WO2022044184A1 - 情報処理システムおよび最適解探索処理方法 - Google Patents
情報処理システムおよび最適解探索処理方法 Download PDFInfo
- Publication number
- WO2022044184A1 WO2022044184A1 PCT/JP2020/032299 JP2020032299W WO2022044184A1 WO 2022044184 A1 WO2022044184 A1 WO 2022044184A1 JP 2020032299 W JP2020032299 W JP 2020032299W WO 2022044184 A1 WO2022044184 A1 WO 2022044184A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- variable
- value
- unit
- memory
- information processing
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N99/00—Subject matter not provided for in other groups of this subclass
Definitions
- the present invention relates to an information processing device, an arithmetic unit, an information processing method, and the like, and relates to a technique for executing an optimum solution search process.
- Patent Document 1 states that "a first memory cell that stores a value representing one spin of the Ising model in a state of three or more and an interaction showing an interaction from another spin that interacts with one spin.
- a second memory cell that stores the action coefficient, a logic circuit that determines the next state of one spin based on a value that expresses the state of another spin and a function that has the interaction coefficient as a constant or a variable.
- a semiconductor device including a plurality of unit units having the above-mentioned is disclosed.
- Patent Document 2 describes a method for realizing an optimum solution search by simultaneously probabilistically updating all spins while satisfying the theoretical background required by the Markov chain Monte Carlo method for an Ising model having an arbitrary coupling. ing.
- Patent Document 3 describes an acquisition means for acquiring candidate information, which is information on candidates for subsets in the partitioning problem, and a singing model corresponding to the partitioning problem based on the candidate information acquired by the acquisition means.
- a generation means for generating the Hamiltonian equation in the above is disclosed.
- An interaction model is defined by a plurality of nodes constituting the model, interactions between the nodes, and if necessary, coefficients that act on each node.
- various models such as the Ising model have been proposed, but all of them can be interpreted as a form of interaction model.
- this energy function is generally a linear sum of multiple terms such as a penalty term representing a constraint. expressed. Since the optimum value of the weight of each term is generally unknown, the solution is searched while changing the weight to obtain an exact solution or a good approximate solution.
- the optimization problem can be calculated in a short time and with low power consumption once the problem to be solved can be expressed by the interaction model.
- the optimization problems that are actually required in the world often impose complicated constraints, and it is difficult to express all of these penalty terms in an interaction model.
- the present invention has been made in view of the above background, and provides a means for solving a complicated optimization problem with various constraints by a mixed quadratic programming problem. The purpose.
- a preferred aspect of the present invention is an information processing system including an arithmetic unit and a computer that controls the arithmetic unit.
- the computer generates solution candidates for an optimization problem including a plurality of constraints, extracts solution candidates satisfying at least a part of the constraints from the solution candidates, and is a mixed integer quadratic based on the extracted solution candidates. It includes a preprocessing unit that generates a planning problem, and an interaction calculation execution unit that inputs data based on the mixed integer quadratic programming problem to the calculation device and executes the calculation.
- the arithmetic unit performs an operation to update a variable of the mixed integer quadratic programming problem, and outputs the variable having the maximum or minimum objective function as a solution.
- a more preferable aspect of the present invention is an information processing system including an arithmetic unit and a computer for controlling the arithmetic unit.
- the computer creates a candidate list by excluding a part of the array from the intermediate input data consisting of a set of arrays, and generates a mixed integer quadratic planning problem from the candidate list, and the mixed integer. It is provided with an interaction calculation execution unit for inputting data based on a secondary planning problem into the calculation device and executing the calculation.
- the arithmetic unit performs an operation to update a variable of the mixed integer quadratic programming problem, and outputs the variable having the maximum or minimum objective function as a solution.
- Another preferable aspect of the present invention is an arithmetic unit and an optimum solution search processing method executed by a computer that controls the arithmetic unit.
- the preprocessing unit of the computer creates a candidate list by excluding a part of the array from the intermediate input data consisting of a set of arrays, and generates secondary planning format problem data from the candidate list.
- the weight setting unit of the computer is a weight signal.
- the sixth step of reading the value of each variable from the variable memory and performing conversion based on the definition area data is executed.
- Notations such as “first”, “second”, and “third” in the present specification and the like are attached to identify components, and do not necessarily limit the number, order, or contents thereof. is not it. Further, the numbers for identifying the components are used for each context, and the numbers used in one context do not always indicate the same composition in the other contexts. Further, it does not prevent the component identified by a certain number from functioning as the component identified by another number.
- a variable memory that stores a value indicating the state of a variable in a mixed integer quadratic design problem and a non-linear coefficient that stores the non-linear coefficient of the state transition calculation block corresponding to the variable memory.
- the memory, the linear coefficient memory that stores the linear coefficient of the state transition calculation block corresponding to the variable memory, the weight input line that receives the weight signal of the state transition calculation block, and the temperature signal of the state transition calculation block are received.
- a difference calculation block that calculates a difference calculation using a temperature input line, a weight signal of the state transition calculation block, a nonlinear coefficient of the state transition calculation block, and a linear coefficient of the state transition calculation block, and a state transition calculation block.
- a sampling block that randomly samples from a probability distribution with interval constraints using the weight signal, the temperature signal of the state transition calculation block, and the output value of the difference calculation block, and the output value of the sampling block and the output value of the variable memory are read out. It is an arithmetic circuit including a next state calculation block for calculating the next state of a variable using the above values.
- an integer programming problem is an optimization problem that includes integer variables.
- a variable that takes an integer value and a variable that takes a real value are mixed, it is called a mixed integer programming problem.
- a mixed integer programming problem that is a quadratic programming problem is called a mixed integer quadratic programming problem.
- a mixed integer quadratic programming problem in which variables that take binary values and variables that take real values are mixed is referred to as a mixed binary quadratic programming problem.
- the purchase ratio of financial products may be 0% or 10% to 100%. Of course, if you do not purchase it, it will be 0%, and if you purchase it, it will be 10% or more of the minimum unit.
- Equation 3 the sets of subscripts ⁇ b and ⁇ c are defined as in Equation 3.
- Equation 2 can also be expressed as Equation 4.
- this optimization problem is a combinatorial optimization problem called the ground state search problem of the Ising model.
- an optimization problem including a search for the base state of an Ising model an optimum solution or an approximate solution is searched for by an algorithm utilizing Markov Chain Monte Carlo methods (hereinafter referred to as MCMC (Markov Chain Monte Carlo methods)).
- FIG. 1 is a conceptual diagram showing a landscape of objective function values for a variable array.
- the horizontal axis of the graph is the variable array s, and the vertical axis is the objective function H (s).
- MCMC repeats a probabilistic transition from the current state s to a state s'near the state s.
- the probability of transition from the state s to the state s' is referred to as a transition probability P (s, s'). Examples of the transition probability P include the metropolis method and the heat-bath algorithm.
- the transition probability has a parameter called temperature, which indicates the ease of transition between states.
- temperature indicates the ease of transition between states.
- SA Simulated Annealing
- MA momentum annealing proposed in Non-Patent Document 1
- the goal of the embodiment shown in the present application is to search for the optimum solution of Eq. 2, but it is possible to obtain the desired solution s + even if the transformation of Eq. 6 is obtained after solving the optimum solution s * of Eq. 5. ..
- the function sgn is a function that returns +1 if the argument is 0 or more, and -1 otherwise.
- the matrix W diag (w 1 , ..., W N ) is an arbitrary diagonal matrix, and vi is a real number moving [ -1 , +1].
- the equation 9 which is the minimization problem of H'(s, v) is introduced.
- Equation 5 can be paraphrased as the minimization problem of Equation 11 (may be treated as a maximization problem).
- Equation 2 the optimum solution of the mixed quadratic programming problem represented by Equation 2 can be obtained from the solution of the constrained quadratic programming problem shown in Equation 11. MCMC is used to find this solution.
- FIG. 2 is a graphical model showing the relationship between the variables of the objective function G in Equation 11.
- the relationship between the variables of the function G can be represented by a complete bipartite graph.
- the only variables that can be multiplied by the variable x i in the function G are y 1 , ..., y N and x i .
- MCMC uses the value of the variable related to the variable when updating the variable value stochastically. That is, when updating the value of the variable x 1 , y 1 , ..., y N and x 1 are obtained, and the other variables (here, x 2 , ..., X N ) are not referred to. This also applies to updating the value of other variables, such as x2 . Therefore, if the value of the variable array y is constant, the theoretical requirement of MCMC is not broken even if the values of the array x are independently and stochastically updated at the same time.
- each value of the array y can be independently and stochastically updated at the same time.
- FIG. 3 is an example of a fully connected graph.
- MCMC is applied directly to the minimization problem of Equation 2, which is the original problem, only one variable is used at a time because the relation of the variable array s is represented by a fully connected graph as shown in FIG. Probability update is not possible and is limited to sequential update.
- variable A i is a value obtained by the equation 13.
- the variables x i and y i are
- the range in which x i can move is ⁇ (2-
- the next state of xi may be sampled. In this method, the next state is determined regardless of the current state of xi .
- y i In the present specification, when the variables of x and y are not distinguished, they may be expressed as s.
- Random numbers that follow the standard normal distribution can be generated by the Box-Muller method. Since the domain is limited here, the algorithm shown in Non-Patent Document 2 may be used.
- Non-Patent Document 3 proposes an over-mitigation method.
- K states are sampled from the Boltzmann distribution at temperature T.
- x c K + 1-r is adopted as the next state. In this method, the next state depends on the current state of xi .
- FIGS. 4 to 6 show the configuration of the information processing apparatus that realizes the present invention.
- FIG. 4 is an example of an information processing device that searches for an optimum solution for a mixed binary quadratic programming problem.
- the information processing device 10 includes a processor 11, a main storage device 12, an auxiliary storage device 13, an input device 14, an output device 15, a communication device 16, one or more arithmetic units 20, and these.
- a system bus 5 for communicably connecting the devices is provided. Even if the information processing device 10 is partially or wholly realized by using a virtual information processing resource such as a cloud server provided by a cloud system (CloudSystem), for example. good. Further, the information processing device 10 may be realized by, for example, a plurality of information processing devices that are communicably connected and operate in cooperation with each other.
- CloudSystem cloud system
- the processor 11 is configured by using, for example, a CPU (Central Processing Unit) or an MPU (Micro Processing Unit).
- the main storage device 12 is a device for storing programs and data, for example, ROM (ReadOnlyMemory), SRAM (StaticRandomAccessMemory), NVRAM (NonVolatileRAM), mask ROM (MaskReadOnlyMemory), PROM. (Programmable ROM), etc.), RAM (RandomAccessMemory) (DRAM (DynamicRandomAccessMemory), etc.), etc.
- the auxiliary storage device 13 is a hard disk drive (Hard Disk Drive), a flash memory (Flash Memory), an SSD (Solid State Drive), an optical storage device (CD (Compact Disc), DVD (Digital Versatile Disc), etc.) and the like. ..
- the programs and data stored in the auxiliary storage device 13 are read into the main storage device 12 at any time.
- the input device 14 is a user interface that receives information input from the user, and is, for example, a keyboard, a mouse, a card reader, a touch panel, or the like.
- the output device 15 is a user interface that provides information to the user, and is, for example, a display device (LCD (Liquid Crystal Display), graphic card, etc.) that visualizes various information, an audio output device (speaker), a printing device, and the like. ..
- the communication device 16 is a communication interface that communicates with other devices, and is, for example, an NIC (NetworkInterfaceCard), a wireless communication module, a USB (UniversalSerialInterface) module, a serial communication module, and the like.
- the arithmetic unit 20 is a device that executes processing related to the search for the optimum solution of the mixed binary quadratic programming problem.
- the arithmetic unit 20 may take the form of an expansion card to be mounted on the information processing unit 10, such as a GPU (Graphics Processing Unit).
- the arithmetic unit 20 is composed of 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 unit 20 includes a control device, a storage device, an interface for connecting to the system bus 5, and sends / receives commands and information to / from the processor 11 via the system bus 5.
- the arithmetic unit 20 may be connected to the other arithmetic unit 20 so as to be communicable via a communication line, and may operate in cooperation with the other arithmetic unit 20.
- the function realized by the arithmetic unit 20 may be realized, for example, by causing a processor (CPU, GPU, etc.) to execute a program.
- the arithmetic unit 20 shown in FIG. 4 will be described later in FIG.
- One or a plurality of arithmetic units 20 can be mounted.
- FIG. 5 is a diagram for explaining the operating principle of the arithmetic unit 20, and is a block diagram of a circuit (hereinafter, referred to as an arithmetic circuit 500) constituting the arithmetic unit 20.
- the arithmetic circuit 500 realizes a function of sampling the variable array x 1 , ..., X N or the variable array y 1 , ..., Y N from the Boltzmann distribution (Equation 12) at the temperature T.
- the operating principle of the arithmetic unit 20 will be described with reference to the figure.
- the arithmetic circuit 500 includes a variable memory 511, a nonlinear coefficient memory 512, a linear coefficient memory 513, a difference calculation block 514, a sampling block 515, and a next state determination block 516.
- variable memory 511 of each arithmetic circuit 500 stores information indicating the variables x 1 , ..., XN and y1, ..., Y N described above (see FIG. 2).
- Information representing the matrix J is stored in the nonlinear coefficient memory 512.
- the matrix J is generally a symmetric matrix, and this symmetry can be used to reduce the usage of the nonlinear coefficient memory 512.
- Information representing the vector h is stored in the linear coefficient memory 513.
- control signal EN As shown in the figure, the control signal EN, the weight signal SW, and the temperature signal TE are input to the arithmetic circuit 500.
- the signal EN is a signal that periodically repeats the values of H (high) and L (low), and represents which of the variable arrays x and y is updated. For example, when EN is H, the variable array x is updated, and when L, y is updated. By this signal EN, the variables x 1 , ..., X N are updated at the same time, and the variables y 1 , ..., Y N are updated at the same time.
- the signal EN is input only to the sampling block 515 for simplification, but it is similarly applied to other places where this signal is required, such as a variable memory.
- the signal SW is a signal representing a vector of N elements representing diagonal components of the diagonal matrix W.
- the difference calculation block 514 the value of the matrix J stored in the nonlinear coefficient memory 512, the vector h stored in the linear coefficient memory 513, the signal SW, and the variable s (x or) stored in the variable memory 511. y) is input.
- the difference calculation block 514 outputs y + h when the signal EN is H (J + diag (w 1 , ..., w N )) and x + h when EN is L (J + diag (w 1 , ..., w N )). do. This output value corresponds to the above- mentioned Ai.
- the sampling block 515 receives the output and signal SW of the difference calculation block 514, the signal TW holding the value of the temperature parameter, the signal EN, and the values of other variables. And as the i-th element, when the signal EN is H- (2-
- the next state determination block 516 determines the next state of the variable based on one or more values output from the sampling block 515. If the MCMC update rule is defined as a simple heat bath method, the next state determination block 516 may receive only one output value of the sampling block 515 and write it as it is to the variable memory 511. Further, if a known over-relaxation method is used as the update rule of MCMC, the next state determination block 516 receives a plurality of values from the sampling block 515 and the current value of the variable to be updated from the variable memory 511, and according to the over-relaxation method. Select one and write it to the variable memory 511. As is well known, in the over-mitigation method, the next state is determined so that the correlation with the immediately preceding state is negative.
- FIG. 6 shows the main functions (software configuration) of the information processing apparatus 10.
- the information processing apparatus 10 includes a storage unit 600, a model conversion unit 611, a model coefficient setting unit 612, a weight setting unit 613, a variable value initialization unit 614, a temperature setting unit 615, and an interaction calculation execution unit. It includes 616, a variable value reading unit 617, and a preprocessing unit 618. These functions are realized by the processor 11 reading and executing the program stored in the main storage device 12, or by the hardware provided in the arithmetic unit 20.
- the information processing device 10 may have other functions such as an operating system, a file system, a device driver, and a DBMS (DataBase Management System).
- DBMS DataBase Management System
- the storage unit 600 stores the problem data 601, the secondary plan format problem data 602, the definition area data 603, the arithmetic unit control program 604, the intermediate input data 619, and the candidate list 620 in the main storage device 12 or the auxiliary storage. Store in the device 13.
- the problem data 601 is data in which, for example, an optimization problem or the like is described in a known predetermined description format.
- the problem data 601 is created by the preprocessing unit 618 from the candidate list 620 described later.
- the analysis input data input by the user via the user interface is converted into intermediate input data 619 composed of a combination of candidates by the preprocessing unit 618.
- the analysis input data is data to be analyzed, such as personnel data and order data.
- Personnel data is, for example, data such as a person in charge ID, a place of work, a qualification, and a work area.
- the order data is, for example, data related to orders such as product / service contents, location, and delivery date. These are just examples, and if the data can be treated as a material for an optimization problem, it is not necessary to limit the data.
- the intermediate input data 619 is generated from the analysis input data.
- the above example includes, for example, a combination of an array of candidates representing which person is in charge of which order.
- the solution is composed of a formula expressed by superposition of limited partial decompositions.
- the preprocessing unit 618 further removes a predetermined candidate that does not satisfy the constraint condition from the intermediate input data 619 to generate a candidate list 620. Such preprocessing eliminates the need to represent complex constraints in an interaction model.
- the preprocessing unit 618 generates problem data 601 from the candidate list 620.
- the quadratic programming format problem data 602 is data generated by the model transformation unit 611 converting the problem data 601 into data in a format that matches the format of the quadratic programming problem represented by Equation 4. In this conversion, the domain of each given variable is written in the domain data 603. The domain data indicates, for example, whether each variable takes a binary value or a real value.
- the arithmetic unit control program 604 is a program that is executed when the interaction arithmetic execution unit 616 controls the arithmetic unit 20, or is loaded by the interaction arithmetic execution unit 616 into each arithmetic unit 20 and executed by the arithmetic unit 20. be.
- the model transformation unit 611 converts the problem data 601 into the quadratic programming format problem data 602, which is the format of the quadratic programming problem.
- the function of deriving the equation 11 from the equation 1 may be implemented in the model transformation unit 611 as software or hardware.
- the function of the model conversion unit 611 does not necessarily have to be implemented in the information processing device 10, and the information processing device 10 inputs the secondary plan format problem data 602 generated by another information processing device or the like to the input device 14 or communication. It may be taken in via the device 16.
- the model coefficient setting unit 612 sets the matrix J of the equation 11 in the nonlinear coefficient memory 512 and the vector h in the linear coefficient memory 513 based on the quadratic planning format problem data 602.
- the variable value initialization unit 614 initializes the value of each variable stored in the variable memory 511 of the arithmetic unit 20.
- the variable value initialization unit 614 may determine, for example, by randomly sampling the value of each variable uniformly from -1 or more and +1 or less. At this time, care must be taken to satisfy the restrictions on variables
- the temperature setting unit 615 sets the temperature T used when the interaction calculation execution unit 616 searches for the optimum solution.
- the interaction operation execution unit 616 is an operation for searching the variable arrays x and y that minimize the function G represented by the equation 11 for each temperature T set by the temperature setting unit 615 (hereinafter referred to as an interaction operation). Is executed by the arithmetic unit 20. In the interaction calculation, the interaction calculation execution unit 616 changes, for example, the temperature T from the higher side to the lower side.
- the variable value reading unit 617 reads the variable arrays x and y stored in the variable memory 511 when the optimum solution search by the interaction calculation execution unit 616 is completed.
- the value read here is the solution of Equation 11.
- the domain data 603 is read out, and the vector s + obtained by the equation 6 is output to the output device 15 and the communication device 16 as the final solution. That is, if the i-th domain is found to be ⁇ -1, +1 ⁇ in the domain data 603, sgn (s * i ) is output, and if the i-th domain is [-1, +1], s i itself is output. Is to do. In this way, a solution according to the defined range is obtained.
- FIG. 7 is a flowchart illustrating a process (hereinafter referred to as an optimum solution search process S700) performed by the information processing apparatus 10 when searching for the optimum solution.
- the optimum solution search process S700 will be described with reference to the figure.
- the letter "S" attached before the reference numeral means a processing step.
- the optimum solution search process S700 is started by receiving an instruction from the user or the like via the input device 14, for example.
- the preprocessing unit 618 first generates the intermediate input data 619 from the analysis input data, generates the candidate list 620 from the intermediate input data 619, and generates the problem data 601 from the candidate list 620 (S710). ).
- the model transformation unit 611 converts the problem data 601 into the quadratic programming format problem data 602 (S711).
- the quadratic programming format problem data expresses, for example, the matrix J and the vector h in the function H expressed by the equation 1 in an arbitrary format. If the storage unit 600 has already stored the quadratic programming format problem data 602, the process S711 is omitted.
- the processing of S711 and the processing after S712 may be executed by different devices. Further, the processing of S711 and the processing after S712 may be executed at different timings (for example, it is conceivable to perform the processing of S711 in advance).
- the model coefficient setting unit 612 sets the values of the matrix J and the vector h in the nonlinear coefficient memory 512 and the linear coefficient memory 513 (S712).
- the memory value can also be set or edited by the user via a user interface (eg, implemented by an input device 14, an output device 15, a communication device 16, etc.).
- the weight setting unit 613 determines the value of the signal SW.
- the signal SW is allowed to take an arbitrary value in searching for the optimum solution. Therefore, the signal value may always be 0. In this case, the calculation load can be reduced.
- it may be determined from the eigenvalues of the matrix J. Alternatively, it may be determined from the sum of rows of the matrix J.
- the calculation of the value calculation of the signal SW may be executed in the arithmetic unit 20 or in the processor 11. Alternatively, the user may set it by himself (S713).
- variable value initialization unit 614 initializes the value of each variable stored in the variable memory 511 (S714).
- the value stored in the variable memory 511 is a continuous value. As mentioned earlier, the initial value may be random. With the above, the parameter expressing the equation 11 is set.
- the above-mentioned subscript k represents the type of temperature T to be set.
- the method of Patent Document 1 can be adopted.
- the interaction calculation execution unit 616 executes stochastic simultaneous update of the variable array by the calculation of the calculation circuit 500 shown in FIG. 5 (S716).
- the interaction calculation execution unit 616 determines whether or not the stop condition is satisfied (for example, whether or not the temperature T has reached a preset minimum temperature) (S717).
- the interaction calculation execution unit 616 determines that the stop condition is satisfied (S717: YES)
- the process proceeds to S718.
- the interaction calculation execution unit 616 determines that the stop condition is not satisfied (S717: NO)
- the process returns to S716.
- variable value reading unit 617 reads the value of the variable stored in the variable memory 511 and the definition area of each variable of the secondary plan format problem data 602 stored in the definition area data 603. Then, the vector through the transformation based on the equation 6 is calculated and output as the solution of the equation 2 or the equation 4. This completes the optimal solution search process S700.
- the information processing apparatus 10 of the present embodiment it is possible to efficiently search for the optimum solution of the mixed quadratic programming problem. Therefore, the optimization problem can be solved efficiently. Since the information processing device 10 (including the arithmetic unit 20) has a simple structure, it can be manufactured inexpensively and easily.
- the arithmetic circuit 500 may be configured by software or hardware as long as it has a function of executing a calculation for solving the optimization problem described above.
- the annealing method not only the hardware mounted by an electronic circuit (digital circuit or the like) but also the method of mounting by a superconducting circuit or the like may be used.
- hardware that realizes the Ising model other than the annealing method may be used.
- a laser network method optical parametric oscillation
- a quantum neural network and the like are known.
- a quantum gate method in which the calculation performed by the Ising model is replaced with gates such as an Adamal gate, a rotation gate, and a control NOT gate can also be adopted as the configuration of this embodiment.
- CMOS Complementary Metal-Oxide Semiconductor
- FPGA Field Programmable Gate Array
- FIG. 8 is a block diagram showing a circuit configuration example when the SRAM technology is applied to the arithmetic circuit 500 of this embodiment.
- a plurality of units 801 constitute an array unit 802. Such a configuration can be manufactured by applying semiconductor manufacturing technology.
- One unit 801 includes a multi-valued memory 901 that stores one of the variables x 1 , ..., XN and y1, ..., Y N , and a configuration for updating the value of the multi-valued memory 901. .. That is, 2N units 801 are prepared.
- the configuration example of FIG. 8 will be described with reference to the generalized configuration of FIG.
- the data stored in the nonlinear coefficient memory 512 and the linear coefficient memory 513 is set by the model coefficient setting unit 612.
- the nonlinear coefficient memory 512 stores an N ⁇ N matrix J, which is commonly used by all units 801.
- the linear coefficient memory 513 stores the N-dimensional vector h, which is commonly used by all the units 801. In order to reduce the circuit scale, these memories are common to each unit 801. Therefore, the nonlinear coefficient memory 512 and the linear coefficient memory 513 supply the coefficients J and h to all the units 801 but omit the signal line for that purpose in FIG.
- each unit 801 may individually include the nonlinear coefficient memory 512 and the linear coefficient memory 513.
- the weight memory 803 stores a vector of N elements (w 1 , ..., W N ) representing the diagonal components of the diagonal matrix W. This data is set by the weight setting unit 613. Since the i-th unit that stores x i and y i uses the i -th component wi, it is necessary to switch the value of the signal SW for each unit 801. In FIG. 8, the signal line for supplying the signal SW to the unit 801 is omitted.
- the temperature signal TE supplied from the temperature setting unit 615 is supplied to all units 801.
- the function and configuration of the temperature signal follow the conventional technology.
- the signal line that supplies the signal TE to the unit 801 is omitted.
- the interaction driver 804 alternately inputs a signal permitting the update of the variable x and a signal permitting the update of the variable y to each unit 801.
- a signal permitting the update of the variable x and a signal permitting the update of the variable y to each unit 801.
- the variables x 1 to x N are updated at the same time
- the variables y 1 to y N are updated at the same time.
- the SRAM interface 805 writes and reads from the memory that stores the variables of the unit 801 created by applying the circuit configuration of the SRAM.
- the variable read after the processing in the arithmetic circuit 500 is completed is sent to the variable value reading unit 617.
- the variable value reading unit 617 obtains a solution to the mixed binary quadratic planning problem by outputting the read variable as a continuous value or a binary value based on the domain data 603.
- the controller 806 initializes the calculation circuit 500 and reports the end of processing according to the instruction of the interaction calculation execution unit 616.
- FIG. 9 is a diagram showing a circuit configuration example of one unit 801.
- One unit includes a multi-valued memory 901 that stores any one of continuous variables x 1 , ..., XN and y1, ..., Y N.
- the difference calculation circuit 902 realizes the function of the difference calculation block 514.
- the variable stored in the multi-valued memory 901 is any one of x 1 , ..., X N
- the vector of (y 1 , ..., y N ) is input to the difference calculation circuit 902.
- a vector of (x 1 , ..., X N ) is input.
- These variable vectors are generated by reading from the multi-valued memory 901 of the other unit 801 by the SRAM interface 805. Further, the N ⁇ N matrix J and the N-dimensional vector h, which are coefficients, are input. Further, the weight wi is input.
- the difference calculation circuit 902 outputs the value Ai of the i -th row of (J + diag (w 1 , ..., w N )) s + h (s is a variable vector of x or y) for these inputs. Become.
- the sampling circuit 903 realizes the function of the sampling block 515.
- the output A i , the signal EN, the signal SW, the signal TE, and the variable stored in the multi-valued memory 901 are y i
- the variable stored in the multi-valued memory 901 is y i .
- x i is input.
- the candidate of the next state of the variable is sampled from the existence probability p ( si ) of the variable s i based on the equation 12.
- the state determination circuit 904 determines the next state of the variable based on one or a plurality of candidates output from the sampling circuit 903. In the state determination circuit 904, for example, when a plurality of candidates are obtained from the sampling circuit 903 when the over-relaxation method is obeyed, the candidate whose correlation with the state immediately before the multi-valued memory 901 is negative is selected and the next state is selected. decide. The determined next state is stored in the multi-valued memory 901.
- the difference calculation block 514, the sampling block 515, and the next state determination block 516 are assumed to be hardware such as FPGA, but software can be implemented by, for example, a GPU arranged in the unit.
- a GPU arranged in the unit.
- FIG. 10 is a diagram conceptually explaining the meaning of the preprocessing S710 in the flow of FIG. 7.
- the solution candidates of the problem to be solved are generated in advance by a general-purpose computer, and the set of solution candidates is divided into a set partitioning / set cover problem that satisfies the conditions.
- the set division / set cover problem is equivalent to the minimization problem (or maximization problem) of the objective function represented by the interaction model such as the Ising model, and the annealing technology for the interaction model such as quantum annealing and CMOS annealing. It is possible to solve with. This makes it possible to increase the types of problems that can be solved by the annealing technique.
- the analysis input data 1001 includes personnel information such as the ID, qualification, and location of the candidate person in charge, and order information such as the content of the case, the place, the time limit, and the skill required for the response. It also includes evaluation parameters and the like for determining whether or not it is optimal.
- the preprocessing unit 618 of the information processing apparatus 10 generates, for example, tabular intermediate input data 619 from the analysis input data 1001 input by the intermediate input data molding S7101 in the preprocessing S710 and stores it.
- the candidate list 620 is generated from the intermediate input data 619 by the candidate list generation S7102.
- FIG. 11 is an example of the candidate list 620, and is a table diagram illustrating an optimization algorithm called a column generation method. The purpose of this algorithm is to choose some from multiple candidates.
- the candidate here is a predetermined array, and the optimum combination is selected from the set of arrays according to a predetermined condition.
- each candidate represents "a matter in charge of one person in charge”.
- Candidate 1 is "person in charge 1 is in charge of only case 1”
- candidate 3 is "person in charge 2 is in charge of case 3” and the like.
- the problem of finding the combination that minimizes the total cost among the combinations of subsets that include each element (for example, matter) of a given set is called the set partitioning problem.
- the evaluation value c i of the candidate is the evaluation of the candidate, and the method of determining it is arbitrary.
- the preprocessing unit 618 can combine the data of the analysis input data 1001 by a general program to automatically create a set of candidates as shown in FIG. 11 or by editing by the operator to obtain the intermediate input data 619. ..
- the algorithm of an automatic program is the combination generation of mechanical data.
- the candidates 1 to n since the selected candidates are combined to obtain the answer to the entire question, the candidates 1 to n, which are the parts of the entire answer, must be feasible in the first place.
- the intermediate input data 819 has all combinations in which each person in charge is in charge of 1 to 1000 cases from 1000 cases as a candidate. However, if there is "the person in charge is in charge of matter 1 and matter 100" in the intermediate input data 619, and if each matter is as follows, one person in charge may be in charge of both matters. It's impossible. Project 1: Work in Tokyo at 10-12 o'clock on 6/25 (Thursday) Project 100: Work in Osaka at 10-12 o'clock on 6/25 (Thursday)
- the candidate list generation S7102 generates a candidate list 620 by excluding inappropriate candidates based on the problem settings stored in the intermediate input data 619. That is, in the preprocessing S710, the candidates included in the intermediate input data 619 are selected by the conditional search by a general information processing apparatus instead of the optimization calculation. Thus, the preprocessing S710 generates a candidate group of the candidate list 620. Then, the problem data 601 of the optimization calculation is generated from the candidate list 620.
- variable ai indicating the presence or absence of the skill of the person in charge can be compared with the variable indicating the presence or absence of the skill required for the matter 2.
- the candidate can be selected immediately by not making the candidate including the case 2.
- the preprocessing S710 has a problem in the data conversion S711 because the problem generation S710 in the preprocessing S710 can generate the problem data 601 from the candidate list 620 and then deal with the mixed quadratic programming problem.
- the data 601 is further converted into the quadratic programming format problem data 602 to complete the generation of the optimization problem.
- Equation 15 the penalty term for expressing the constraint of x 1 + x 2 ⁇ 1 is as shown in Equation 15.
- Patent Document 3 describes that a Hamiltonian in an Ising model showing a set partitioning problem to be calculated is generated based on the acquired candidate information, but since a continuous variable is not used, only a single auxiliary variable is generated. Cannot be expressed by.
- x i is a binary variable of "1" or "0”
- the variable z is an auxiliary variable having a range of real numbers of 0 or more and 1 or less.
- the penalty function Pi is combined to obtain the objective function H of Eq. 16. This is the function to be minimized.
- the degree of freedom of the value that can be selected as the auxiliary variable z is large.
- a real value can be selected as xi , and the range of application is wide. In the above case, "1" can be replaced with "+1" and "0" can be replaced with "-1".
- the constant ci represents the evaluation value of each candidate.
- the larger the ci the higher the evaluation of the candidate.
- c i may be arbitrarily set by the user or may be set automatically. For example, the higher the number of projects in charge, the higher the evaluation value.
- quadratic programming format problem data 602 is subjected to annealing calculation by the arithmetic unit 20 suitable for the optimization calculation as described above, and the result data 1002 is read out.
- the solution candidates obtained by column generation are evaluated. If the overall solution candidates obtained do not meet expectations, it is possible that the quality of the answers obtained by annealing is not good, or that the evaluation value ci set for each candidate is inappropriate. Therefore, it is included in this process to determine whether the quality of the solution has reached the standard, and if not, adjust the magnitude of the penalty coefficient and the evaluation value for each candidate and recalculate. Answer 1006 is obtained from the finally obtained solution candidate 1004.
- Whether the constraint condition of the problem is processed by the preprocessing S710 or the optimization problem is not particularly limited.
- the part that is complicated to handle in the mixed quadratic programming problem may be processed on a case-by-case basis, such as being processed by S710.
- each of the above configurations, functional units, processing units, processing means, etc. may be realized by hardware by designing a part or all of them by, for example, an integrated circuit.
- each of the above configurations, functions, and the like may be realized by software by the processor interpreting and executing a program that realizes each function.
- Information such as programs, tables, and files that realize each function can be placed in a memory, a hard disk, a recording device such as an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.
- control lines and information lines are shown as necessary for explanation, and not all the control lines and information lines on the mounting are necessarily shown. For example, in practice almost all configurations may be considered interconnected.
- the arrangement form of various functional units, various processing units, and various databases of the information processing apparatus 10 described above is only an example.
- the arrangement form of the various function units, the various processing units, and the various databases can be changed to the optimum arrangement form from the viewpoint of the performance, processing efficiency, communication efficiency, and the like of the hardware and software included in the information processing apparatus 10.
- the configuration of the database (schema, etc.) that stores the various data described above can be flexibly changed from the viewpoints of efficient use of resources, improvement of processing efficiency, improvement of access efficiency, improvement of search efficiency, and the like.
- It can be used for information processing devices, arithmetic units, information processing methods, etc.
- 10 information processing device 11 processor, 12 main storage device, 20 arithmetic device, 511 variable memory, 512 nonlinear coefficient memory, 513 linear coefficient memory, 514 difference calculation block, 515 sampling block, 516th order determination block, 600 storage unit, 601 problem data, 602 secondary plan format problem data, 603 definition area data, 604 arithmetic unit control program, 611 model conversion unit, 612 model coefficient setting unit, 613 weight setting unit, 614 variable value initialization unit, 615 temperature setting unit , 616 Interaction calculation execution unit, 617 Variable value reading unit
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Complex Calculations (AREA)
Abstract
Description
r={(1+x)/2} × {0.1+0.9×(1+y))/2}
とすることで表現できる。
案件1:6/25(木) 10-12時に東京で作業
案件100:6/25(木) 10-12時に大阪で作業
x1+x2=1
なる等式制約で表現される。
また例えば作業員2に関しては、候補3~候補5のうちの一つのみを選択するようにしなければならない。このために与えられる等式制約は、
x3+x4+x5=1
である。
・候補1, 2いずれか1つが選ばれる
・候補1, 2いずれも選ばれない
これを数式で表現すると
x1+x2≦1
になる。
このような不等式制約の制約P1をペナルティー法で表現するため、0から1の間を動く連続変数zを導入してペナルティー項を作成する。
x1+x2+x3+…x100≦20
を表現しなければならない。この場合には、単一の2値変数zだけでは、この不等式を表現するペナルティー関数Pxが作れない。そのため、連続変数zを導入することにより、
Px=(x1+x2+x3+…x100-20z)2
というように、単一の補助変数のみで表現することが可能になる。特許文献3には、取得された候補情報に基づいて、計算対象の集合分割問題を示すイジングモデルにおけるハミルトニアンを生成することが記載されるが、連続変数を利用しないため、単一の補助変数のみで表現することができない。
Claims (14)
- 演算装置と、前記演算装置を制御する計算機を備える情報処理システムであって、
前記計算機は、
複数の制約を含む最適化問題の解候補を生成し、前記解候補から前記制約の少なくとも一部を満たす解候補を抽出し、抽出された解候補に基づいて混合整数二次計画問題を生成する前処理部と、
前記混合整数二次計画問題に基づくデータを前記演算装置に入力して、演算を実行させる相互作用演算実行部と、
を備え
前記演算装置は、
前記混合整数二次計画問題の変数を更新する演算を行い、目的関数を最大または最小とする前記変数を解として出力するものである、
情報処理システム。 - 前記前処理部は、
配列の組からなる中間入力データから配列の一部を除外して候補リストを作成し、前記候補リストから混合整数二次計画問題を生成する、
請求項1記載の情報処理システム。 - 前記演算装置は、
前記混合整数二次計画問題の変数の状態を示す値を記憶する変数メモリと、
前記変数の状態を示す値の次状態を計算する状態遷移計算ブロックと、
前記状態遷移計算ブロックの非線形係数を記憶する非線形係数メモリと、
前記状態遷移計算ブロックの線形係数を記憶する線形係数メモリと、
前記状態遷移計算ブロックの重み信号を受信する重み入力線と、
前記状態遷移計算ブロックの温度信号を受信する温度入力線と、を備え、
前記状態遷移計算ブロックは、
前記重み信号と前記非線形係数と前記線形係数を用いて差分計算を計算する差分計算ブロックと、
前記重み信号と前記温度信号と前記差分計算ブロックの出力値を用いて、区間制約付きの確率分布からランダムにサンプリングするサンプリングブロックと、
前記変数メモリから読み出した値を用いて、変数の次状態を計算する次状態決定ブロックと、を備える、
請求項2記載の情報処理システム。 - 前記変数メモリは、前記変数の状態を示す値x1、…、xNおよびy1、…、yNとして連続値を記憶する、
請求項3記載の情報処理システム。 - 前記混合整数二次計画問題の変数の定義域を記憶する記憶部と、
前記変数メモリから値を読み出し、前記変数の定義域に基づいて、前記変数メモリに格納されている連続値を2値に変換する変数値読出部と、を備える、
請求項4記載の情報処理システム。 - 前記非線形係数Jは、N×N行列であり、
前記線形係数hは、N次元ベクトルであり、
前記重み信号SWは、対角行列Wの対角成分w1、…、wNを表すN要素のベクトルを表す信号である、
請求項4記載の情報処理システム。 - 前記差分計算ブロックには、前記非線形係数J、前記線形係数h、前記重み信号SW、および前記変数メモリに記憶されている値が入力され、(J+diag(w1、・・・、wN))s+hを出力し、
ただし、sはN次元ベクトル(x1、…、xN)および(y1、…、yN)のいずれかである、
請求項6記載の情報処理システム。 - 前記非線形係数Jは対称行列である、
請求項6記載の情報処理システム。 - 前記非線形係数Jのi行i列目の要素は0である、
請求項8記載の情報処理システム。 - 前記サンプリングブロックには、前記差分計算ブロックの出力A、前記重み信号SW、前記温度信号TE、制御信号EN、および、前記変数メモリに記憶されている値が入力され、
前記制御信号ENが第1の値のとき-(2-|yi|)以上(2-|yi|)以下、前記制御信号ENが第2の値のとき-(2-|xi|)以上(2-|xi|)以下を定義域とする正規分布からランダムに1または複数の値をサンプリングして出力し、
前記正規分布は、前記出力A、前記重み信号SW、および前記温度信号TEに基づいて形成される、
請求項7記載の情報処理システム。 - 前記正規分布は、平均Ai/wi、分散T/wiの正規分布であり、
ただし、Aiは前記出力Aのi番目の値、Tは前記温度信号TEの値である、
請求項10記載の情報処理システム。 - 前記変数の状態を示す値x1、…、xNおよびy1、…、yNの一つを記憶する多値メモリを備えるユニットを複数備え、
前記ユニットのそれぞれは、前記差分計算ブロックの一部の機能を実行する差分計算回路と、前記サンプリングブロックの一部の機能を実行するサンプリング回路と、前記次状態決定ブロックの一部の機能を実行する次状態決定回路を備え、
前記変数の状態を示す値xiまたはyiの一つを記憶する多値メモリを備えるユニットにおいては、
前記差分計算回路は、前記非線形係数J、前記線形係数h、対角行列Wのi番目の対角成分wi、および自ユニットの多値メモリが記憶する値がxiのときはN次元ベクトル(y1、…、yN)を、自ユニットの多値メモリが記憶する値がyiのときはN次元ベクトル(x1、…、xN)を入力とし、
Ai=hi+wisi+Σijsj
(ただし、hiは線形係数hのi番目の要素、sは自ユニットの多値メモリが記憶する値がxiのときはy、自ユニットの多値メモリが記憶する値がyiのときはxを示す)
を出力とする、
請求項7記載の情報処理システム。 - 前記サンプリング回路は、前記差分計算回路の出力Ai、前記対角成分wi、前記温度信号TE、制御信号EN、および、前記変数メモリに記憶されている値が入力され、
前記制御信号ENが第1の値のとき-(2-|yi|)以上(2-|yi|)以下、前記制御信号ENが第2の値のとき-(2-|xi|)以上(2-|xi|)以下を定義域とする平均Ai/wi、分散T/wiの正規分布からランダムに1または複数の値をサンプリングして出力する、
(ただし、Tは前記温度信号TEの値である)
請求項12記載の情報処理システム。 - 演算装置と、前記演算装置を制御する計算機で実行する最適解探索処理方法であって、
前記計算機の前処理部が、配列の組からなる中間入力データから配列の一部を除外して候補リストを作成し、前記候補リストから二次計画形式問題データを生成する前処理ステップ、
前記計算機の記憶部に、前記二次計画形式問題データと、前記二次計画形式問題データの変数の定義域である定義域データを記憶する第1のステップ、
前記計算機のモデル係数設定部が、前記二次計画形式問題データに基づいて、非線形係数メモリに非線形係数Jを設定し、線形係数メモリに線形係数hの値を設定する第2のステップ、
前記計算機の重み設定部が重み信号SWの値を決定する第3のステップ、
前記計算機の変数値初期化部が、変数メモリに格納されている各変数の値を初期化する第4のステップ、
前記計算機の相互作用演算実行部が、前記非線形係数J、前記線形係数h、前記重み信号SWを用いて、前記演算装置の状態遷移計算ブロックに前記変数の次状態の計算を実行させる第5のステップ、
前記計算機の変数読出部が、前記変数メモリから各変数の値を読み出し、前記定義域データに基づいて変換を行う第6のステップ、
を実行する最適解探索処理方法。
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/JP2020/032299 WO2022044184A1 (ja) | 2020-08-27 | 2020-08-27 | 情報処理システムおよび最適解探索処理方法 |
JP2022544983A JP7425210B2 (ja) | 2020-08-27 | 2020-08-27 | 情報処理システムおよび最適解探索処理方法 |
AU2020465147A AU2020465147B2 (en) | 2020-08-27 | 2020-08-27 | Information processing system and optimal solution search processing method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/JP2020/032299 WO2022044184A1 (ja) | 2020-08-27 | 2020-08-27 | 情報処理システムおよび最適解探索処理方法 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022044184A1 true WO2022044184A1 (ja) | 2022-03-03 |
Family
ID=80352829
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2020/032299 WO2022044184A1 (ja) | 2020-08-27 | 2020-08-27 | 情報処理システムおよび最適解探索処理方法 |
Country Status (3)
Country | Link |
---|---|
JP (1) | JP7425210B2 (ja) |
AU (1) | AU2020465147B2 (ja) |
WO (1) | WO2022044184A1 (ja) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2000112505A (ja) * | 1998-10-06 | 2000-04-21 | Nec Corp | 生産計画立案装置及び生産計画立案方法並びにプログラムを記録した機械読み取り可能な記録媒体 |
WO2017056368A1 (ja) * | 2015-09-30 | 2017-04-06 | 日本電気株式会社 | 最適化システム、最適化方法および最適化プログラム |
JP2017151810A (ja) * | 2016-02-25 | 2017-08-31 | 新日鉄住金ソリューションズ株式会社 | 情報処理装置、情報処理方法及びプログラム |
-
2020
- 2020-08-27 WO PCT/JP2020/032299 patent/WO2022044184A1/ja active Application Filing
- 2020-08-27 JP JP2022544983A patent/JP7425210B2/ja active Active
- 2020-08-27 AU AU2020465147A patent/AU2020465147B2/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2000112505A (ja) * | 1998-10-06 | 2000-04-21 | Nec Corp | 生産計画立案装置及び生産計画立案方法並びにプログラムを記録した機械読み取り可能な記録媒体 |
WO2017056368A1 (ja) * | 2015-09-30 | 2017-04-06 | 日本電気株式会社 | 最適化システム、最適化方法および最適化プログラム |
JP2017151810A (ja) * | 2016-02-25 | 2017-08-31 | 新日鉄住金ソリューションズ株式会社 | 情報処理装置、情報処理方法及びプログラム |
Non-Patent Citations (1)
Title |
---|
OKUYAMA, TAKUYA ET AL.: "Binary optimization by momentum annealing", PHYSICAL REVIEW E, vol. 100, 10 July 2019 (2019-07-10), pages 012111 - 1, XP055806165, Retrieved from the Internet <URL:https://journals.aps.org/pre/pdf/10.1103/PhysRevE.100.012111> DOI: 10.1103/PhysRevE.100.012111 * |
Also Published As
Publication number | Publication date |
---|---|
JPWO2022044184A1 (ja) | 2022-03-03 |
AU2020465147B2 (en) | 2023-11-09 |
AU2020465147A1 (en) | 2023-03-23 |
JP7425210B2 (ja) | 2024-01-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6925546B1 (ja) | 演算システム、情報処理装置、および最適解探索処理方法 | |
EP3792841A1 (en) | Automated feature generation for machine learning application | |
JP7007520B2 (ja) | 情報処理装置、演算装置、及び情報処理方法 | |
Zheng et al. | A data-driven robust optimization method for the assembly job-shop scheduling problem under uncertainty | |
JPWO2019216277A1 (ja) | 情報処理装置、演算装置、及び情報処理方法 | |
JP7085158B2 (ja) | ニューラルネットワーク学習装置、ニューラルネットワーク学習方法、プログラム | |
Ballinas et al. | Hybrid quantum genetic algorithm with adaptive rotation angle for the 0-1 Knapsack problem in the IBM Qiskit simulator | |
Zhang et al. | Error-feedback stochastic modeling strategy for time series forecasting with convolutional neural networks | |
Zhang et al. | Graph attention reinforcement learning with flexible matching policies for multi-depot vehicle routing problems | |
Wauters et al. | Development of an adaptive infill criterion for constrained multi-objective asynchronous surrogate-based optimization | |
Phillipson et al. | Classification of hybrid quantum-classical computing | |
Horng et al. | Ordinal optimization of G/G/1/K polling systems with k-limited service discipline | |
WO2022044184A1 (ja) | 情報処理システムおよび最適解探索処理方法 | |
Lupo Pasini et al. | Fast and accurate predictions of total energy for solid solution alloys with graph convolutional neural networks | |
Yalaoui et al. | Heuristics for Optimization and Learning | |
JP2022083776A (ja) | 情報処理システム | |
Zhang et al. | Multi-objective cuckoo algorithm for mobile devices network architecture search | |
Simona et al. | Intelligent modeling method based on genetic algorithm for partner selection in virtual organizations | |
Binois et al. | A portfolio approach to massively parallel Bayesian optimization | |
Paliwal et al. | Stock prediction using neural networks and evolution algorithm | |
JP7357795B2 (ja) | 情報処理方法および情報処理システム | |
JP7398401B2 (ja) | 最適化方法、情報処理装置及びそれを用いたシステム | |
Liang et al. | Asynchronous evolution of deep neural network architectures | |
US20230267170A1 (en) | Information processing system, information processing method, and non-transitory computer-readable recording medium for information processing program | |
Wu et al. | A multiple criteria decision for trading capacity between two semiconductor fabs |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 20951437 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2022544983 Country of ref document: JP Kind code of ref document: A |
|
ENP | Entry into the national phase |
Ref document number: 2020465147 Country of ref document: AU Date of ref document: 20200827 Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 20951437 Country of ref document: EP Kind code of ref document: A1 |