Photoelectric collaborative Isin machine computing system and method based on noise-assisted optimization

Abstract

The invention provides a photoelectric collaborative Isin machine calculation system and method based on noise-assisted optimization, and the system comprises a data preprocessing module which carries out the simplification and deformation of any combination optimization problem, generates a corresponding Hamiltonian expression, and provides standardized input data for a photon Isin machine module and a digital Isin calculation module; the photon Isin machine module is used for carrying out rapid annealing by adopting a plurality of spin overturning strategies to obtain a local optimal solution as an initial solution of the digital Isin calculation module; and the digital Isin calculation module adopts a single spin upset strategy to accurately search, further evolves to a global optimal solution on the basis of a local optimal solution provided by a photon Isin machine, and finally solves the problem of combinatorial optimization through cooperative work with the photon Isin machine module. According to the method, the large-scale and high-speed parallel computing characteristics of the photon Isin machine are fully utilized, and the photoelectric cooperative computing technology is combined, so that computing tasks such as solving of the Isin model and the combinatorial optimization problem are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photonic Ising machine computing, in particular to a photoelectric cooperative Ising machine computing system and method based on noise-assisted optimization. BACKGROUND

[0002] With the rapid development of artificial intelligence and big data computing technology, the processing and optimization tasks of complex systems have become increasingly important. The traditional von Neumann computing architecture faces challenges such as low efficiency and high cost when dealing with large-scale computing tasks. Photonic Ising machine is a kind of non-von computing processor based on Ising model, which aims to solve scientific computing problems such as combinatorial optimization. The spin information of the Ising model is encoded in the phase information of the optical field, and the interaction matrix of the Ising model is simulated by adjusting and controlling the optical field and realizing coherent superposition. Finally, the output light intensity of the system reflects the Hamiltonian of the Ising model, and the output result is judged and the light field is changed by the external circuit. In the whole optimization process, the most complex and high-density matrix calculation is borne by the optical circuit, saving a lot of time and computing resources.

[0003] However, as a kind of analog computing method, the calculation accuracy of optical computing is often affected by the resolution of each device in the system and the spontaneous noise. On the one hand, excessive noise will destroy the dynamic process of photonic Ising machine, and even cannot accurately map the Ising model, resulting in failure of problem solving; on the other hand, appropriate noise helps the system Hamiltonian to directly jump out of the local optimal solution in the convergence process, and evolve to the global optimal solution, thereby improving the speed and accuracy of combinatorial optimization problem solving. At present, the calculation accuracy of optical computing is about 8 bits. When the calculation accuracy of optical Ising machine is not enough to accurately reflect the influence of individual spin state flip on the change of system Hamiltonian, the system feedback will fail, thereby destroying the energy convergence dynamics process. This not only limits the possibility of large-scale computing, but also greatly restricts the ability of optical Ising machine in solving complex problems.

[0004] Patent document CN116185125A discloses an optical Ising machine based on Cholesky decomposition, which includes: encoding the interaction matrix of the Ising model to the optical modulation device after Cholesky decomposition; receiving the optical intensity signal carrying the Ising model to realize high-speed calculation of any Ising model.

[0005] However, the patent document CN116185125A only decomposes the Ising model to realize the mapping and solving of any Ising model, and cannot solve the problems of limited precision and noise interference of photonic Ising machine. SUMMARY

[0006] Aiming at the defects in the prior art, the purpose of the present application is to provide an optoelectronic collaborative Ising machine computing system and method based on noise-assisted optimization.

[0007] The optoelectronic collaborative Ising machine computing system based on noise-assisted optimization provided by the present application comprises:

[0008] The data preprocessing module: simplifies and deforms any combinatorial optimization problem to generate a corresponding Hamiltonian expression, and provides standardized input data for the photonic Ising machine module and the digital Ising computing module;

[0009] The photonic Ising machine module: adopts multiple spin-flip strategies for fast annealing to obtain a local optimal solution as the initial solution of the digital Ising computing module;

[0010] The digital Ising computing module: is used for directly searching the ground state energy of the Ising model; adopts a single spin-flip strategy for accurate search, further evolves from the local optimal solution provided by the photonic Ising machine to the global optimal solution, and finally solves the combinatorial optimization problem through the collaborative work with the photonic Ising machine module.

[0011] Preferably, the simplification and deformation of any combinatorial optimization problem to generate a corresponding Hamiltonian expression comprises:

[0012] By introducing intermediate variables or constructing auxiliary blocks, the judgment function of the problem is replaced equivalently in form, so as to meet the form requirement of the Ising model, and the Hamiltonian expression of the Ising model is obtained through simplification.

[0013] Preferably, the input data comprises the scale and interaction matrix of the Ising model, and the weight vector of the spin obtained by decomposing the interaction matrix.

[0014] Preferably, the photonic Ising machine module comprises a laser, an amplitude-type optical modulator, a phase-type optical modulator, an optical Fourier device, a detector and a control unit.

[0015] The laser is responsible for providing a stable coherent light source to provide the required optical signal for the entire optical computing process.

[0016] The amplitude-type optical modulator is used for modulating the amplitude of the light beam to encode the weight vector.

[0017] The phase-type optical modulator is used for modulating the phase of the light beam to perform phase encoding.

[0018] The optical Fourier device is the core part of the photonic Ising machine module, which provides the Fourier transformation function of the light beam.

[0019] The detector is used for receiving and detecting the output light intensity, and feeding back the result to the control unit.

[0020] The control unit is used for controlling the operation of various optical devices and realizing the control of the whole optical circuit.

[0021] Preferably, the digital Ising calculation module comprises a signal input module, a new spin module, a Hamiltonian calculation module, a judgment module, an iteration module and a signal output module.

[0022] The signal input module is responsible for receiving the output state of the photonic Ising machine module as the initial input state.

[0023] The new spin module is responsible for generating a new spin state.

[0024] The Hamiltonian calculation module is responsible for calculating the Hamiltonian corresponding to the spin state generated by the new spin module.

[0025] The judgment module compares the new Hamiltonian calculated by the Hamiltonian calculation module with the current Hamiltonian and judges whether to accept the new spin state as the current spin state.

[0026] The iteration module is responsible for controlling the iteration of the whole spin state optimization process.

[0027] The signal output module is responsible for outputting the final spin state and the corresponding Hamiltonian.

[0028] According to the photoelectric collaborative Ising machine calculation method based on noise-assisted optimization provided by the application, the method comprises the following steps:

[0029] Model mapping step: mapping the combinatorial optimization problem to the Ising model, preprocessing the data, simplifying to obtain the Hamiltonian expression of the Ising model, and determining the interaction matrix of the Ising model.

[0030] Fast annealing step: according to the interaction matrix and the structure of the photonic Ising calculation system, setting the required parameters in the photonic Ising machine, loading the weight coefficient through the optical amplitude modulation device, and realizing the encoding of the interaction matrix.

[0031] Precise search step: taking the local optimal solution as the initial solution of the precise search stage, further optimizing the solution provided by the photonic Ising machine by using the dynamic evolution process of the Ising model until the system stably outputs the final solution.

[0032] Preferably, in the model mapping step, for different combinatorial optimization problems, the cost function or objective function of the original problem is converted into the general form of Hamiltonian through variable substitution and formula transformation, so as to determine the interaction matrix J, spin state X and spin scale N of the Ising model, and the general form of Hamiltonian is as follows:

[0033]

[0034] where J ij represents the interaction matrix, N represents the spin size, x i , x j represent the spin state of the i-th and j-th spin, respectively.

[0035] Preferably, for each Ising model, the coefficients J i,j of the interaction matrix between each group of spins are calculated respectively according to different photonic Ising machines.

[0036] Preferably, the fast annealing step comprises the following sub-steps:

[0037] Step S2.1: randomly generate an initial spin state X0, encode the wavefront of the light beam by using a phase-type optical modulator, and encode the i-th spin state x i of the Ising model into the phase information of the i-th pixel on the phase-type optical modulator

[0038] Step S2.2: directly or indirectly encode the coefficients of the interaction matrix of the Ising model into the amplitude information of the light beam by using an amplitude-type optical modulator;

[0039] Step S2.3: randomly change the state of any number of spins to generate a new spin state X, and encode the new spin state by using a phase-type optical modulator;

[0040] Step S2.4: realize the coherent superposition of the light beam by using an optical Fourier device, and realize the indirect calculation of the Hamiltonian of the Ising model;

[0041] Step S2.5: detect the light intensity at the center point of the output light, and the light intensity is proportional to the Hamiltonian of the Ising model;

[0042] Step S2.6: determine whether the output light intensity increases, if yes, keep the new spin state; if no, accept the new spin state according to the metropolis criterion;

[0043] Repeat steps S2.3 to S2.6 until a predetermined number of iterations T is reached or a global optimal solution is reached.

[0044] Preferably, the exact search step comprises the following sub-steps:

[0045] Step S3.1: take the spin state output by the fast annealing step as the initial spin state, calculate the Hamiltonian according to the obtained Ising model, and the calculation formula is as follows:

[0046]

[0047] where J ij represents an interaction matrix, N represents a spin size, x i , x j represent the spin states of the i-th and j-th spins, respectively;

[0048] Step S3.2: randomly generating a new spin state;

[0049] Step S3.3: calculating the Hamiltonian in the new spin state, and determining whether the Hamiltonian is less than the current value, if yes, accepting the new spin state; if no, continuing to maintain the current spin state;

[0050] Steps S3.2 to S3.3 are repeatedly executed until the Hamiltonian corresponding to the spin state is the ground state Hamiltonian, and then the ground state spin state and the ground state Hamiltonian of the Ising model are output, and the optimal solution of the combinatorial optimization problem is calculated.

[0051] Compared with the prior art, the present application has the following beneficial effects:

[0052] 1. The present application completely discloses a noise-assisted optimization optoelectronic collaborative Ising machine system and operation method, filling the gap in the related field. The system not only builds a complete, practical application-oriented computing system, but also provides a general method with high universality and practicality.

[0053] 2. The present application significantly improves the computing performance of the system by combining optoelectronic Ising machine collaborative computing. Compared with traditional photonic Ising machines or electronic Ising models, the computing process of the present application combines the advantages of fast annealing of optical computing systems and precise search of digital computing systems. This collaborative working method not only enhances the global search capability, but also enables the system to obtain more accurate results when facing complex computing tasks.

[0054] 3. The modular design of the system in the present application not only retains the structural design and computing advantages of each module, but also enables each functional module to be independently upgraded and optimized.

[0055] 4. The present application improves the general iterative process of the Ising model in the optoelectronic collaborative computing system. By adopting different spin flipping strategies for different Ising computing modules, the system can find or approach the optimal solution more quickly, effectively avoiding the problem of being trapped in a local optimal solution, thereby improving the efficiency of solving combinatorial optimization problems. BRIEF DESCRIPTION OF DRAWINGS

[0056] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments, made with reference to the accompanying drawings:

[0057] Figure 1 The overall flowchart of the noise-assisted optimized optoelectronic collaborative Ising machine system provided by the present application is shown in the figure;

[0058] Figure 2 The schematic diagram of the noise-assisted optimized optoelectronic collaborative Ising machine system of the embodiment of the present application is shown in the figure;

[0059] Figure 3 The flowchart of the operation method of the noise-assisted optimized optoelectronic collaborative Ising machine of the embodiment of the present application is shown in the figure;

[0060] Figure 4 The structural schematic diagram of the spatial photonic Ising machine of the embodiment of the present application is shown in the figure;

[0061] Figure 5 The iteration result example diagram for solving the maximum cut problem of the embodiment of the present application is shown in the figure.

[0062] The figure shows the structure of the present application. DETAILED DESCRIPTION

[0063] The present application will be described in detail below with specific embodiments. The following embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any form. It should be pointed out that for those skilled in the art, without departing from the concept of the present application, a number of changes and improvements can be made. These all belong to the protection scope of the present application.

[0064] The present application utilizes the self-generated noise of the system to assist in constructing a smooth Ising model, improves the possibility of jumping out of the local optimal solution of the simulated annealing algorithm, accelerates the convergence speed of the Hamiltonian, and thus provides a stable local minimum value; the precise search module constructed based on the digital Ising model takes the spin state searched by the photonic Ising machine as the initial state, accurately calculates the Hamiltonian of the system by using an electronic computer, gradually approaches the ground state through continuous iteration, and finally accurately finds the global minimum value, so as to realize efficient calculation. The present application fully utilizes the characteristics of large-scale and high-speed parallel calculation of the photonic Ising machine, combines with the optoelectronic collaborative calculation technology, realizes the calculation tasks of the Ising model and the solution of the combinatorial optimization problem, etc. The present application improves the calculation accuracy and efficiency of the photonic Ising machine by optimizing the noise management, so as to enhance its ability to solve complex problems.

[0065] Embodiment one

[0066] According to the noise-assisted optimized optoelectronic collaborative Ising machine calculation system provided by the present application, as shown in the figures, Figure 1 and Figure 2 , it comprises:

[0067] Data preprocessing module: Simplify and deform any combinatorial optimization or other scientific computing problems, realize the formal equivalent replacement of the problem's judging function by introducing intermediate variables or constructing auxiliary blocks, so as to make it meet the formal requirements of the Ising model, and simplify to get the Hamiltonian expression of the Ising model. The role of this module is to convert any combinatorial optimization or other scientific computing problems into the Ising model, generate the corresponding Hamiltonian expression and other standardized input data for subsequent photonic Ising machine calculation and digital Ising calculation.

[0068] Specifically, for different combinatorial optimization problems, the way to convert into the Ising model is different. What needs to be ensured in the conversion process is that the final expression does not contain high-order terms higher than the quadratic term, and meets the formal requirements of the Ising model. According to the converted Hamiltonian expression, the size and interaction matrix of the Ising model are extracted respectively. Further, the weight vector of the spin is obtained by decomposing the interaction matrix.

[0069] Photonic Ising machine module: The photonic Ising machine module is an optical computing module designed based on the Ising physical model, which adopts multiple spin flipping strategies for fast annealing to provide a reliable initial solution for the digital Ising calculation module. According to the interaction value and size of the given Ising model, the intensity modulation value and phase modulation value of the incident light are set, wherein the intensity modulation value, i.e. the weight vector, is determined by the interaction value of the Ising model, and the phase modulation value is determined by the spin state; the Fourier optical device is used to complete the calculation of the Hamiltonian; the multiple spin flipping strategy is adopted to speed up the evolution of the system to the ground state; and the spontaneous noise of the photonic Ising machine is used to assist the simulated annealing algorithm to jump out of the local optimal solution, thereby improving the global search ability.

[0070] The photon Ising machine module includes a laser, an amplitude optical modulator, a phase optical modulator, an optical Fourier device, a detector, and a control unit. The laser is responsible for providing a stable coherent light source to provide the required optical signal for the entire optical calculation process. Through the stable laser output, the accuracy and reliability of the optical calculation are ensured. The amplitude optical modulator is used to modulate the amplitude of the light beam for weight vector encoding. By adjusting the amplitude of the light beam, the mapping of the interaction matrix of the Ising model is realized, thereby simulating the interaction relationship in the Ising model. The phase optical modulator is used to modulate the phase of the light beam for phase encoding. By adjusting the phase of the light beam, the update of the spin state is realized, thereby simulating the spin flipping process in the Ising model. The optical Fourier device is the core part of the photon Ising machine module, which provides the Fourier transform function of the light beam. Through the Fourier transform, the multiplication and addition operation of a certain specific matrix is realized, thereby completing the calculation of the system Hamiltonian. The high efficiency and parallelism of optical calculation are used to improve the calculation speed and efficiency. The detector is used to receive and detect the output light intensity, and the results are fed back to the control unit. According to the measurement results of the detector, the energy information of the system can be obtained, which provides the basis for subsequent iterative calculation. The control unit is used to control the work of various optical devices and realize the control of the entire optical circuit. At the same time, the control unit also uses the simulated annealing algorithm to judge the measurement results of the optical circuit, and controls the iteration process of the circuit according to the judgment results. Through the application of the simulated annealing algorithm, the search performance of the photon Ising machine can be effectively optimized.

[0071] As an analog computing platform, the photon Ising machine module fully utilizes the high-speed parallelism and inherent Fourier convolution characteristics of optical calculation, and has high tolerance to noise. Rational use of noise resources helps to build a smooth Ising model, which is helpful for the rapid convergence of the simulated annealing algorithm.

[0072] Digital Ising calculation module: The digital Ising calculation module is a digital calculation module designed based on the Ising model, which is used to directly search for the ground state energy of the Ising model; adopt a single spin flipping strategy to accurately search, further evolve from the local optimal solution provided by the photon Ising machine to the global optimal solution, and through the cooperative work with the photon Ising machine module, the combinatorial optimization problem is finally solved, and the accuracy of the calculation result is improved.

[0073] The digital Ising computing module comprises a signal input module, a new spin module, a Hamiltonian computing module, a judging module, an iteration module and a signal output module. The signal input module is responsible for receiving the output state of the photonic Ising machine module as an initial input state, and calculating the Hamiltonian of the initial state according to the Ising model to provide basic data for subsequent spin state optimization. The new spin module is responsible for generating a new spin state. The Hamiltonian computing module is responsible for calculating the Hamiltonian corresponding to the spin state generated by the new spin module. The judging module compares the new Hamiltonian calculated by the Hamiltonian computing module with the current Hamiltonian and judges whether to accept the new spin state as the current spin state. If the new Hamiltonian is smaller than the current Hamiltonian, it means that the change of the spin state is beneficial, so the change is accepted; otherwise, the spin state before the change is kept unchanged. This judgment mechanism ensures that the system always evolves towards a state with lower energy and higher stability. The iteration module is responsible for controlling the iteration of the whole spin state optimization process. By repeatedly triggering the new spin module, the Hamiltonian computing module and the judging module process, the system continuously tries new spin states and judges their advantages and disadvantages until the spin state corresponding to the lowest Hamiltonian is found. At this time, the system considers that the spin ground state of the Ising model has been reached, that is, the global optimal solution. The signal output module is responsible for outputting the final spin state and the corresponding Hamiltonian. According to this spin state, the system can further calculate the optimal solution of the original combinatorial optimization problem. This optimal solution is obtained by decoding the spin state, which represents the best configuration or strategy in the actual problem.

[0074] The digital Ising computing module fully utilizes the high-precision advantage of digital calculation, and further improves the precision and accuracy of the system.

[0075] Embodiment two

[0076] According to the photoelectric collaborative Ising machine computing method based on noise-assisted optimization provided by the application, as shown in the formula (1), the method comprises the following steps: Figure 3 As shown in the formula (1), the method comprises the following steps:

[0077] The model mapping step: mapping the combinatorial optimization problem to the Ising model, preprocessing the data, simplifying to obtain the Hamiltonian expression of the Ising model, and determining the interaction matrix of the Ising model. By determining the interaction matrix of the Ising model, it is converted into the form of vector product by using characteristic decomposition or other decomposition methods, which provides a basis for the subsequent model mapping process. For different combinatorial optimization problems, the cost function or objective function of the original problem is transformed into the general form of Hamiltonian through variable substitution and formula transformation, so as to determine the interaction matrix J of the Ising model, the spin state X and the spin scale N. The general form of Hamiltonian is as follows:

[0078]

[0079] Among them, J ij represents the interaction matrix, N represents the spin scale, x i 、x j Represent the spin states of the i-th and j-th spins respectively. For each Ising model, according to different photon Ising machines, the coefficients J of the interaction matrix between each group of spins need to be calculated separately. i,j For the space photon Ising machine system, it is necessary to further calculate the coefficient J of the interaction matrix i,j The weight vector W obtained after decomposition i ,W j If the rank of the interaction matrix J is 1, then the corresponding weight vector J = W can be obtained by vector decomposition. T *W. If the rank of the interaction matrix J is greater than 1, the interaction matrix can be decomposed by eigenvalue. Where N is the size of the Ising model, λ i is the i-th eigenvector W i The corresponding eigenvalue. Each eigenvector can be encoded onto the amplitude-type spatial light modulator in sequence using the time-division multiplexing method. The corresponding Hamiltonian can be written as the sum of the Hamiltonians of N rank-1 Ising models:

[0080]

[0081] Among them, w i,l represents the weight value of the lth spin of the rank-one Ising model of the i-th group after eigenvalue decomposition, x l represents the lth spin, and N is the number of spins. In addition to the eigenvalue decomposition described above, other decomposition methods such as Cholesky decomposition can also be used to transform the high-rank Ising model into the sum of multiple rank-1 Ising models. Eigenvalue decomposition is the most commonly used decomposition method.

[0082] Rapid annealing step: According to the interaction matrix and the structure of the photon Ising computing system, the required parameters are set in the photon Ising machine, including: model mapping parameters: Ising model scale N, interaction matrix J; annealing related parameters: initial temperature T0, annealing rate q, final temperature T end Initialize relevant parameters: initial spin state X0, initial Hamiltonian H0, etc. Load weight coefficients through optical amplitude modulation devices to achieve encoding of the interaction matrix. Use a spatial light modulator to load spin information and achieve coherent superposition of optical signals in the optical domain. The output light intensity of the system is used as the cost function of the simulated annealing algorithm to guide the spin flip in the new round of iteration. The updated spin information is re-input into the photonic Ising machine, so that the photonic Ising model continuously completes new iterations until it stably outputs the local optimal solution. The rapid annealing step includes the following sub-steps:

[0083] Step S2.1: Randomly generate an initial spin state X0, and encode the wavefront of the light beam by using a phase-type optical modulator. The i-th spin state x i = ±1 of the Ising model is encoded into the phase information of the light beam The two states of the spin {±1} correspond to two opposite phase states of the light field, for example, {0, π} or

[0084] Step S2.2: Directly or indirectly encode the interaction matrix coefficients of the Ising model into the amplitude information of the light beam by using an amplitude-type optical modulator.

[0085] Step S2.3: Randomly change the state of any number of spins to generate a new spin state X, and encode the new spin state by using a phase-type optical modulator. That is, modulate the phase distribution φ of the light beam according to the new spin state. In actual operation, the number n of spins flipped each time can be determined according to the relationship between the current iteration number t and the total iteration number and the total spin size N:

[0086]

[0087] where ρ is a randomly generated proportionality coefficient (0 < ρ < 1). This flipping strategy improves the convergence rate of the system in the initial stage of iteration compared to flipping only a single spin each time, achieving the effect of fast annealing, while ensuring that when the Hamiltonian evolves to the vicinity of a local optimal solution, the system will perform a small step search to avoid skipping the optimal solution.

[0088] Further, according to the new spin state distribution X, a corresponding phase distribution φ is generated. When the state of a spin x is 1 (-1), the spin phase of the corresponding pixel point on the optical phase modulation device should be 0 (π).

[0089] Step S2.4: Realize the coherent superposition of the light beams by using an optical Fourier device to realize the indirect calculation of the Hamiltonian of the Ising model. The optical Fourier device mainly realizes coherent superposition by beam combination or focusing of coherent light beams. The light intensity at the center point after beam combination can be represented as:

[0090]

[0091] where w l is the weight value of the l-th spin, is the phase corresponding to the l-th spin. At this time, the output light intensity I can indirectly represent the Hamiltonian of the Ising model where the interaction matrix coefficient J i,j = w i w j ​.

[0092] Step S2.5: detecting the light intensity at the center point of the output light using the optical detection device, the light intensity being proportional to the Hamiltonian of the Ising model, which is expressed as follows:

[0093]

[0094] where J ij represents the interaction matrix, N represents the spin scale, x i , and x j represent the spin states of the i-th and j-th spins, respectively. This step provides a direct measurement of the energy or cost function of the current spin state.

[0095] Step S2.6: determining whether the output light intensity is increased using the electronic computer, if yes, retaining the new spin state; if no, accepting the new spin state according to the metropolis criterion. Specifically, determining whether the output light intensity I_new of the new spin state is greater than the output light intensity I of the current state. If I_new>I, retaining the new spin state; otherwise, accepting the new spin state according to the metropolis criterion. This step ensures that the algorithm can search in the direction of lower energy.

[0096] Repeating steps S2.3 to S2.6 until a predetermined number of iterations T is reached or a global optimal solution is reached.

[0097] Precise search step: using the dynamic evolution process of the Ising model to further optimize the solution provided by the photonic Ising machine, taking the local optimal solution as the initial solution of the precise search phase, until the system stably outputs the final solution. The precise search step includes the following sub-steps:

[0098] Step S3.1: taking the spin state output by the fast annealing step as the initial spin state, calculating the Hamiltonian according to the obtained Ising model, and the calculation formula is as follows:

[0099]

[0100] where J ij represents the interaction matrix, N represents the spin scale, x i , and x j represent the spin states of the i-th and j-th spins, respectively.

[0101] Step S3.2: randomly generating a new spin state, specifically, in each iteration, randomly selecting one spin and flipping its state, and generating a new spin state by flipping one spin each time. This step ensures that different regions of the solution space are accurately explored in small steps.

[0102] Step S3.3: Calculate the Hamiltonian in the new spin state, and determine whether the Hamiltonian is less than the current value, if yes, accept the new spin state; if no, continue to maintain the current spin state.

[0103] Repeat steps S3.2 to S3.3 until the Hamiltonian corresponding to the spin state is the ground state Hamiltonian, then output the ground state spin state and the ground state Hamiltonian of the Ising model, and calculate the optimal solution of the combinatorial optimization problem.

[0104] The present application aims to improve the ability of Ising machine to solve problems, and uses the spontaneous noise of photonic Ising machine and the method of optoelectronic cooperative calculation to improve the search accuracy and the accuracy of solving problems of Ising machine system. Digital calculation is used to compensate for the defects of limited precision of analog calculation, thereby improving the performance of Ising machine calculation system. In addition, the system spontaneous noise of optoelectronic cooperative Ising machine can assist the simulated annealing algorithm to jump out of the local optimal solution.

[0105] Embodiment three

[0106] Embodiment three is a preferred example of embodiment one, and on the basis of embodiment one, the data preprocessing module is further described in this embodiment.

[0107] In the data preprocessing module of this embodiment, a spatial photonic Ising machine is selected as the photonic Ising machine system, and a standard spatial photonic Ising machine should include:

[0108] 1) Laser, as a light source, emits a coherent light beam to provide stable light input for the entire system.

[0109] 2) Beam expander, used to expand the diameter of the light beam and reduce the divergence angle of the light beam to improve the quality of the light beam.

[0110] 3) Spatial light modulator, as the core element of the system, used for wavefront control of the light beam to realize spin coding.

[0111] 4) Lens, used for spatial optical Fourier transformation to realize coherent superposition of spin signals.

[0112] 5) Detector, as a photodetector, used to capture the focused light spot, convert it into an electrical signal and output it.

[0113] 6) Electronic computer, as the control center of the system, receives the electrical signal output by the detector, processes and judges, generates a new spin state, and refreshes the phase of the spatial light modulator through feedback control.

[0114] For example, Figure 4As shown, the spatial photonic Ising machine system in this embodiment comprises: a helium-neon laser 1, a polarizer 2, a beam expander 3, a beam splitter 4, a spatial light modulator 5, a plano-convex lens 6, a CCD camera 7 and an electronic computer 8. Among them, the optical circuit comprises: a coherent light beam emitted by the helium-neon laser 1, after adjusting the polarization state through the polarizer 2, the light beam diameter is enlarged through the beam expander 3, and then divided into multiple paths through the beam splitter 4. One of the light beams is incident on the spatial light modulator 5, and the reflected light after modulation is focused through the plano-convex lens 6, and is detected at the focal point by the CCD camera 7. The electrical circuit comprises: the CCD camera 7 converts the detected light intensity signal into an electrical signal and inputs it into the electronic computer 8. The electronic computer 8 compares the newly detected light intensity value with the currently saved light intensity value, and judges whether to retain the new spin state as the current state according to the comparison result. Then, the electronic computer 8 generates a new spin state and refreshes the phase of the spatial light modulator 5 through feedback control, realizing the simulation and optimization of the Ising model.

[0115] The spatial modulator 5 is a phase-type spatial light modulator, wherein: the pixels in the modulation area are divided into two regions, and the phases of the corresponding pixels in the two regions are modulated as i and i respectively according to a weight and wherein π represents the spin state phase ±1, and i = cos -1 w i is the weight state phase. By using the coherent superposition of the light beams of the two regions, the weight and state information of the spin can be encoded at the same time. Further, in order to ensure successful encoding, 10x10 pixels are taken as a pixel unit for simultaneous encoding in actual operation.

[0116] It is worth noting that, in addition to the spatial photonic Ising machine, other types of photonic Ising machines that can completely map the Ising model and adapt to the simulated annealing algorithm can also be used as the photonic Ising machine system of module M2 for fast annealing calculation.

[0117] Embodiment Four

[0118] Embodiment Four is a preferred example of Embodiment One, which more specifically illustrates the advantages of the present application.

[0119] This embodiment takes the maximum cut problem as an example to verify that the solving result of the present application is better than that of the traditional photonic Ising machine system when solving a combinatorial optimization problem. The maximum cut problem is a weighted maximum cut problem of 20736 vertices, and the weight of each vertex is selected as a random integer of -10-10.

[0120] As Figure 5As shown, the system and method of the present embodiment are used to solve the maximum cut problem, and the iterative results of 10 searches are compared with the traditional photon Ising machine and electronic Ising model. The results show that under the condition of the same number of iterations, the present method has a significant advantage, providing the maximum cut value result 204193274, which is much higher than the search result 176701819 of the photon Ising machine and the search result 130222153 of the electronic Ising model. Further, according to the multiple iteration distribution shown in the figure, it can be seen that the method of the present application combines the stability of digital calculation, so that the convergence result is more consistent, and the result variance is much lower than that of the photon Ising machine, which further verifies the effectiveness and superiority of the present method.

[0121] The present embodiment verifies that under the condition of large-scale complex graph problems, the optoelectronic collaborative Ising machine can not only find better solutions than the traditional photon computer and digital Ising model, but also has the advantages of high noise tolerance and high precision.

[0122] In summary, the present application provides an optoelectronic collaborative Ising machine computing system and method based on noise-assisted optimization. The system combines the fast annealing technology of the photon Ising machine and the precise search algorithm of the digital Ising model to achieve efficient simulation and optimization of the Ising model. Further, the system has a compact structure design, simple operation process, fast operation speed and excellent precision performance, providing a new solution for solving combinatorial optimization problems.

[0123] At present, with the continuous development of special computing devices related to the Ising model, the present application is expected to provide strong support for breaking through the inherent defects of various optical and electronic Ising machines. Through the optoelectronic collaborative method of the present application, it is expected to promote the speed, efficiency and accuracy of the Ising machine in solving problems to achieve a qualitative leap. This will not only promote the technological progress in related fields, but also provide more powerful and efficient tools for solving complex problems in practical applications.

[0124] Those skilled in the art know that in addition to implementing the system provided by the present application and each device, module, unit thereof in pure computer readable program code, the same function can be achieved by logically programming the method steps in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers and embedded microcontrollers. Therefore, the system provided by the present application and each device, module, unit thereof can be considered as a hardware component, and the devices, modules and units included therein for achieving various functions can also be considered as structures within the hardware component. The devices, modules and units for achieving various functions can also be considered as both software modules and structures within the hardware component.

[0125] The specific embodiments of the present application are described above. It needs to be understood that the present application is not limited to the specific embodiments described above, and various changes or modifications can be made by those skilled in the art within the scope of the claims, which does not affect the essential content of the present application. The embodiments of the present application and the features in the embodiments can be combined with each other at will without conflict.

Claims

1. A noise-assisted optimization based opto-electronic cooperative Ising machine computing system, characterized by, The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system.

2. The noise-assisted optimization based optoelectronic cooperative Ising machine computing system of claim 1, wherein, The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system.

3. The noise-assisted optimization based optoelectronic cooperative Ising machine computing system of claim 1, wherein, The application relates to a photonic Ising machine and a digital Ising computing system.

4. The noise-assisted optimization based optoelectronic cooperative Ising machine computing system of claim 1, wherein, The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system.

5. The noise-assisted optimization based optoelectronic cooperative Ising machine computing system of claim 1, wherein, The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system.

6. A method of photoelectric synergistic Ising machine computation based on noise- assisted optimization, characterized in that, The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Ising computing system. The application relates to a photonic Ising machine and a digital Is The precise search step: taking the local optimal solution as the initial solution of the precise search stage, further optimizing the solution provided by the photonic Ising machine by using the dynamic evolution process of the Ising model until the system stably outputs the final solution.

7. The noise-assisted optimization-based optoelectronic cooperative Ising machine calculation method according to claim 6, wherein, In the model mapping step, for different combinatorial optimization problems, the cost function or objective function of the original problem is transformed into the general form of the Hamiltonian by variable substitution and formula transformation, so as to determine the interaction matrix J, spin state X and spin scale N of the Ising model, and the general form of the Hamiltonian is as follows: where J ij represents the interaction matrix, N represents the spin size, x i , x j represent the spin state of the i-th and j-th spins, respectively.

8. The noise-assisted optimization based optoelectronic cooperative Ising machine computing method of claim 6, wherein, For each Ising model, the coefficients J of the interaction matrix between spins of each group are calculated according to different photonic Ising machines i,j .

9. The noise-assisted optimization based optoelectronic cooperative Ising machine computing method of claim 6, wherein, The fast annealing step includes the following sub-steps: Step S2.1 : Randomly generate an initial spin state X0, encode the wavefront of the light beam with a phase-type optical modulator, and encode the spin state x = ±1 of the Ising model into the phase information of the light beam i,j = ±1 wherein the subscripts i, j represent the spin state of the i-th and j-th spins, respectively; Step S2.2: using an amplitude-type optical modulator to directly or indirectly encode the interaction matrix coefficients of the Ising model into the amplitude information of the light beam respectively; Step S2.3: randomly changing the state of any number of spins to generate a new spin state X, and using a phase-type optical modulator to encode the new spin state; Step S2.4: using an optical Fourier device to realize the coherent superposition of the light beam, thereby realizing the indirect calculation of the Hamiltonian of the Ising model; Step S2.5: detecting the light intensity at the center point of the output light, and the light intensity is proportional to the Hamiltonian of the Ising model; Step S2.6: determining whether the output light intensity increases, if yes, the new spin state is retained; if not, the new spin state is accepted according to the metropolis criterion; Repeat steps S2.3 to S2.6 until a predetermined number of iterations T is reached or a global optimal solution is reached.

10. The noise-assisted optimization based optoelectronic cooperative Ising machine computing method of claim 6, wherein, The precise search step includes the following sub-steps: Step S3.1: taking the spin state output by the fast annealing step as the initial spin state, calculating the Hamiltonian according to the obtained Ising model, and the calculation formula is as follows: where J ij represents the interaction matrix, N represents the spin size, x i , x j represent the spin state of the i-th and j-th spins, respectively; Step S3.2: randomly generating a new spin state; Step S3.3: calculating the Hamiltonian under the new spin state, and determining whether the Hamiltonian is less than the current value, if yes, the new spin state is accepted; if not, the current spin state is maintained; Repeat steps S3.2 to S3.3 until the spin state corresponds to the ground state Hamiltonian, then output the ground state spin state and the ground state Hamiltonian of the Ising model, and calculate the optimal solution of the combinatorial optimization problem.

Images