating unit 22 judges whether or not a termination condition of the repetitive process starting from step S21 is satisfied (step S31). The termination condition may be previously determined as the number of times that the repetitive process starting from step S21 has been performed a predetermined number of times. However, the above example is one example of the termination condition, and the termination condition is not limited to the above example.

[0137] If the end condition is not satisfied (No in step S31), the processes from step S21 onwards are repeated.

[0138] If the termination condition is met, the process proceeds to step S32, which is shown in FIG.

[0139] In step S32, the communication unit 21 transmits the values ​​and indexes of the variables assigned to the dynamical system update unit 20 to the variable assignment device 10.

[0140] In step S32, each dynamical system update unit 20 transmits the variable value and index to the variable assignment device 10, and in step S13 (see FIG. 4), the communication unit 13 of the variable assignment device 10 receives the variable value and index. Then, the variable assignment device 10 performs the operations in and after step S13.

[0141] According to this embodiment, the variable assignment unit 11 of the variable assignment device 10 assigns, for each pair of variables for which a constraint is defined, the variables belonging to the pair to one of the dynamical system update units 20. Therefore, the next state generation unit 22 of the dynamical system update unit 20 can generate, as the next state, a next state in which multiple variables belonging to the pair are updated so as to maintain a state in which the constraint defined for the pair of variables is satisfied. Therefore, in multiple dynamical system update units 20, processing can be executed in which the technology described in Patent Document 4 is applied to technology T.

[0142] In other words, in order to enable the application of the technology described in Patent Document 4 to technology T in multiple dynamical system updating units 20, the variables of the combinatorial optimization problem can be appropriately assigned to multiple dynamical system updating units 20 that perform processing in parallel.

[0143] 11 is a schematic block diagram showing an example of the configuration of a computer related to a variable assignment device 10 of the embodiment of the present invention. The computer 1000 includes a CPU 1001, a main memory device 1002, an auxiliary memory device 1003, an interface 1004, and a communication interface 1005.

[0144] A variable assignment device 10 according to an embodiment of the present invention is realized, for example, by a computer 1000. The operation of the variable assignment device 10 is stored in the form of a variable assignment program in an auxiliary storage device 1003. A CPU 1001 reads out the program, expands the program in a main storage device 1002, and executes the processing of the variable assignment device 10 described in the embodiment of the present invention in accordance with the program.

[0145] The auxiliary storage device 1003 is an example of a non-transient tangible medium. Other examples of non-transient tangible media include a magnetic disk, a magneto-optical disk, a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), a semiconductor memory, and the like, which are connected via the interface 1004. In addition, when the variable assignment program is distributed to the computer 1000 via a communication line, the computer 1000 that receives the program may load the program into the main storage device 1002 and execute the processing of the variable assignment device 10 described in the above embodiment in accordance with the program.

[0146] Furthermore, some or all of the components of the variable assignment device 10 may be realized by general-purpose or dedicated circuits, processors, etc., or a combination of these. These may be configured by a single chip, or may be configured by multiple chips connected via a bus. Some or all of the components may be realized by a combination of the above-mentioned circuits, etc., and a program.

[0147] These points also apply to the dynamical system update unit 20.

[0148] Next, an overview of the present invention will be described. Fig. 12 is a block diagram showing an example of an overview of a variable assignment device of the present invention. The variable assignment device of the present invention is a plurality of parallel processing means to which variables of a combinatorial optimization problem are assigned and which perform in parallel the process of finding values ​​of the assigned variables, and assigns variables of the combinatorial optimization problem to each of a plurality of parallel processing means (e.g., a plurality of dynamical system update units 20) in a set order. The variable assignment device includes variable assignment means 71 and matrix conversion means 72.

[0149] The variable assignment means 71 (e.g., the variable assignment unit 11) assigns, for each set of variables for which constraints are defined, the variables belonging to the set to one of the parallel processing means, and after assigning all the variables to one of the parallel processing means, converts the indexes of the variables assigned to the parallel processing means in accordance with the order of the parallel processing means.

[0150] The matrix conversion means 72 (for example, the matrix conversion unit 12) converts a matrix used in an evaluation function for a combinatorial optimization problem according to the converted index of the variable.

[0151] With such a configuration, variables for a combinatorial optimization problem can be appropriately assigned to multiple devices performing processing in parallel.

[0152] The configuration may include an index inverse conversion means (for example, the index inverse conversion unit 14) that inversely converts the index of each variable to the index before conversion when the value of the assigned variable is returned from each parallel processing means.

[0153] A priority may be predefined for the constraints defined for the set of variables depending on the type of the constraint, and the variable assignment means 71 may be configured to assign variables belonging to the set of variables for which the constraints are defined to one of the parallel processing means in accordance with the priority of the constraint.

[0154] The variable allocation means 71 may be configured to allocate variables belonging to a set of variables for which constraints are defined to a parallel processing means having the largest difference between the upper limit of the number of variables that can be allocated to the parallel processing means and the number of variables already allocated to the parallel processing means.

[0155] The variable assignment means 71 may be configured to allow variables belonging to a set of variables for which constraints are defined to be assigned to multiple parallel processing means when at least two of the variables belonging to the set are already assigned to different parallel processing means.

[0156] When some of the variables belonging to a set of variables for which constraints are defined have already been assigned to only one parallel processing means, the variable assignment means 71 may be configured to assign the variables belonging to the set, other than the part of the variables, to the parallel processing means if the difference between the upper limit of the number of variables that can be assigned to the parallel processing means and the number of variables already assigned to the parallel processing means is equal to or greater than the number of variables belonging to the set, other than the part of the variables.

[0157] The variable allocation means 71 may be configured such that, when some of the variables belonging to a set of variables for which a constraint is defined are already assigned to only one parallel processing means, the variable allocation means 71 allows the variables belonging to the set to be assigned to multiple parallel processing means if the difference between the upper limit of the number of variables that can be assigned, defined for the parallel processing means, and the number of variables already assigned to the parallel processing means is smaller than the number of variables other than the some of the variables belonging to the set.

[0158] Although the present invention has been described above with reference to the embodiment, the present invention is not limited to the above embodiment. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

[0159] The present invention is preferably applied to a variable assignment device that assigns variables for a combinatorial optimization problem to a plurality of devices performing processing in parallel. [Explanation of symbols]

[0160] 10 Variable assignment device 11 Variable assignment 12 Matrix transformation section 13. Communications Department 14 Index Reverse Conversion Unit 20 Dynamical system update part

Claims

1. A variable assignment device which assigns variables of a combinatorial optimization problem to a plurality of parallel processing means which are assigned variables of the combinatorial optimization problem and perform processes for determining values ​​of the assigned variables in parallel, the parallel processing means having a predetermined order, the variable assignment device comprising: a variable assignment means for assigning, for each set of variables for which a constraint is defined, the variables belonging to the set to one of the parallel processing means, and after allocating all the variables to one of the parallel processing means, converting the indexes of the variables assigned to the parallel processing means in accordance with the order of the parallel processing means; a matrix conversion means for converting a matrix used in an evaluation function for the combinatorial optimization problem according to an index after conversion of the variable; A variable assignment device comprising:

2. The parallel processing means includes an index inversion means for inversely converting the index of each variable to the index before conversion when the value of the assigned variable is returned from each parallel processing means. The variable assignment device according to claim 1.

3. Priority is predefined for the constraints defined on the set of variables according to the type of constraint. The variable assignment means According to the priority of the constraint, the variables belonging to the set of variables for which the constraint is defined are assigned to one of the parallel processing means. The variable assignment device according to claim 1 or 2.

4. The variable assignment means A variable belonging to a set of variables for which a constraint is defined is assigned to a parallel processing means having the largest difference between the upper limit of the number of variables that can be assigned to the parallel processing means and the number of variables already assigned to the parallel processing means. The variable assignment device according to any one of claims 1 to 3.

5. The variable assignment means When at least two variables among the variables belonging to a set of variables for which a constraint is defined are already assigned to different parallel processing means, the variables belonging to the set are allowed to be assigned to a plurality of parallel processing means. The variable assignment device according to any one of claims 1 to 4.

6. The variable assignment means When some of the variables belonging to a set of variables for which a constraint is defined have already been assigned to only one parallel processing means, if the difference between the upper limit of the number of variables that can be assigned, defined for the parallel processing means, and the number of variables already assigned to the parallel processing means is equal to or greater than the number of variables other than the part of the variables belonging to the set, the variables other than the part of the variables belonging to the set are assigned to the parallel processing means. The variable assignment device according to any one of claims 1 to 5.

7. The variable assignment means When some of the variables belonging to a set of variables for which a constraint is defined have already been assigned to only one parallel processing means, if the difference between the upper limit of the number of variables that can be assigned, defined for the parallel processing means, and the number of variables already assigned to the parallel processing means is smaller than the number of variables other than the some of the variables belonging to the set, the variables belonging to the set are allowed to be assigned to multiple parallel processing means. The variable assignment device according to any one of claims 1 to 6.

8. A variable assignment method for assigning variables of a combinatorial optimization problem to a plurality of parallel processing means that perform processes in parallel to find values ​​of the assigned variables, the parallel processing means having a predetermined order, the method comprising: The computer for each set of variables for which a constraint is defined, assigning the variables belonging to the set to one of the parallel processing means, and after allocating all the variables to one of the parallel processing means, converting the indexes of the variables assigned to the parallel processing means in accordance with the order of the parallel processing means; A matrix used in an evaluation function for the combinatorial optimization problem is converted according to an index after the conversion of the variable. A variable assignment method comprising:

9. A variable assignment program for causing a computer to execute a process of assigning variables of a combinatorial optimization problem to each of a plurality of parallel processing means having a predetermined order, the parallel processing means being configured to assign variables of a combinatorial optimization problem to the plurality of parallel processing means and performing a process of determining values ​​of the assigned variables in parallel, the program comprising: The computer includes: a variable assignment process for assigning, for each set of variables for which constraints are defined, the variables belonging to the set to one of the parallel processing means, and after allocating all the variables to one of the parallel processing means, converting the indexes of the variables assigned to the parallel processing means in accordance with the order of the parallel processing means; and A matrix transformation process that transforms a matrix used in an evaluation function for the combinatorial optimization problem according to an index after the transformation of variables. A variable assignment program for executing the above.

Citation Information

Patent Citations

  • Information processing device, ising device and method for controlling information processing device

    JP2018041351A

  • Computer, calculation program, recording medium, and calculation method

    JP2019145010A

  • Ising model calculation device

    WO2019078354A1

  • Solution system, solution method, and solution program

    WO2021059338A1