Information processing apparatus and information processing method
The information processing device efficiently solves unconstrained and mixed binary quadratic programming problems by initializing variables and dynamically switching between binary and continuous treatments for optimal solutions.
Patent Information
- Application Number
- JP2023078529
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-11
- Publication Date
- 2024-11-21
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing technologies lack efficient methods for solving unconstrained binary quadratic optimization problems and mixed binary quadratic programming problems, particularly in transitioning to optimal solutions.
An information processing device and method that initializes variables, calculates quantities related to state updates, and dynamically switches between treating binary and continuous variables based on probability distributions to efficiently find optimal solutions using simulated annealing and Markov chain Monte Carlo methods.
Facilitates efficient search for optimal solutions in unconstrained binary quadratic optimization and mixed binary quadratic programming problems by optimizing state transitions.
Smart Images

Figure 2024162709000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to an information processing device and an information processing method. [Background technology]
[0002] Various physical and social phenomena can be expressed using interaction models. An interaction model is defined by multiple nodes that make up the model, nonlinear coefficients (or interaction coefficients) between the nodes, and, if necessary, linear coefficients (or bias coefficients and external magnetic field coefficients) that act on each node. In the fields of physics and social science, various models including the Ising model have been proposed, and all of them can be interpreted as a form of interaction model.
[0003] In relation to this point, Patent Document 1 describes a method for converting the interaction relationships into a complete bipartite graph structure for mixed binary quadratic programming problems, including the ground state search problem of the Ising model, to speed up the ground state search based on simulated annealing.
[0004] Furthermore, Patent Document 2 describes a method for converting the interaction relationship of the Ising model into a complete bipartite graph structure to speed up ground state search based on simulated annealing. [Prior art documents] [Patent documents]
[0005] [Patent Document 1] WO2021 / 220445 [Patent Document 2] WO2019 / 216277 Summary of the Invention [Problem to be solved by the invention]
[0006] Finding node states that minimize or maximize indicators related to interaction models is important in solving social problems. Examples include the problem of detecting cliques in social networks and portfolio optimization problems in the financial sector. In the field of operations research, these are broadly classified as unconstrained binary quadratic optimization problems and mixed binary quadratic programming problems. In addition, these can be reduced to quadratic programming problems by relaxing the binary variables to continuous variables with interval constraints.
[0007] However, neither Patent Document 1 nor Patent Document 2 proposes any method for promoting a state transition to an optimal solution when solving an unconstrained binary quadratic optimization problem or a mixed binary quadratic programming problem by reducing it to a quadratic programming problem, and further, does not propose any method for efficiently searching for an optimal solution.
[0008] The present invention has been made in consideration of the above-mentioned background, and has an object to provide an information processing device and an information processing method for efficiently searching for an optimal solution when reducing an unconstrained binary quadratic optimization problem or a mixed binary quadratic programming problem to a quadratic programming problem and solving it. [Means for solving the problem]
[0009] In view of the above, the present invention provides "an information processing device for solving a mixed binary quadratic programming problem having an objective function that is a quadratic function for binary variables or continuous variables, comprising: a model setting unit that sets the mixed binary quadratic programming problem; a variable value initialization unit that initializes variables of the mixed binary quadratic programming problem; a product-sum operation execution unit that calculates a quantity related to state update based on the variables in the mixed binary quadratic programming problem; and a state update control unit that updates the state of the variables based on the quantity related to state update, wherein the product-sum operation execution unit calculates an index of the magnitude of state transition in a probabilistic state transition as the quantity related to state update based on a probability distribution function defined by the original objective function and a probability distribution function defined by an objective function obtained by relaxing the binary variables to continuous values, and the state update control unit updates the variables while switching state transition methods based on the quantity related to state update to search for an optimal solution."
[0010] The present invention also provides an information processing method for solving a mixed binary quadratic programming problem having an objective function that is a quadratic function for binary variables or continuous variables using a computer, the computer having a model setting unit that sets the mixed binary quadratic programming problem and a calculation unit, the calculation unit initializes variables of the mixed binary quadratic programming problem, calculates quantities related to state update based on the variables in the mixed binary quadratic programming problem, executes a process of updating the states of the variables based on the quantities related to state update, and calculates an index of the magnitude of state transition in a probabilistic state transition as a quantity related to state update based on a probability distribution function defined by the original objective function and a probability distribution function defined by an objective function obtained by relaxing the binary variables to continuous values, and updates the variables while switching state transition methods based on the quantities related to state update to search for an optimal solution. Effect of the Invention
[0011] According to the present invention, it is possible to efficiently search for an optimal solution when reducing an unconstrained binary quadratic optimization problem or a mixed binary quadratic programming problem to a quadratic programming problem and solving it. Problems, configurations, and effects other than those described above will become apparent from the following description of the embodiment of the invention. [Brief description of the drawings]
[0012] [Figure 1] FIG. 1 is a diagram showing an example of a basic hardware configuration of an information processing apparatus according to an embodiment of the present invention. [Diagram 2] 2 is a diagram showing main functions (software configuration) of an information processing device 10. [Diagram 3] 3 is a process flow of the ground state search executed by the processing unit (21-26) in FIG. [Figure 4] Diagram showing the concept of the energy landscape of the interaction model. [Diagram 5] A schematic showing the concept of gradually changing the temperature parameter T in simulated annealing SA. [Figure 6] A complete graph of the interaction relationships between variables in an interaction model. [Figure 7]4 is a detailed process flow of processing step S14 in FIG. 3. [Figure 8] Diagram showing probability distribution functions for binary and continuous variables. [Figure 9] A schematic diagram showing how repeated random sampling moves between local optima in state space. [Figure 10] A graph showing the value of standard deviation SD versus gi / T. [Figure 11] FIG. 13 is a diagram showing the overall processing contents of the ground state search processing and the variable update processing. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0013] Hereinafter, the embodiments will be described in detail with reference to the drawings. In the following description, the same or similar configurations may be given common reference numerals and duplicated descriptions may be omitted. Furthermore, when there are multiple elements having the same or similar functions, different subscripts may be given to the same reference numerals. Furthermore, when there is no need to distinguish between multiple elements, the subscripts may be omitted. EXAMPLES
[0014] An information processing device according to an embodiment of the present invention is realized using a computer. Fig. 1 shows an example of a basic hardware configuration of an information processing device according to an embodiment of the present invention.
[0015] In FIG. 1, an information processing device 10 includes a processor 11, a main memory device 12, an auxiliary memory device 13, an input device 14, an output device 15, a communication device 16, one or more arithmetic devices 20, and a system bus 5 that communicatively connects these devices.
[0016] The information processing device 10 may be realized, for example, by using a virtual information processing resource such as a cloud server provided in part or in whole by a cloud system. The information processing device 10 may also be realized, for example, by a plurality of information processing devices connected to each other so as to be able to communicate with each other and to operate in cooperation with each other.
[0017] The processor 11 is configured using, for example, a CPU (Central Processing Unit) or an MPU (Micro Processing Unit). The main storage device 12 is a device that stores programs and data, and is, for example, a ROM (Read Only Memory) (SRAM (Static Random Access Memory), NVRAM (Non Volatile RAM), Mask ROM (Mask Read Only Memory), PROM (Programmable ROM), etc.), a RAM (Random Access Memory) (DRAM (Dynamic Random Access Memory)), etc. The auxiliary storage device 13 is a hard disk drive, a flash memory, an SSD (Solid State Drive), an optical storage device (CD (Compact Disc), DVD (Digital Versatile Disc), etc.), etc. The programs and data stored in the auxiliary storage device 13 are read into the main storage device 12 as needed.
[0018] The input device 14 is a user interface that accepts information input from a user, and is, for example, a keyboard, a mouse, a card reader, a touch panel, etc. The output device 15 is a user interface that provides information to a user, and is, for example, a display device (LCD (Liquid Crystal Display), a graphic card, etc.) that visualizes various information, an audio output device (speaker), a printer, etc. The communication device 16 is a communication interface that communicates with other devices, and is, for example, a NIC (Network Interface Card), a wireless communication module, a USB (Universal Serial Interface) module, a serial communication module, etc.
[0019] The arithmetic device 20 is a device that executes a ground state search, the details of which will be described later. The arithmetic device 20 may take the form of an expansion card that is attached to the information processing device 10, such as a GPU (Graphics Processing Unit). The arithmetic device 20 is configured by hardware such as a CMOS (Complementary Metal Oxide Semiconductor) circuit, an FPGA (Field Programmable Gate Array), and an ASIC (Application Specific Integrated Circuit). The arithmetic device 20 includes a control device, a storage device, an interface for connecting to a system bus 5, and transmits and receives commands and information to and from the processor 11 via the system bus 5. The arithmetic device 20 may be communicably connected to another arithmetic device 20 via a communication line, for example, and may operate in cooperation with the other arithmetic device 20. The functions realized by the arithmetic device 20 may be realized by, for example, having a processor (CPU, GPU, etc.) execute a program.
[0020] Fig. 2 shows the main functions (software configuration) of the information processing device 10. Here, the processing functions executed by the processor 11 or the arithmetic device 20 in Fig. 1 are described as processing units (21-26). These processing functions are a model setting unit 21, a variable value initialization unit 22, a temperature parameter control unit 23, a product-sum operation execution unit 24, a state transition control unit 25, and a variable value reading unit 26.
[0021] These functions are realized by the processor 11 reading and executing a program stored in the main storage device 12, or by hardware included in the arithmetic device 20. In addition to the above functions, the information processing device 10 may also include other functions such as an operating system, a file system, a device driver, and a DBMS (DataBase Management System).
[0022] In addition, the memory unit DB, which is composed of the main memory device 12 or the auxiliary memory device 13 in the information processing device 10, stores mixed binary quadratic programming problem MBQP format problem data D1 in an MBQP format problem database DB1, and the calculation device control program D2 in a calculation device control program database DB2.
[0023] Here, the mixed binary quadratic programming problem MBQP format problem data D1 is data in which a combinatorial optimization problem is input in a predetermined description format, and holds nonlinear coefficients and linear coefficients. Details of the mixed binary quadratic programming problem will be described later. For example, the MBQP format problem data D1 is set by a user via a user interface (input device, output device, communication device, etc.). The arithmetic device control program D2 is a program that is executed when the product-sum operation execution unit 24 controls the arithmetic device 20, or that is loaded by the product-sum operation execution unit 24 into each arithmetic device 20 and executed by the arithmetic device 20.
[0024] The outline of the processing contents of the processing unit (21-26) is as follows. First, the model setting unit 21 passes model information to the calculation device 20 based on the MBQP format problem data D1. The variable value initialization unit 22 initializes the values stored in the variable memory of the calculation device 20. The temperature parameter control unit 23 controls the temperature parameter T in SA. The product-sum calculation execution unit 24 calculates the amount related to the state update according to SA for the interaction model (the value of g in equation (5) described later), and is a part of the calculation device 20. The state update method control unit 25 is a device that gives an instruction to switch the update method of variables to the calculation device 20 as described in the embodiment. When SA is completed in the calculation device 20, the variable value reading unit 26 reads the value stored in the variable memory and outputs the read value to the output device 15 or the communication device 16, thereby completing the ground state search.
[0025] FIG. 3 shows a flow diagram of the ground state search process executed by the processing units (21-26) of FIG. 2, in which processing step S11 corresponds to the processing of the model setting unit 21, processing step S12 corresponds to the processing of the variable value initialization unit 22, processing step S13 corresponds to the processing of the temperature parameter control unit 23 and the product-sum operation execution unit 24, processing step S14 corresponds to the processing of the state transition control unit 25, and processing step S16 corresponds to the processing of the variable value reading unit 26.
[0026] The ground state search process in FIG. 3 is started, for example, by receiving an instruction from a user via the input device 14 in FIG.
[0027] After the start, in the first processing step S11 of the series of processing in Fig. 3, the model coefficient setting unit 21 sets information on the model coefficients in the arithmetic device 20. The value can also be set or edited by the user via a user interface (realized by, for example, the input device 14, the output device 15, the communication device 16, etc.).
[0028] Setting the model coefficients means setting a mixed binary quadratic programming problem, and here, an interaction model is formed by setting the model coefficients. In the following explanation, the interaction model will first be explained in detail.
[0029] The interaction model is defined by multiple nodes that make up the model, nonlinear coefficients acting between the nodes, and linear coefficients acting on each node. Here, the variable x corresponding to each node i (i = 1 to N, N is the number of all nodes) is i is a binary variable x i ∈{-1, 1} or continuous variable x i ∈[-1, 1].
[0030] The energy function H(x) (generally called Hamiltonian) is defined based on the model coefficients (nonlinear and linear coefficients) of the interaction model input by the user. In particular, when an interaction relationship is defined between two nodes, the energy function becomes a quadratic equation shown in equation (1). Note that it is also possible to think that the first term in equation (1) represents the interaction between the nodes, and the second term represents the bias toward the node.
[0031]
number
[0032] In general, an interaction model is expressed as an undirected graph, and nonlinear terms are specified by the set of variables contained therein. Therefore, the matrix J in equation (1) is a real symmetric matrix (J ij =J ji ) and the binary variable x i ∈{-1, 1} squared is x 2 i = 1, so the binary variable x i Diagonal element J ii can be set to 0.
[0033] The above interaction model is a concept that includes the Ising model. In fact, in the interaction model, all variables are expressed as binary variables x i When we restrict it to ∈{-1, 1}, we get the Ising model, which is used, for example, in statistical mechanics as a lattice model to describe magnetic materials, with +1 / -1 corresponding to the up / down spin directions.
[0034] Searching for the minimum energy state (ground state) of an interaction model is an optimization problem that seeks the variable array x that minimizes the above energy function. This is equivalent to a mixed binary quadratic programming problem (hereafter referred to as MBQP (Mixed Binary Quadratic Programming)) with interval constraints on variables, with the energy function as the objective function to be minimized.
[0035] Figure 4 is a conceptual diagram of the energy landscape of the interaction model. The horizontal axis of the graph is the variable array, and the vertical axis is the total energy of the system. In Figure 4, state A has the smallest energy, and updating one of the variables from state A from -1 to 1 results in state B with more energy, and updating yet another variable from 1 to -1 results in state C with less energy. Changing the variables in this way changes the value of the energy function, and moves around the energy landscape.
[0036] In this embodiment, the ground state of the interaction model is searched for using Markov Chain Monte Carlo methods (hereafter referred to as MCMC). In the stochastic transition in the Markov Chain Monte Carlo methods MCMC, stochastic transition from the current state x to another state is repeated. Gibbs sampling (or heat bath method) is a method for sequentially executing stochastic state transitions. Let p(x) be the probability distribution function that state x follows, and x be the state generated in the i-th step by Gibbs sampling. (i) ={x1 (i) , …, x N (i)} In this case, the i+1th state is generated sequentially according to the conditional probability of equation (2) for n=1, .., N.
[0037]
number
[0038] When the Markov chain Monte Carlo method (MCMC) is run by introducing a parameter T that controls the above-mentioned stochastic state transitions and gradually suppressing the transitions, the state converges asymptotically to the ground state. Simulated annealing (hereafter referred to as SA) is a method that utilizes this to find the optimal solution to a minimization problem.
[0039] Figure 5 shows a schematic diagram of the case where the parameter T is gradually changed in simulated annealing SA. The parameter T is called the temperature parameter, corresponding to the real annealing. The maximum and minimum values of the temperature parameter are T H and T L For example, when T H From T L The geometric cooling method, in which the temperature parameter is lowered by steps of values interpolated by an arithmetic progression between , is commonly used. Of course, various cooling methods can be implemented, such as lowering the temperature parameter by steps of values interpolated by an arithmetic progression.
[0040] Figure 6 is a diagram showing the interaction relationships between the variables of the interaction model as a complete graph. The upper diagram in Figure 6 shows the interaction relationships between the variables of the interaction model. The number of variables N=6. Since there are interaction relationships between all variables, it is a complete graph (fully connected graph). Note that the parts indicated with squares are binary variables of {-1, 1}, and the parts indicated with circles are continuous variables between [-1, 1]. The lower diagram in Figure 6 is an interaction model in which the binary variables of the interaction model represented in the upper diagram in Figure 6 are relaxed to continuous values.
[0041] Here, the binary variable x in equation (1), which is the interaction model, i For the energy function H(x) of the interaction model, we focus on all variables x i Let x' be a continuous variable. i The energy function of the interaction model replaced by ∈[-1, 1] is H(x'), and the ground states of H(x) and H(x') are x * and x' * Then, as described in Patent Document 1, the following formula (3) holds.
[0042]
number
[0043] Here, the sign function sign(x) is a function that returns +1 if x ≥ 0, and -1 if x < 0. The ground state of H(x) is found by searching the ground state of H(x') using equation (3). In other words, in searching the ground state of an interaction model that represents a mixed binary quadratic programming problem (MBQP), there is a degree of freedom in whether to treat binary variables as they are or as continuous variables.
[0044] When applying the Markov chain Monte Carlo method (MCMC) or simulated annealing SA to an interaction model, the variable values are probabilistically updated based on equation (2) or the like. Here, variables that do not have a nonlinear term between them are independent in the energy function and therefore the probability density function, and it is possible to simultaneously apply state transitions based on equation (2). Therefore, by updating independent variables in parallel, it is possible to speed up the processing of the Markov chain Monte Carlo method (MCMC) and simulated annealing SA.
[0045] Such speedup can be achieved by using an interaction model in which the interaction relationship has a bipartite graph structure, i.e., each subgraph of a node does not have an interaction relationship. Patent Document 2 proposes a method of solving the ground state search problem of an Ising model having a complete bipartite graph structure instead of the ground state search problem of a fully connected Ising model, thereby speeding up variable updates by parallel processing to find the ground state. Patent Document 1 proposes a method of solving the ground state search problem of an interaction model having a complete bipartite graph structure in which all binary variables of a mixed binary quadratic programming problem MBQP are relaxed to continuous values, and then speeding up variable updates by parallel processing to find the ground state.
[0046] The relationship between Patent Documents 1 and 2 corresponds to the relationship between whether to treat binary variables as they are in the ground state search of the interaction model expressing the mixed binary quadratic programming problem (MBQP) described above, or to treat them as continuous variables after relaxing the continuous values. Conventionally, either of the two treatments has been used, and since the efficiency of state search changes depending on how the variables are treated, the appropriate method may differ depending on the problem.
[0047] Returning to FIG. 3, in order to efficiently solve the mixed binary quadratic programming problem MBQP set as above, in the next process step S12, the variable value initialization unit 23 initializes the variable values of the calculation device 20, and the temperature parameter control unit 23 sets the initial value of the temperature parameter of the calculation device 20. For example, the initial value of the variables is selected as one of the values {-1, +1} with a probability of 50% for binary variables, and is selected by random sampling from a uniform distribution in the interval [-1, +1] for continuous variables. Also, the initial value of the temperature parameter is selected as the maximum value T of the temperature parameter in simulated annealing SA. H Set it in.
[0048] In and after processing step S13 in FIG. 3, a method for solving the mixed binary quadratic programming problem MBQP is specifically processed. In the processing, taking into consideration the background in Patent Documents 1 and 2 mentioned above, when performing a ground state search of an interaction model that represents the mixed binary quadratic programming problem MBQP, a method is adopted in which the efficiency of the ground state search is improved by dynamically switching between treating binary variables as they are or as continuous variables by referring to the statistics of the probability distribution function that follows the state transition.
[0049] Here, the state transition is performed according to Gibbs sampling for the energy function of equation (1). Note that, as the temperature parameter T, the state x of the interaction model on the right-hand side of equation (7) described later appears with the probability density function p(x) based on the Boltzmann distribution shown in equation (4) according to the Markov chain Monte Carlo method MCMC.
[0050]
number
[0051] In equation (4), Z is the normalization factor of the Boltzmann distribution, which is called the partition function. By applying Gibbs sampling as shown in equation (2) to this Boltzmann distribution, it is possible to probabilistically update the variable x.
[0052] Then, in processing step S13 in the computing device 20, the temperature parameter control unit 23 updates the value of the temperature parameter if necessary, and the product-sum operation execution unit 24 updates the amount required for updating the state. The temperature parameter is updated according to the geometric cooling method described above or the like.
[0053] In this example, it is best to use g in equation (5) as the quantity required for state update. In equation (5), which shows the specific equation for update, the following vector quantity is defined for this purpose, and g in equation (5) is related to the quantity obtained by differentiating the energy function in equation (1) with respect to x.
[0054]
number
[0055] 3, the state update control unit 25 updates the state of the variable based on the value of g in the formula (5) and the temperature parameter in the process step S14. This process is performed by updating the state of the variable x i The process may be performed in order from 1 to N for the subscript i, or in a random order for the subscript i from 1 to N. Details of this process will be described with reference to FIG.
[0056] Fig. 7 shows the process step S14 in Fig. 3 in more detail. In the variable update process in Fig. 7, it is determined whether it is appropriate to process the variable xi in binary or continuous, and a switching process is performed. The reason for this switching process will be explained below.
[0057] Here, we first consider the original interaction model that represents the mixed binary quadratic programming problem (MBQP), and then consider the continuous variable x iConsider the interaction matrix J ii The case is distinguished depending on whether is non-zero or zero.
[0058] For example, the interaction matrix J ii When is non-zero, the distribution functions of (4) to (6) are obeyed. Conversely, the interaction matrix J ii When is 0, the distribution functions (4) to (7) are followed.
[0059]
number
[0060]
number
[0061] By using the inverse function method or the rejection sampling method for equations (6) and (7), it is possible to randomly sample the next state from these distribution functions.
[0062] Figure 8 shows the probability distribution functions for binary and continuous variables. Next, in the original interaction model representing the mixed binary quadratic programming problem MBQP, the binary variable x i In this case, as mentioned above, ii If we treat it as a binary variable, the next state x i ∈{-1, +1} follows the discrete probability distribution function of equation (8) as shown in the left diagram of Figure 8. On the other hand, when the binary variable is relaxed to a continuous value, it follows the continuous probability distribution function of equation (7) as shown in the right diagram of Figure 8.
[0063]
number
[0064] Figure 9 is a schematic diagram showing how the state moves between local optima in the state space when random sampling of the next state is repeated according to equations (7) and (8). When the variables are binary values, that is, when the transition follows equation (8), it is easy to move a large amount of state space. However, because it is necessary to transition to a state with a large energy value once between local optima, the transition becomes more difficult as the temperature parameter becomes smaller.
[0065] On the other hand, when the variables are continuous, that is, when the transition follows equation (7), it is easy to move finely through the state space. However, because it is possible to move finely, it is thought that even if the temperature parameter is small, the state will not remain at a local optimum solution and some state transitions will occur.
[0066] From equation (3), the variables that are binary in the original interaction model are x i ∈{-1, +1}. Therefore, it is desirable to transition between -1 and +1 efficiently. However, the temperature and g i Which transition is appropriate depends on the value of . The standard deviation can be considered as an index showing the size of the state transition, that is, the spread of the probability distribution function. The probability distribution function p(x i ), the mean under the condition E[·] is the standard deviation SD[x i ] can be expressed by equation (9).
[0067]
number
[0068] Calculating the standard deviation SD using equation (9) for equation (7) gives equation (10). Calculating the standard deviation SD using equation (9) for equation (8) gives equation (11). Here, coth(·) is the hyperbolic cotangent function, and sech(·) is the hyperbolic secant function.
[0069]
number
[0070]
number
[0071] g i The values of the standard deviation SD for / T as expressed by equations (10) and (11) are shown in Figure 10. i The size relationship is actually reversed depending on the size of / T.
[0072] In the original interaction model that represents the mixed binary quadratic programming problem (MBQP), the binary variable x i In updating, the state space can be searched more efficiently by calculating a statistic that indicates the ease of a variable transition, such as the standard deviation mentioned above, and selecting the transition in equation (7) or (8). Of course, it is also possible to switch state transitions by evaluating the ease of a state space transition using other statistics other than the standard deviation, such as the expected value of the distance between the current state and the next state.
[0073] Based on the above findings, the present invention executes the processing flow of Fig. 7. Fig. 7 is a flowchart for explaining the processing (hereinafter referred to as variable update processing) performed by the information processing device 10 when updating each variable in processing step S14 of the ground state search in Fig. 3.
[0074] Here, we start with the variable x i The process starts when an instruction to start updating is given. The variable x to be updated i In the original interaction model before being relaxed to a continuous value, it is judged based on the information of the model setting unit S21 whether the variable x i It is determined whether or not was originally binary, and if it was binary as a variable of the original interaction model, the process proceeds to processing step S22, whereas if it was not a binary variable as a variable of the original interaction model, i.e., it was a continuous variable, the process proceeds to processing step S24.
[0075] In the case where it is binary, the process in step S22 is performed on the corresponding variable x i For the probability distribution function when is a continuous variable and when is a binary variable, a statistical value such as a standard deviation is calculated, and then in process step S23, an update method is selected according to the statistical value using the relationship in Fig. 10. For example, a continuous probability distribution function or a discrete probability distribution function with a larger standard deviation is selected. If the discrete type is selected, the process proceeds to process step S25, and if the continuous type is selected, the process proceeds to process step S24.
[0076] In process step S25, random sampling is performed from the discrete probability distribution function, and in process step S24, random sampling is performed from the continuous probability distribution function to obtain the variable x i This will update the state of the variable.
[0077] After the processing of FIG. 7, the process returns to processing step S15 of FIG. 3, and then the arithmetic device control program (FIG. 2) determines whether or not the simulated annealing SA end condition is satisfied. This is, for example, determining whether or not state update has been executed while changing the temperature parameter T a predetermined number of times. If it is determined that the simulated annealing SA end condition is satisfied (processing step S15: YES), the process proceeds to processing step S16. On the other hand, if the product-sum operation execution unit 24 determines that the stop condition is not satisfied (processing step S15: NO), the process returns to processing step S13, and is repeatedly executed until the end condition is satisfied.
[0078] If it is determined that the simulated annealing SA termination condition is satisfied, then the variable value reading unit 26 reads out the value stored in the variable memory and stores it as the result of the ground state search (processing step S16), and the ground state search process in Figure 3 ends.
[0079] The overall picture of the ground state search process shown in Fig. 3 and the variable update process shown in Fig. 7 is shown in Fig. 11. As mentioned above, the variable update process is iThe index i of is sequentially run from 1 to N, or the index i is randomly run from 1 to N. This can accelerate the magnitude of state updates of binary variables in simulated annealing, and thus make the ground state search more efficient.
[0080] In summary, the information processing device of the present invention is an information processing device for solving a mixed binary quadratic programming problem having an objective function that is a quadratic function for binary variables or continuous variables, and includes a model setting unit that sets the mixed binary quadratic programming problem, a variable value initialization unit that initializes variables of the mixed binary quadratic programming problem, a product-sum operation execution unit that calculates quantities related to state update based on the variables in the mixed binary quadratic programming problem, and a state update control unit that updates the states of the variables based on the quantities related to state update, and the product-sum operation execution unit calculates an index of the magnitude of state transition in a probabilistic state transition as the quantity related to state update based on a probability distribution function defined by the original objective function and a probability distribution function defined by an objective function obtained by relaxing the binary variables to continuous values, and the state update control unit updates the changes while switching state transition methods based on the quantities related to state update, and searches for an optimal solution.
[0081] Although one embodiment has been described in detail so far, the present invention is not limited to the above embodiment, and it goes without saying that various modifications can be made without departing from the gist of the present invention. For example, the above embodiment has been described in detail to explain the present invention in an easy-to-understand manner, and the present invention is not necessarily limited to those having all of the configurations described. In addition, it is possible to add, delete, or replace part of the configuration of the above embodiment with other configurations.
[0082] In addition, the above-mentioned configurations, functional units, processing units, processing means, etc. may be realized in part or in whole by hardware, for example, by designing them as integrated circuits. In addition, the above-mentioned configurations, functions, etc. may be realized in software by a processor interpreting and executing a program that realizes each function. Information such as the program, table, file, etc. that realizes each function can be stored in a memory, a hard disk, a recording device such as an SSD (Solid State Drive), an IC card, an SD card, a DVD, or other recording medium.
[0083] In addition, in each of the above figures, the control lines and information lines are shown as those considered necessary for the explanation, and do not necessarily show all the control lines and information lines in the actual implementation. For example, it may be considered that almost all the components are actually connected to each other.
[0084] The above-described layout of the various functional units, processing units, and databases of the information processing device 10 is merely an example. The layout of the various functional units, processing units, and databases may be changed to an optimal layout in terms of the performance, processing efficiency, communication efficiency, and the like of the hardware and software included in the information processing device 10.
[0085] Furthermore, the configuration (schema, etc.) of the database that stores the various types of data described above can be flexibly changed from the perspective of efficient use of resources, improved processing efficiency, improved access efficiency, improved search efficiency, and the like. [Explanation of symbols]
[0086] 5: System bus 10: Information processing device 11: Processor 12: Main memory 13: Auxiliary storage device 14: Input device 15: Output device 16:Communication equipment 12: Main memory 20: Arithmetic device DB: Storage section D1: MBQP format problem data D2: Calculation device control program 21: Model setting section 22: Variable value initialization section 23: Temperature parameter control section 24: Multiply-and-accumulate execution unit 25: Status update method control section 26: Variable value reading section
Claims
1. An information processing apparatus for solving a mixed binary quadratic programming problem having an objective function that is a quadratic function for binary variables or continuous variables, comprising: a model setting unit that sets a mixed binary quadratic programming problem; a variable value initialization unit that initializes variables of the mixed binary quadratic programming problem; a product-sum operation execution unit that calculates amounts related to state update based on the variables in the mixed binary quadratic programming problem; and a state update control unit that updates states of variables based on the amounts related to state update, the product-sum operation execution unit calculates an index of a magnitude of a state transition in a probabilistic state transition as a quantity related to the state update, based on a probability distribution function defined by an original objective function and a probability distribution function defined by an objective function obtained by relaxing a binary variable to a continuous value; The information processing device, wherein the state update control unit searches for an optimal solution by updating the variables while switching state transition methods based on a quantity related to the state update.
2. 2. The information processing device according to claim 1, a temperature parameter control unit for controlling a temperature parameter of the mixed binary quadratic programming problem, The variable value initialization unit performs initialization including a temperature parameter of the mixed binary quadratic programming problem, The information processing device, wherein the state update control unit updates the variables and the temperature parameter to search for an optimal solution.
3. 2. The information processing device according to claim 1, The information processing device, characterized in that the product-sum operation execution unit calculates the quantity related to the state update based on the statistics of a probability distribution function defined by an original objective function and the statistics of a probability distribution function defined by an objective function obtained by relaxing binary variables to continuous values.
4. 4. The information processing device according to claim 3, The information processing device is characterized in that the product-sum operation execution unit calculates the standard deviation of a probability distribution function for the binary variable and a continuous variable obtained by relaxing the binary variable to a continuous value, and selects a probability distribution function to be used for a state transition.
5. 5. The information processing device according to claim 4, 2. An information processing apparatus comprising: a multiply-and-accumulate execution unit that executes the stochastic state transition method by simulated annealing.
6. 6. The information processing device according to claim 5, 2. An information processing apparatus comprising: a processor for processing the simulated annealing data;
7. 2. The information processing device according to claim 1, The objective function is an energy function, and in order to calculate the energy function, a multiply-and-accumulate operation of values of the variables and a coefficient of the objective function is performed in the multiply-and-accumulate execution unit.
8. 1. An information processing method for solving a mixed binary quadratic programming problem having an objective function that is a quadratic function for binary variables or continuous variables, using a computer, comprising the steps of: The computer includes a model setting unit that sets a mixed binary quadratic programming problem, and a calculation unit. The calculation unit executes a process of initializing variables of the mixed binary quadratic programming problem, calculating a quantity related to a state update based on the variables in the mixed binary quadratic programming problem, and updating a state of the variable based on the quantity related to the state update, Based on a probability distribution function defined by the original objective function and a probability distribution function defined by an objective function obtained by relaxing the binary variable to a continuous value, an index of the magnitude of a state transition in a probabilistic state transition is calculated as a quantity related to the state update; An information processing method, comprising: updating the variables while switching state transition methods based on quantities related to the state update, and searching for an optimal solution.
9. 9. The information processing method according to claim 8, The calculation unit executes a process of controlling a temperature parameter of the mixed binary quadratic programming problem, and Initialize the mixed binary quadratic programming problem including a temperature parameter, An information processing method comprising updating the variables and the temperature parameter to search for an optimal solution.
10. 9. The information processing method according to claim 8, An information processing method characterized in that a quantity related to the state update is obtained based on statistics of a probability distribution function defined by an original objective function and statistics of a probability distribution function defined by an objective function obtained by relaxing binary variables to continuous values.
Citation Information
Patent Citations
Information processing device, calculation device, and information processing method
WO2019216277A1
Computation system, information processing device, and optimum solution search processing method
WO2021220445A1
