CN114065121A - Calculation method and equipment for solving Itanium model - Google Patents
Calculation method and equipment for solving Itanium model Download PDFInfo
- Publication number
- CN114065121A CN114065121A CN202010747171.6A CN202010747171A CN114065121A CN 114065121 A CN114065121 A CN 114065121A CN 202010747171 A CN202010747171 A CN 202010747171A CN 114065121 A CN114065121 A CN 114065121A
- Authority
- CN
- China
- Prior art keywords
- spin
- signals
- complexity
- signal
- group
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F7/00—Methods or arrangements for processing data by operating upon the order or content of the data handled
- G06F7/38—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation
- G06F7/48—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation using non-contact-making devices, e.g. tube, solid state device; using unspecified devices
- G06F7/50—Adding; Subtracting
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Optimization (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Amplifiers (AREA)
Abstract
A computing method and equipment for solving an Esino model are used for meeting the requirements of solving the Esino model in different application scenes. In the method, a computing device determines an adjustment mode of a spin signal according to a problem matrix and a solving strategy of the problem matrix, wherein the problem matrix is used for indicating first data to be operated; after the first group of spin signals are acquired, the complexity of each spin signal in the first group of spin signals can be adjusted according to the adjustment mode of the spin signals, and a second group of spin signals are output; and then, performing operation by using the second group of spin signals and the problem matrix, and outputting a group of feedback signals, wherein the group of feedback signals are used for indicating an intermediate operation result of performing Yixin calculation on the first data. When solving the Escison model, the computing equipment can adjust the complexity of a group of spinning signals participating in operation so as to meet the solving strategy of the problem matrix, and the solving mode of the Escison module is suitable for different scenes.
Description
Technical Field
The present application relates to the field of communications technologies, and in particular, to a computing method and device for solving an ixing model.
Background
The Esinon model describes a complex system comprising a large number of spin nodes, each having spin states of both "+ 1" and "-1" values. In this system, there is an interaction between spin nodes, and the interaction between spin nodes can change the spin state of the spin nodes. Based on the interaction between the spin nodes, the Esin model gradually realizes the annealing process, namely the Hamiltonian of the system is gradually reduced until convergence. The combinatorial optimization problem can be changed into an Esinc model through conversion, and parameters in the combinatorial optimization problem are represented by utilizing the interaction between the spin nodes. And obtaining the optimal solution of the combined optimization problem by solving the Itanium model.
At present, when an Esin model is solved, a spin node in the Esin model is represented by a spin signal (such as an electric signal or an optical signal), a problem matrix is obtained by performing mathematical simulation on a combined optimization problem, the problem matrix can indicate the interaction between the spin signals, and the process of solving the Esin model is converted into multiplication between the spin signal and the problem matrix.
However, the complexity of the spin signal directly relates to the calculation speed and calculation accuracy of the solution of the ixing model, for example, the spin signal with high complexity is used to solve the ixing model, although the calculation accuracy is high and a more accurate calculation result can be obtained, the calculation process is more complex and the calculation speed is slower. And the spin signal with low complexity is utilized to solve the Esino model, although the operation process is simpler and the operation speed is higher, the operation precision is poorer and the accuracy of the calculation result is lower.
In some application scenarios, some concern about the operation speed of solving the Exin model, and some concern about the operation precision of solving the Exin model, the complexity of the spin signal in the current method for solving the Exin model is fixed, and only one requirement of the operation speed and the operation precision can be met, but the method cannot be applied to various application scenarios.
Disclosure of Invention
The application provides a calculation method and equipment for solving an Esino model, which are used for meeting the requirement of solving the Esino model in different application scenes.
In a first aspect, an embodiment of the present application provides a computing method for solving an ixing model, where the method is executed by a computing device, and in the method, the computing device may first determine an adjustment mode of a spin signal according to a problem matrix and a solution strategy of the problem matrix, where the problem matrix is used to indicate first data to be operated; after the first group of spin signals are acquired, the complexity of each spin signal in the first group of spin signals can be adjusted according to the adjustment mode of the spin signals, and a second group of spin signals are output; and then, performing operation by using the second group of spin signals and the problem matrix, and outputting a group of feedback signals, wherein the group of feedback signals are used for indicating an intermediate operation result of performing Yixin calculation on the first data.
After outputting the set of feedback signals, the computing device may further execute the above method again, that is, the computing device may determine a third set of spin signals according to the first set of spin signals and the set of feedback signals, and then adjust the complexity of each spin signal in the third set of spin signals according to the adjustment mode of the spin signals, and output a fourth set of spin signals; and performing operation by using the fourth group of spin signals and the problem matrix, outputting another group of feedback signals, and circulating the operation until the Escip model converges.
By the method, when the computing equipment solves the Itanium model, the complexity of a group of spinning signals participating in operation can be adjusted to meet the solving strategy of the problem matrix, so that the solving mode of the Itanium model is suitable for different scenes.
In a possible implementation manner, the solving strategy can be to improve the operation speed and the operation precision.
By the method, the method for solving the Itanium module can be suitable for application scenes with requirements on operation speed or operation precision.
In one possible implementation, when determining the adjustment mode of the spin signals according to the problem matrix and the solution strategy of the problem matrix, the computing device may first determine the number of spin signals in the first set of spin signals according to the problem matrix; and then determining the adjustment mode of the spin signals according to the number of the spin signals and the solution strategy.
By the method, the computing equipment can determine the number of the spin signals in the first spin signal which needs to participate in operation subsequently through the problem matrix, and then determine a proper adjustment mode according to the data and the solving strategy.
In a possible implementation manner, the number of the spin signals is greater than a first threshold, when the computing device determines the adjustment mode of the spin signals according to the number of the spin signals and the solution strategy, if the solution strategy is to increase the operation speed, the computing device determines the adjustment mode of the spin signals to reduce the complexity of each spin signal in the first set of spin signals to a first complexity; if the solving strategy is to improve the operation precision, the computing device determines the adjusting mode of the spin signals to reduce the complexity of each spin signal in the first group of spin signals to a second complexity, wherein the first complexity is smaller than the second complexity.
By the method, under the condition that the number of the spin signals in the first group of spin signals is large, the complexity of the spin signals in the first group of spin signals can be adjusted to be the spin signals with the corresponding complexity corresponding to different solving strategies, so that the solving strategy of the problem matrix is met.
In a possible implementation manner, the number of the spin signals is not greater than a first threshold, when the computing device determines the adjustment mode of the spin signals according to the number of the spin signals and the solving strategy, if the solving strategy is to increase the operation speed, the adjustment mode of the spin signals is determined to be an adaptive adjustment mode, and the adaptive adjustment mode is to adjust the complexity of each spin signal in the first set of spin signals according to the spin signal in the first set of spin signals; and if the solving strategy is to improve the operation precision, determining the adjustment mode of the spin signals to reduce the complexity of each spin signal in the first group of spin signals to a second complexity.
By the method, under the condition that the number of the spin signals in the first group of spin signals is small, the complexity of the spin signals in the first group of spin signals can be adjusted to the spin signals with the corresponding complexity according to different solving strategies, so that the solving strategy of the problem matrix is met.
In one possible implementation, if the adjustment mode of the spin signal is the adaptive adjustment mode, when the computing device adjusts the complexity of each spin signal in the received first set of spin signals according to the adjustment mode of the spin signal, the computing device first determines data of the spin signal in the first set of spin signals within a preset range of signal values, and if the number of spin signals in the first set of spin signals within the preset range of signal values is greater than the second threshold, reduces the complexity of each spin signal in the first set of spin signals to the first complexity. And if the number of the spin signals in the preset signal value range in the first group of spin signals is not more than a second threshold value, reducing the complexity of each spin signal in the first group of spin signals to a second complexity, wherein the first complexity is less than the second complexity.
By the method, the computing equipment can adaptively adjust the complexity of the first group of spin signals according to the distribution condition of the signal values of the spin signals in the first group of spin signals, so that the method for solving the Esino model is more flexible and efficient.
In a second aspect, an embodiment of the present invention provides a computing device for solving an ixing model, where the computing device has a function of implementing a behavior of the computing device in the above method example. The functions may be implemented by hardware, or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above-described functions. The computing device includes a determination unit, an adjustment unit, and an arithmetic unit.
The determining unit is used for determining an adjustment mode of the spin signal according to a problem matrix and a solving strategy of the problem matrix, wherein the problem matrix is used for indicating first data to be operated;
the adjusting unit is used for adjusting the complexity of each spin signal in the received first group of spin signals according to the adjusting mode of the spin signals and outputting a second group of spin signals;
and the operation unit is used for performing operation by utilizing the second group of spin signals and the problem matrix and outputting a group of feedback signals, and the group of feedback signals are used for indicating an intermediate operation result of performing Yixin calculation on the first data.
In one possible embodiment, the solving strategy is to increase the operation speed or increase the operation precision.
In a possible implementation manner, when determining the adjustment mode of the spin signals according to the problem matrix and the solution strategy of the problem matrix, the determining unit may first determine the number of spin signals in the first set of spin signals according to the problem matrix; and then, determining the adjustment mode of the spin signals according to the number of the spin signals and the solution strategy.
In a possible implementation manner, the number of the spin signals is greater than a first threshold, and when the determining unit determines the adjustment mode of the spin signals according to the number of the spin signals and the solving strategy, if the solving strategy is to increase the operation speed, the determining unit determines the adjustment mode of the spin signals to reduce the complexity of each spin signal in the first set of spin signals to a first complexity; and if the solving strategy is to improve the operation precision, determining the adjustment mode of the spin signals to reduce the complexity of each spin signal in the first group of spin signals to a second complexity, wherein the first complexity is less than the second complexity.
In a possible implementation manner, the number of the spin signals is not greater than the first threshold, and when determining the adjustment mode of the spin signals according to the number of the spin signals and the solving strategy, if the solving strategy is to increase the operation speed, the determination unit determines that the adjustment mode of the spin signals is an adaptive adjustment mode, where the adaptive adjustment mode is to adjust the complexity of each spin signal in the first set of spin signals according to the spin signal in the first set of spin signals. And if the solving strategy is to improve the operation precision, determining the adjustment mode of the spin signals to reduce the complexity of each spin signal in the first group of spin signals to a second complexity.
In a possible embodiment, if the adjustment mode of the spin signal is the adaptive adjustment mode, the adjusting unit determines the number of spin signals in the first set of spin signals within a preset range of signal values when adjusting the complexity of each spin signal in the received first set of spin signals according to the adjustment mode of the spin signal, and reduces the complexity of each spin signal in the first set of spin signals to the first complexity if the number is greater than the second threshold. And reducing the complexity of each spin signal in the first group of spin signals to a second complexity under the condition that the number of spin signals in the preset range of the signal value in the first group of spin signals is not more than a second threshold value, wherein the first complexity is less than the second complexity.
In a third aspect, the present application further provides a computer-readable storage medium having stored therein instructions, which, when executed on a computer, cause the computer to perform the method of the first aspect and any possible implementation manner of the first aspect.
In a fourth aspect, the present application also provides a computer program product comprising instructions which, when run on a computer, cause the computer to perform the method of the first aspect described above and any possible implementation of the first aspect.
In a fifth aspect, the present application further provides a computer chip, where the chip is connected to a memory, and the chip is configured to read and execute a software program stored in the memory, and perform the method in the first aspect and any possible implementation manner of the first aspect.
Drawings
FIG. 1 is a schematic diagram of a method for solving an Esin model according to the present disclosure;
FIG. 2 is a schematic diagram of a structure for implementing a multiplication operation of a spin signal and a problem matrix according to the present application;
FIG. 3A is a schematic diagram of a multiplication of a spin signal with a problem matrix according to the present application;
FIG. 3B is a schematic diagram of a multiplication of a spin signal with a problem matrix according to the present application;
fig. 4 is a schematic structural diagram of a computing device provided in the present application.
Detailed Description
The embodiment of the application provides a calculation method for solving an Esin model, which is executed by a calculation device, wherein the calculation device can solve the Esin model, when the calculation device solves the Esin model, an adjustment mode of a spin signal can be determined according to a problem matrix and a solving strategy of the problem matrix, and then the complexity of each spin signal in a first group of received spin signals is adjusted according to the adjustment mode of the spin signal, and a second group of spin signals is output. After obtaining the second set of spin signals, the computing device may perform an operation using the second set of spin signals and the problem matrix, outputting a set of feedback signals indicative of a result of an intermediate operation of the first data to perform an yixin computation. The computing device may further continue to perform an operation using the set of feedback signals, for example, the computing device may generate a third set of spin signals according to the set of feedback signals and the first set of spin signals, adjust the complexity of each spin signal in the third set of spin signals according to the adjustment mode of the spin signals, output a fourth set of spin signals, perform an operation using the fourth set of spin signals and the problem matrix, and output another set of feedback signals, which is repeated until the ixing model converges, and when the ixing model converges, the generated set of spin signals is an operation result of the first data.
In the embodiment of the application, when the Esin model is solved, the complexity of a group of spin signals participating in operation can be adjusted to meet the solving strategy of the problem matrix, so that the solving mode of the Esin model can be suitable for different scenes.
The calculation method for solving the ixing model provided by the present application is further described below with reference to the accompanying drawings, and with reference to fig. 1, the method includes:
step 101: the calculation equipment firstly determines an adjustment mode of the spin signal according to a problem matrix and a solving strategy of the problem matrix, wherein the problem matrix is used for indicating first data to be operated.
After the problem matrix is obtained, the problem matrix is analyzed by the computing equipment, the number of spin signals in a first group of spin signals which are subsequently operated with the problem matrix can be determined by the computing equipment according to the matrix scale of the problem matrix, namely the size of the problem matrix, and the number of spin signals in the first group of spin signals is equal to the number of rows or columns of the problem matrix.
After determining the number of spin signals in the first set of spin signals, the computing device may determine an adjustment mode for the spin signals in conjunction with a solution strategy for the problem matrix.
In the embodiment of the present application, the solution strategy can be divided into two types, one type is to increase the operation speed, that is, to ensure the timeliness of solving the ixing model, so that the operation result can be obtained in a short time. The other is to improve the operation precision, that is, the accuracy of the operation result needs to be ensured.
The determining method of the solving strategy is not limited in the embodiment of the application, the solving strategy is configured in advance, or the solving strategy can be determined by the computing device according to the operation scene, for example, the current operation scene explicitly indicates the time for solving the ixing model, if the time is greater than the time threshold, the computing device determines that the solving strategy is the lifting operation precision, and if the time is not greater than the time threshold, the computing device determines that the solving strategy is the lifting operation speed; for another example, the current operation scenario explicitly indicates a time for solving the izod model, the computing device determines that the solving strategy is to increase the operation speed, the current operation scenario does not indicate a time for solving the izod model, and the computing device determines that the solving strategy is to increase the operation precision.
The following describes a manner in which the computing device determines the adjustment mode of the spin signal according to the number of spin signals and the solution strategy:
(1) the number of spin signals is greater than a first threshold.
The solving strategy is to increase the operation speed, and the spin signals are adjusted in a mode of reducing the complexity of each spin signal in the first group of spin signals to a first complexity.
The solving strategy is to improve the operation precision, and the spin signal is adjusted in a mode of keeping the complexity of each spin signal in the first group of spin signals or reducing the complexity of each spin signal in the first group of spin signals to a second complexity. The first complexity is less than the second complexity. The embodiment of the application does not limit the specific values of the first threshold, the first complexity and the second complexity, and can be set according to a specific scene.
In the embodiment of the present application, the complexity of the spin signal is used to indicate the accuracy of the spin signal, and the bit number of the spin signal may be used as the complexity of the spin signal to indicate the accuracy of the spin signal. Other parameters, such as the difference between the bit number of the spin signal and a specific value, may also be used as the complexity of the spin signal, and the embodiment of the present application does not limit the representation manner of the complexity of the spin signal, and any manner capable of indicating the accuracy of the spin signal is applicable to the embodiment of the present application.
The complexity of each spin signal in the first set of spin signals serves as an initial complexity, which may be a higher complexity. For example, the number of bits of any one spin signal in the first set of spin signals is 16 bits or a higher number of bits than 16 bits. The bit number of any spin signal in the first set of spin signals is 16 bits, which means that a 16-bit array is required to characterize one spin signal in the first set of spin signals.
When the number of the spin signals is greater than the first threshold, it indicates that the number of the spin signals is greater, in order to increase the operation speed, the spin signals with lower complexity may be used to perform an operation with the problem matrix, and before performing an operation with the problem matrix, the complexity of each spin signal in the first set of spin signals needs to be reduced to a greater extent.
In this case, the spin signals are adjusted in a manner to reduce the complexity of each spin signal in the first set of spin signals to a first complexity that is lower than the initial complexity. For example, the first complexity may be 1 bit, i.e., a 1 bit array is used to characterize a spin signal.
If the operation is performed with the problem matrix in order to improve the operation accuracy, the spin signal with higher complexity is usually used for performing the operation with the problem matrix, and before performing the operation with the problem matrix, the complexity of each spin signal in the first set of spin signals may be maintained to ensure that the complexity of the spin signal subsequently participating in the operation is higher, and the complexity of each spin signal in the first set of spin signals may also be reduced to a lesser extent.
If the complexity of each spin signal in the first set of spin signals is maintained or reduced to a lesser degree, the spin signal adjustment mode may be to reduce the complexity of each spin signal in the first set of spin signals to a second complexity that is equal to or less than the initial complexity and higher than the first complexity. For example, the second complexity may be 2 bits, i.e. a 2-bit array is used to represent a spin signal.
(2) The number of spin signals is not greater than a first threshold.
The solving strategy is to improve the operation speed, the adjusting mode of the spin signals is a self-adaptive adjusting mode, and the self-adaptive adjusting mode is to adjust the complexity of each spin signal in the first group of spin signals according to the spin signals in the first group of spin signals.
The solving strategy is to improve the operation precision, and the spin signal is adjusted in a mode of reducing the complexity of each spin signal in the first group of spin signals to a third complexity or keeping the complexity of each spin signal in the first group of spin signals. The third complexity is less than the initial complexity, and the third complexity may be equal to or greater than the second complexity.
When the number of the spin signals is not greater than the first threshold, it indicates that the number of the spin signals is small, and the time consumed for the operation of the problem matrix of the spin signals with small number is relatively small.
In this case, the adjustment mode of the spin signals is an adaptive mode, and the complexity of adjusting each spin signal in the first set of spin signals according to the spin signal in the first set of spin signals is required. The adaptive mode will be explained in step 102.
If the operation is performed with the problem matrix in order to improve the operation accuracy, the spin signal with higher complexity is usually used for performing the operation with the problem matrix, and before performing the operation with the problem matrix, the complexity of each spin signal in the first set of spin signals may be maintained to ensure that the complexity of the spin signal subsequently participating in the operation is higher, and the complexity of each spin signal in the first set of spin signals may also be reduced to a lesser extent.
If the complexity of each spin signal in the first set of spin signals is maintained or reduced to a lesser degree, the spin signal adjustment mode may be to reduce the complexity of each spin signal in the first set of spin signals to a third complexity, which is equal to or less than the initial complexity and higher than the first complexity. For example, the third complexity may be 4 bits or 2 bits, i.e. a 4-bit or 2-bit array is used to represent a spin signal.
After determining the adjustment mode for the spin signals, the computing device may perform step 102 to adjust the complexity of each spin signal in the first set of spin signals.
Step 102: the computing device adjusts the complexity of each spin signal in the received first set of spin signals according to the adjustment mode of the spin signals and outputs a second set of spin signals.
The following describes a manner in which the computing device adjusts the complexity of each spin signal in the received first set of spin signals and outputs the second set of spin signals in different spin signal adjustment modes:
1. the spin signals are adjusted in a manner that reduces the complexity of each spin signal in the first set of spin signals to a first complexity.
The embodiment of the present application does not limit the way in which the computing device reduces the complexity of each spin signal in the first set of spin signals to obtain the second set of spin signals, for example, the computing device may intercept each spin signal in the first set of spin signals, and only reserve an array of some bits in each spin signal in the first set of spin signals as the second set of spin signals.
The following describes a manner in which a computing device provided by an embodiment of the present application obtains a second set of spin signals by reducing the complexity of each spin signal in the first set of spin signals.
Taking the bit number of the spin signal as the complexity of the spin signal, wherein the first complexity is N bits, and the computing equipment adds half of the value range of the spin signal which can be represented by the first complexity to the spin signal in the first group of spin signals, namely half 2 of the maximum value of the spin signal value under the first complexityN2-1/2, then in the Nbit range (0-2)N-1)Performing saturation truncation, i.e. greater than 2NThe value of-1 is taken to be 2N-1, taking the value less than 0 as 0, then randomly rounding up and rounding down the truncated data to obtain spin signals of a first complexity, i.e. spin signals in the second set of spin signals, using TjRepresenting spin signals in the second set of spin signals, σjIs the spin signal in the first set of spin signals. Spin signal T in the second set of spin signalsjWith the spin signal σ in the first set of spin signalsjThe relationship of (a) to (b) is as follows:
Tj＝[σj]+2N/2-1/2
it should be noted that, after the computing device adds the spin signal in the first set of spin signals to half of the range of values of the spin signal that can be represented by the first complexity, the generated data may also be adjusted, for example, an offset is added to the generated data, where the offset may be an empirical value or a value determined according to the performance of the computing device.
2. The spin signals are adjusted in a mode that reduces the complexity of each spin signal in the first set of spin signals to a second complexity or a third complexity.
For the case where the spin signal adjustment mode is to reduce the complexity of each spin signal in the first set of spin signals to the second complexity or the third complexity, the computing device executes step 102 in a similar manner to the case where the spin signal adjustment mode is to reduce the complexity of each spin signal in the first set of spin signals to the first complexity, except that the complexity of each spin signal in the first set of spin signals is reduced to a different degree, the foregoing description may be specifically referred to.
3. The adjustment mode of the spin signal is an adaptive mode.
In the adaptive mode, the computing device needs to analyze the states of the spin signals in the first set of spin signals and adjust the complexity of each spin signal in the set of spin signals based on the spin signals in the first set of spin signals.
For example, the computing device may determine the number of spin signals in a preset range of signal values in the first set of spin signals, where the signal value of a spin signal may be understood as a specific value of a spin signal, and if the spin signal is an optical signal, the signal value of the spin signal may be an amplitude of the optical signal; if the spin signal is an electrical signal, the signal value of the spin signal may be a voltage value of the electrical signal. The embodiment of the present application does not limit the specific range indicated by the signal value preset range, and the signal value preset range may be determined according to an application scenario or may be an empirical value.
If the number of spin signals in the first group of spin signals within the preset range of signal values is larger than the second threshold, the preset range of signal values is part or all of the following: the signal value is smaller than the first preset value or larger than the second preset value.
The number of the spin signals in the preset signal value range in the first group of spin signals is large, the spin signals in the later stage of solving the Esino model can be low in complexity, and the operation speed can be improved. The computing device may reduce the complexity of each spin signal in the first set of spin signals to a first complexity. The way in which the computing device can reduce the complexity of each spin signal in the first set of spin signals to the first complexity can be referred to the foregoing description, and is not described here in detail.
If the number of the spin signals in the signal value preset range in the first group of spin signals is not greater than the second threshold, it is indicated that the number of the spin signals in the signal value preset range in the first group of spin signals is smaller, and in the initial stage of solving the Esino model, the spin signals with higher complexity can be adopted, and a better operation result can be obtained. The computing device may reduce the complexity of each spin signal in the first set of spin signals to a second complexity. The way in which the computing device can reduce the complexity of each spin signal in the first set of spin signals to the second complexity is similar to the way in which the computing device can reduce the complexity of each spin signal in the first set of spin signals to the first complexity, which may be referred to in detail in the foregoing description and is not described here again.
After obtaining the second set of spin signals, the computing device may perform step 103.
Step 103: the computing device performs an operation using the second set of spin signals and the problem matrix and outputs a set of feedback signals, wherein the set of feedback signals is an intermediate operation result of the first data.
The operation of the second group of spin signals and the problem matrix is multiplication operation, and the result of multiplication of the second group of spin signals and the problem matrix is a group of feedback signals.
The embodiment of the present application does not limit the way of obtaining a set of feedback signals by multiplying the second set of spin signals by the problem matrix, and the second set of spin signals is taken as { T }1，T2，…，TjThe problem matrix is
For example, a way of performing a multiplication operation on the second set of spin signals and the problem matrix provided by the embodiment of the present application is described:
the matrix multiplication between the second set of spin signals and the problem matrix is:
wherein m isiIs the number of 0 elements in the ith row in the problem matrix.
Referring to fig. 2, a plurality of cascaded adders may be included in the computing device, each adder being connected to a plurality of multipliers, one for implementing multiplication of one spin signal of the second set of spin signals by one element of a column in the problem matrix. Each adder is connected with a plurality of multipliers to obtain the operation result of each multiplier, and the result of each multiplier is summed.
The number of multipliers connected to each adder and the connection mode of each adder and each multiplier are not limited in the embodiments of the present application. In fig. 2, each adder is connected to four multipliers, of which three multipliers are connected to a secondary adder which can sum the results of the three multipliers. The secondary adder and the other multiplier are connected with the adder through a selector (MUX), and the selector can select to obtain the calculation result of the secondary adder or obtain the calculation result of the other multiplier under the control of the calculation device.
In fig. 2, a secondary selector is arranged after every 3 cascaded adders, and the secondary selector can change the connection state of the previous adder and the next adder of the secondary selector under the control of a computing device.
In the embodiment of the present application, based on the structure as shown in fig. 2, the computing device may adopt the following two operation modes in the process of implementing the operation of the problem matrix and the second set of spin signals.
In the first mode, a second set of spin signals with higher complexity (such as the second complexity) is used to operate with the problem matrix.
With the second set of spin signals as { T1，T2，…，T9The problem matrix is
For example, the parameters input to the multiplier are shown in FIG. 3A. The computing device controls the selector to obtain the computation result of the secondary adder. The first row of three cascaded adders in fig. 3A obtains the multiplications of the second set of spin signals with each element in the first column of the problem matrix, the computing device controls the secondary selector such that the previous adder and the next adder of the secondary selector are disconnected, and the last adder of the first row of three cascaded adders can output the multiplications of the second set of spin signals with each element in the first column of the problem matrix. The three cascaded adders in the second row obtain the multiplication sum of the second group of spin signals and each element in the second column of the problem matrix, and the last adder of the three cascaded adders in the second row can output the multiplication sum of the second group of spin signals and each element in the first column of the problem matrix; third row threeThe cascaded adders obtain the multiplication sum of the second group of spin signals and each element in the third column of the problem matrix, and the last adder in the third row of the three cascaded adders can output the multiplication sum of the second group of spin signals and each element in the first column of the problem matrix. The way in which the second set of spin signals is multiplied by the first three columns of the problem matrix is shown in figure 3A, and the way in which the second set of spin signals is multiplied by the other columns of the problem matrix can be seen in the way shown in figure 3A, with the difference that the parameters of the multiplier inputs are different.
In the second mode, a second set of spin signals with lower complexity (such as the first complexity) is used to operate with the problem matrix.
The computing device may change the connection of the cascaded adders so that the plurality of cascaded adders becomes a plurality of groups of adders, each group of adders includes a part of the adders in the plurality of adders, and the adders in each group of adders are connected in a cascaded manner.
Still using the second set of spin signals as { T1，T2，…，T9The problem matrix is
For example, the parameters input to the multiplier are shown in FIG. 3B. The computing device controls the selector to obtain the result of the computation of the other multiplier, i.e. the multiplier not connected to the secondary adder. In fig. 3B, the first row of three cascaded adders obtains the multiplication sum of the first three spin signals of the second set of spin signals and the first three elements of the first column of the problem matrix, the second row of three cascaded adders obtains the multiplication sum of the middle three spin signals of the second set of spin signals and the first three elements of the first column of the problem matrix, and the third row of three cascaded adders obtains the multiplication sum of the last three spin signals of the second set of spin signals and the last three elements of the first column of the problem matrix. The computing device controls the secondary selector so that the preceding adder and the following adder of the secondary selector are connected, and the multiplication sum of the second set of spin signals and each element of the first column of the problem matrix can be obtained by nine cascaded adders, only the implementation of the second set of spin signals being shown in fig. 3BThe way in which the spin signals are multiplied by the first column of the problem matrix and the way in which the second set of spin signals are multiplied by the other columns of the problem matrix can be seen in the way shown in fig. 3B, with the difference that the parameters input to the multipliers are different.
In the foregoing, it is also mentioned that the computing device may continue to perform the operation after obtaining a set of feedback signals, and the computing device may sum the set of feedback signals and the first set of spin signals to obtain a third set of spin signals, adjust the complexity of each spin signal in the third set of spin signals according to an adjustment mode of the spin signals, output a fourth set of spin signals, perform matrix multiplication on the fourth set of spin signals and the problem matrix, and output another set of feedback signals, and loop until the ixing model converges, and when the ixing model converges, the generated set of spin signals is an operation result of the first data.
There are various ways to determine whether the ixing model converges, for example, a threshold of the number of operations may be set, one operation is performed every time a group of feedback signals is obtained, when the number of operations reaches the threshold, the ixing model is considered to converge, and the last operation obtains a group of spin signals, which is the operation result of the first data. Namely, each time the operation is executed, whether the total number of times of the operation reaches the threshold value is determined, if the total number of times of the operation does not reach the threshold value, a group of feedback signals obtained by the operation is summed with the first group of spin signals again to obtain a new group of spin signals, the complexity of the group of spin signals is continuously adjusted, and then the spin signals with the adjusted complexity and the problem matrix are used for operation to obtain a new group of feedback signals. And if the threshold value is reached, determining that a group of spin signals are obtained in the operation process.
For another example, the computing device may calculate a hamiltonian quantity of the evans model, the hamiltonian quantity being determined from a set of feedback signals obtained during the operation, and if the hamiltonian quantity is not decreased, indicating that the evans model has converged, stop the operation, and obtain a set of spin signals generated during the last operation.
Based on the same inventive concept as the method embodiment, an embodiment of the present application further provides a computing device, configured to execute the method in the method embodiment shown in fig. 1, where relevant features may refer to the method embodiment, and are not described herein again, referring to fig. 4, where the computing device 400 includes a determining unit 401, an adjusting unit 402, and an arithmetic unit 403, and optionally, may further include a spin generating unit 404.
The determining unit 401 determines an adjustment mode of the spin signal according to a problem matrix and a solution strategy of the problem matrix, where the problem matrix is used to indicate first data to be operated; the determination unit 401 may be a circuit formed by a Complementary Metal Oxide Semiconductor (CMOS), such as a central processing unit (cpu), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or a Complex Programmable Logic Device (CPLD), or may be another device or a unit formed by a plurality of devices.
A spin generation unit 404 for generating a first set of spin signals. The spin generating unit may be a spin array or a superconducting line that can generate an optical pulse, and if the type of the first spin signal is a polaron, the spin generating unit 404 may be a unit that can generate a polaron.
And an adjusting unit 402, configured to adjust complexity of each spin signal in the received first set of spin signals according to an adjustment mode of the spin signal, and output a second set of spin signals. The adjusting module may also be a circuit formed by CMOS, such as a central processor, ASIC, FPGA or CPLD, or may also be another device or a unit formed by a plurality of devices.
And an operation unit 403 for performing an operation using the second set of spin signals and the problem matrix, and outputting a set of feedback signals indicating an intermediate operation result for performing an inching calculation on the first data. The present embodiment is not limited to the configuration of the arithmetic unit, and the configuration of the arithmetic unit is as shown in fig. 2.
As a possible implementation manner, the solving strategy may be to increase the operation speed, and may also be to increase the operation precision.
As a possible implementation manner, when determining the adjustment mode of the spin signals according to the problem matrix and the solution strategy of the problem matrix, the determining unit 401 may first determine the number of spin signals in the first set of spin signals according to the problem matrix; and then, determining the adjustment mode of the spin signals according to the number of the spin signals and the solution strategy.
As a possible implementation manner, when the number of spin signals is greater than the first threshold, the determining unit 401 determines, according to the number of spin signals and the solving strategy, an adjustment mode of the spin signals, if the solving strategy is to increase the operation speed, the adjustment mode of the spin signals is determined to reduce the complexity of each spin signal in the first set of spin signals to the first complexity; and if the solving strategy is to improve the operation precision, determining the adjustment mode of the spin signals to reduce the complexity of each spin signal in the first group of spin signals to a second complexity, wherein the first complexity is less than the second complexity.
As a possible implementation manner, the number of the spin signals is not greater than the first threshold, when the determining unit 401 determines the adjustment mode of the spin signals according to the number of the spin signals and the solving strategy, if the solving strategy is to increase the operation speed, the adjustment mode of the spin signals is determined to be an adaptive adjustment mode, and the adaptive adjustment mode is to adjust the complexity of each spin signal in the first set of spin signals according to the spin signal in the first set of spin signals; and if the solving strategy is to improve the operation precision, determining the adjustment mode of the spin signals to reduce the complexity of each spin signal in the first group of spin signals to a second complexity.
As a possible implementation manner, if the adjustment mode of the spin signal is the adaptive adjustment mode, the adjusting unit 402 may first determine the number of spin signals in the first set of spin signals within a preset range of signal values when adjusting the complexity of each spin signal in the received first set of spin signals according to the adjustment mode of the spin signal; reducing the complexity of each spin signal in the first group of spin signals to a first complexity under the condition that the number of spin signals in the first group of spin signals within the preset range of the signal value is larger than a second threshold value; and reducing the complexity of each spin signal in the first group of spin signals to a second complexity under the condition that the number of spin signals in the preset range of the signal value in the first group of spin signals is not more than a second threshold value, wherein the first complexity is less than the second complexity.
It should be noted that the division of the unit in the embodiment of the present application is schematic, and is only a logic function division, and there may be another division manner in actual implementation. The functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The above embodiments may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented in software, the above-described embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded or executed on a computer, cause the flow or functions according to embodiments of the invention to occur, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, which may be any available medium that can be accessed by a computer. The usable medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium may be a Solid State Drive (SSD).
It will be apparent to those skilled in the art that embodiments of the present application may be provided as a method, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present application without departing from the scope of the application. Thus, if such modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is intended to include such modifications and variations as well.
Claims (12)
1. A computational method for solving an ixing model, the method comprising:
determining an adjustment mode of a spin signal according to a problem matrix and a solving strategy of the problem matrix, wherein the problem matrix is used for indicating first data to be operated;
adjusting the complexity of each spin signal in the received first group of spin signals according to the adjustment mode of the spin signals, and outputting a second group of spin signals;
and performing operation by using the second group of spin signals and the problem matrix, and outputting a group of feedback signals, wherein the group of feedback signals are used for indicating an intermediate operation result of performing Yixin calculation on the first data.
2. The method of claim 1, wherein the solution strategy is to increase operational speed or increase operational accuracy.
3. The method of claim 1 or 2, wherein determining the adjustment mode of the spin signal according to the problem matrix and a solution strategy for the problem matrix comprises:
determining a number of spin signals in the first set of spin signals from the problem matrix;
and determining an adjustment mode of the spin signals according to the number of the spin signals and the solving strategy.
4. The method of claim 3, wherein the number of spin signals is greater than a first threshold, and wherein determining the adjustment mode for the spin signals based on the number of spin signals and the solution strategy comprises:
if the solving strategy is to increase the operation speed, determining the adjustment mode of the spin signals to reduce the complexity of each spin signal in the first group of spin signals to a first complexity; or
And if the solving strategy is to improve the operation precision, determining the adjustment mode of the spin signals to reduce the complexity of each spin signal in the first group of spin signals to a second complexity, wherein the first complexity is smaller than the second complexity.
5. The method of claim 3, wherein the number of spin signals is not greater than a first threshold, and wherein determining the adjustment mode for the spin signals based on the number of spin signals and the solution strategy comprises:
if the solving strategy is to increase the operation speed, determining that the adjustment mode of the spin signals is a self-adaptive adjustment mode, wherein the self-adaptive adjustment mode is to adjust the complexity of each spin signal in the first group of spin signals according to the spin signal in the first group of spin signals; or
And if the solving strategy is to improve the operation precision, determining the adjustment mode of the spin signals to reduce the complexity of each spin signal in the first group of spin signals to a second complexity.
6. The method of claim 5, wherein if the spin signal adjustment mode is an adaptive adjustment mode, the adjusting the complexity of each spin signal in the received first set of spin signals according to the spin signal adjustment mode comprises:
reducing the complexity of each spin signal in the first group of spin signals to a first complexity under the condition that the number of spin signals in the first group of spin signals within a preset range of signal values is larger than a second threshold value; or
And reducing the complexity of each spin signal in the first group of spin signals to a second complexity under the condition that the number of spin signals in the first group of spin signals within the preset signal value range is not greater than a second threshold value, wherein the first complexity is less than the second complexity.
7. A computing device to solve an ixing model, the device comprising:
the device comprises a determining unit, a calculating unit and a calculating unit, wherein the determining unit is used for determining an adjusting mode of a spin signal according to a problem matrix and a solving strategy of the problem matrix, and the problem matrix is used for indicating first data to be operated;
the adjusting unit is used for adjusting the complexity of each spin signal in the received first group of spin signals according to the adjusting mode of the spin signals and outputting a second group of spin signals;
and the operation unit is used for performing operation by utilizing the second group of spin signals and the problem matrix and outputting a group of feedback signals, and the group of feedback signals are used for indicating an intermediate operation result of executing the Yixin calculation on the first data.
8. The computing device of claim 7, wherein the solution policy is to increase operational speed or increase operational precision.
9. The computing device according to claim 7 or 8, wherein the determining unit, when determining the adjustment mode of the spin signal according to the problem matrix and the solution strategy of the problem matrix, is specifically configured to:
determining a number of spin signals in the first set of spin signals from the problem matrix;
and determining an adjustment mode of the spin signals according to the number of the spin signals and the solving strategy.
10. The computing device of claim 9, wherein the number of spin signals is greater than a first threshold, the determination unit, when determining the adjustment mode of the spin signals based on the number of spin signals and the solution strategy, being specifically configured to:
if the solving strategy is to increase the operation speed, determining the adjustment mode of the spin signals to reduce the complexity of each spin signal in the first group of spin signals to a first complexity; or
And if the solving strategy is to improve the operation precision, determining the adjustment mode of the spin signals to reduce the complexity of each spin signal in the first group of spin signals to a second complexity, wherein the first complexity is smaller than the second complexity.
11. The computing device of claim 9, wherein the number of spin signals is not greater than a first threshold, the determination unit, when determining the adjustment mode of the spin signals based on the number of spin signals and the solution strategy, being specifically configured to:
if the solving strategy is to increase the operation speed, determining that the adjustment mode of the spin signals is a self-adaptive adjustment mode, wherein the self-adaptive adjustment mode is to adjust the complexity of each spin signal in the first group of spin signals according to the spin signal in the first group of spin signals; or
And if the solving strategy is to improve the operation precision, determining the adjustment mode of the spin signals to reduce the complexity of each spin signal in the first group of spin signals to a second complexity.
12. The computing device of claim 11, wherein if the spin signal adjustment mode is an adaptive adjustment mode, the adjustment unit, when adjusting the complexity of each spin signal in the received first set of spin signals according to the spin signal adjustment mode, is specifically configured to:
reducing the complexity of each spin signal in the first group of spin signals to a first complexity under the condition that the number of spin signals in the first group of spin signals within a preset range of signal values is larger than a second threshold value; or
And reducing the complexity of each spin signal in the first group of spin signals to a second complexity under the condition that the number of spin signals in the first group of spin signals within the preset signal value range is not greater than a second threshold value, wherein the first complexity is less than the second complexity.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010747171.6A CN114065121A (en) | 2020-07-29 | 2020-07-29 | Calculation method and equipment for solving Itanium model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010747171.6A CN114065121A (en) | 2020-07-29 | 2020-07-29 | Calculation method and equipment for solving Itanium model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114065121A true CN114065121A (en) | 2022-02-18 |
Family
ID=80226895
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010747171.6A Pending CN114065121A (en) | 2020-07-29 | 2020-07-29 | Calculation method and equipment for solving Itanium model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114065121A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115858999A (en) * | 2023-02-07 | 2023-03-28 | 华南理工大学 | Combined optimization problem processing circuit based on improved simulated annealing algorithm |
| CN115907005A (en) * | 2023-01-05 | 2023-04-04 | 华南理工大学 | Large-scale full-connection Esino model annealing processing circuit based on network on chip |
-
2020
- 2020-07-29 CN CN202010747171.6A patent/CN114065121A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115907005A (en) * | 2023-01-05 | 2023-04-04 | 华南理工大学 | Large-scale full-connection Esino model annealing processing circuit based on network on chip |
| CN115858999A (en) * | 2023-02-07 | 2023-03-28 | 华南理工大学 | Combined optimization problem processing circuit based on improved simulated annealing algorithm |
| CN115858999B (en) * | 2023-02-07 | 2023-04-25 | 华南理工大学 | Combined optimization problem processing circuit based on improved simulated annealing algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114065121A (en) | Calculation method and equipment for solving Itanium model | |
| JP7110278B2 (en) | Method, apparatus, apparatus, computer readable storage medium and computer program for load balancing | |
| EP3561739A1 (en) | Data accelerated processing system | |
| CN110109646A (en) | Data processing method, device and adder and multiplier and storage medium | |
| WO2022135440A1 (en) | Llc resonant circuit control method and control apparatus, and terminal device | |
| Shu et al. | High energy efficiency FPGA-based accelerator for convolutional neural networks using weight combination | |
| CN112668639A (en) | Model training method and device, server and storage medium | |
| CN114021708B (en) | Data processing method, device and system, electronic equipment and storage medium | |
| CN115408061B (en) | Hardware acceleration method, device, chip and storage medium for complex matrix operation | |
| US20220358183A1 (en) | Matrix multiplier and operation method thereof | |
| CN112799637B (en) | High-throughput modular inverse computation method and system in parallel environment | |
| CN110659731B (en) | Neural network training method and device | |
| JP7036224B2 (en) | Arithmetic processing unit and control method of arithmetic processing unit | |
| CN116644813B (en) | Method and device for determining optimal combination scheme by utilizing quantum circuit | |
| CN111061675A (en) | Hardware implementation method of system transfer function identification algorithm, computer equipment and readable storage medium for running method | |
| CN113298843B (en) | Data quantization processing method, device, electronic equipment and storage medium | |
| KR20200095951A (en) | GPU-based AI system using channel-level architecture search for deep neural networks | |
| CN111931919B (en) | Sparse neural network computing method and device based on systolic array | |
| CN116136752B (en) | Method and system for determining array input strategy | |
| CN117492371B (en) | Optimization method, system and equipment for active power filter model predictive control | |
| KR102628658B1 (en) | Neural processor and control method of neural processor | |
| CN111400981B (en) | Simulation circuit control system, method, electronic device, and storage medium | |
| CN116402149B (en) | Quantum circuit optimization method and device, storage medium and electronic device | |
| CN111767984A (en) | 8-bit integer full-quantization inference method and device based on fixed displacement | |
| US20230367553A1 (en) | Apparatus and method for multiplication operation based on outer product |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |