Noise-robust correction update method and system for optical Ising machine and medium

Abstract

The application provides a noise-resistant correction updating method and system of an optical Ising machine and a medium, and belongs to the technical field of optical computing and optical optimization solving. The method aims to solve the problem of inaccurate and unstable solving of the optical Ising machine caused by physical noise and system uncertainty. The core lies in that in the iterative solving process, a plurality of candidate spin flip schemes are first generated, and a reliability index is constructed for each scheme to perform preselection and determine the candidate state to be measured that is most worthy of measurement. Then, optical measurement is performed on the candidate state to be measured, and correction acceptance judgment is performed in combination with the theoretical optical response trend and the actual measurement value, and the spin state is updated if the judgment is accepted. Finally, the stability index is continuously monitored in the iteration, and when the preset freezing condition is met, it is judged that the system converges and the optimal solution is output. The combination mechanism of preselection, correction and freezing introduced in the application can effectively resist noise interference, and significantly improve the accuracy, stability and repeatability of the optical Ising machine solving.

Description

Technical Field

[0001] This invention relates to the field of optical computing and optical optimization solution technology, and more specifically, to a method, system, and computer-readable storage medium for an optical ising machine. Background Technology

[0002] An optical isosing machine is a specialized computing device that utilizes optical systems to accelerate the solution of combinatorial optimization problems, such as quadratic unconstrained binary optimization problems. Its basic principle is to map the problem matrix to be solved into physical parameters of the optical system, such as intensity modulation or phase modulation, and then find the optimal solution through optical field evolution and measurement. Existing solutions typically involve encoding the problem matrix into optical modulation information, loading it onto devices such as spatial light modulators, constructing and measuring the optical field, and then searching for the optimal solution based on the measurement results through iterative updates and simulated annealing mechanisms.

[0003] However, there is a gap between the ideal theoretical model and the real physical system. In practical optical isolators, the entire measurement chain is inevitably affected by various noise and uncertainty factors, such as shot noise and thermal noise of photodetectors, fluctuations in the response of optical devices themselves, optical path drift, system alignment deviations, and environmental temperature disturbances. In existing technologies, a simple decision-making method is usually adopted, which involves performing an optical measurement on a randomly selected candidate solution and directly judging acceptance or rejection based on the measurement result. This simple decision-making method faces severe challenges in physical systems with significant noise: theoretically favorable candidate solutions may produce very weak signal changes that are easily drowned out by noise and cannot be correctly identified; conversely, an unfavorable candidate solution may be incorrectly accepted due to the random superposition of noise. This leads to an unstable solution process, resulting in unsatisfactory accuracy and repeatability of the final solution, which severely restricts the practical performance of optical isolators.

[0004] In the prior art, patent CN118974723A discloses a general system optimization method implemented at the pure computational level. Its core lies in receiving multiple candidate parameter value sets, applying an optimization algorithm to each candidate set to generate candidate states, and evaluating and selecting the optimal candidate state based on an "optimality metric." This method does not address the noise and uncertainty issues in the physical measurement system. However, this invention applies the idea of ​​"multi-candidate scheme pre-selection" to the physical implementation of the optical Ising method. After obtaining the optical measurement values, it does not directly adopt them but compares them with the theoretical optical response trend to determine acceptance, rejection, or other specific processing. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a noise-resistant correction and update method and system for optical ising machines. This method and system are intended to solve the technical problem that, in the physical implementation of existing optical ising machines, uncertainties such as detection noise, device fluctuations, and optical path drift lead to errors in the evaluation and decision-making of candidate solutions, thereby affecting the accuracy, stability, and repeatability of the solution.

[0006] According to the present invention, a noise-resistant spin correction update method for an optical ising machine includes at least one spatial light modulator (SLM) and a digital micromirror device (DMD). The at least one SLM is used to load fixed optical modulation information corresponding to the problem to be optimized, and the DMD is used to load state optical modulation information corresponding to the current spin state. The method includes: Repeat the iterative steps until the preset freezing condition is met; the iterative steps include: Multiple candidate spin-flip schemes are generated, and a candidate state reliability index is constructed for each candidate spin-flip scheme. Based on the candidate state reliability index, multiple candidate states corresponding to the multiple candidate spin-flipping schemes are pre-selected to determine at least one candidate state to be tested. The optical modulation information of the state loaded on the digital micromirror device (DMD) is updated according to the candidate state to be tested, and optical measurement is performed on the candidate state to be tested to obtain the measurement value. The theoretical optical response trend of the candidate state to be tested is determined, and a correction acceptance decision is made by combining the theoretical optical response trend with the measured value; wherein, if the decision is to accept, the spin state is updated based on the candidate state to be tested; if the decision is to reject, at least one of the following processes is performed: The following options are available: reject the candidate state to be tested, restore the spin state to the state before the pre-selection in the current iteration, reduce the priority of the candidate state to be pre-selected in subsequent iteration steps, or perform a retest on the candidate state to be tested. During the iteration process, stability indicators are monitored, and the spin state with the lowest Hamiltonian energy value corresponding to the problem to be optimized is recorded and updated as the historical preferred state, wherein the preset freezing condition is determined based on the stability indicators. When the freezing condition is met, output the current spin state and / or the historical preferred state.

[0007] Preferably, the candidate state reliability index is constructed based on one or more of the following information: theoretical light intensity change information, theoretical optical response change information, affected item information, historical measurement statistics, noise characterization information, and device / optical path stability information.

[0008] Preferably, the correction acceptance decision considers at least one of the following: the consistency between the theoretical optical response trend and the measured value, and the distinguishability of the measured value relative to the noise background.

[0009] Preferably, the stability index includes at least one of the following: state acceptance rate within a predetermined window, rate of change of measured value, and state flip frequency.

[0010] Preferably, the correction acceptance decision is a probabilistic decision, and its acceptance probability depends on the degree of consistency between the theoretical optical response trend and the measured value.

[0011] According to the present invention, a noise-resistant correction and update method for an optical isolator is provided, applicable to a coherent optical isolator. The method includes: Based on the problem to be optimized, fixed optical modulation information is generated, and state optical modulation information is generated based on the current spin state. The fixed optical modulation information and the state optical modulation information are jointly loaded onto at least one spatial light modulator. The iterative steps are repeated until the preset freezing condition is met.

[0012] Preferably, the fixed optical modulation information and the state optical modulation information are jointly loaded onto a single spatial light modulator through time-division multiplexing, space-division multiplexing, code superposition, or partition loading.

[0013] According to the present invention, an anti-noise correction and update system for an optical ising machine includes a control computer, a light source, at least one spatial light modulator (SLM), a digital micromirror device (DMD), an optical system, and a photodetector. The at least one SLM is used to load fixed optical modulation information corresponding to the problem to be optimized. The DMD is used to load state optical modulation information corresponding to the current spin state. The light beam emitted by the light source interacts with the fixed optical modulation information and the state optical modulation information through the optical system, and is then detected by the photodetector to obtain a measurement value. The control computer includes: The pre-selection unit is configured to generate multiple candidate spin-flip schemes in each iteration, construct a candidate state reliability index for each candidate spin-flip scheme, and perform pre-selection based on the candidate state reliability index to determine at least one candidate state to be tested. The measurement control unit is configured to update the state optical modulation information loaded on the digital micromirror device (DMD) according to the candidate state to be measured, and to acquire the measurement value measured by the photodetector. The determination and update unit is configured to determine the theoretical optical response trend of the candidate state to be tested, and combine the theoretical optical response trend with the measured value to perform a correction acceptance determination. When the determination is accepted, the spin state is updated. When the determination is not accepted, at least one of the following processes is performed: rejecting the candidate state to be tested, restoring the spin state to the state before the pre-selection unit determined the at least one candidate state to be tested, reducing the priority of the candidate state to be tested being pre-selected in subsequent iterations, or instructing the measurement control unit to perform a retest on the candidate state to be tested. The monitoring and output unit is configured to monitor stability indices during the iteration process, record and update the spin state with the lowest Hamiltonian energy value corresponding to the problem to be optimized as the historical preferred state, and control the output of the current spin state and / or the historical preferred state when the stability indices meet the preset freezing conditions.

[0014] Preferably, the fixed optical modulation information loaded onto the at least one spatial light modulator (SLM) remains unchanged during the solution process, while the state optical modulation information loaded onto the digital micromirror device (DMD) changes dynamically with the spin state update during the iteration process; the photodetector is a single-pixel photodetector.

[0015] According to the present invention, a computer-readable storage medium is provided thereon storing a computer program, which, when executed by a processor, implements the noise-resistant correction and update method for the optical ising machine.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. Improved solution accuracy and stability: By actively filtering low-quality candidate solutions and correcting noisy measurements, the probability of erroneous decisions is significantly reduced, enabling the solution process to converge more stably to the global optimum, thus improving the accuracy and reliability of the final solution.

[0017] 2. Enhanced result repeatability: Effectively suppresses result fluctuations caused by random noise and drift in the physical system, enabling more consistent results to be obtained when solving the same problem in different experimental rounds, thus improving repeatability stability.

[0018] 3. Reduced extreme requirements on hardware: The algorithm in this application can compensate for the imperfections of the hardware system to a certain extent, thereby reducing the extreme requirements on the stability of optical devices, detector sensitivity and environmental control, and enhancing the engineering practicality of the method.

[0019] 4. Wide range of applications: The core noise reduction and correction algorithm mechanism has good platform independence and can be applied to optical ising machines with various physical implementation schemes.

[0020] 5. Synergistic Gain of Architecture and Algorithm: The SLM+DMD optoelectronic hybrid system provided by this invention carries fixed optical modulation information and state optical modulation information in a synergistic manner by SLM and DMD, respectively, forming a hardware architecture suitable for performing optical measurement feedback updates, providing a reliable physical implementation basis for the core polarization correction algorithm; the core polarization correction algorithm fully utilizes the measurement feedback capability of this architecture, and the two complement each other and synergistically enhance the solution performance. Attached Figure Description

[0021] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart illustrating an anti-noise correction and update method for an optical ising machine, as provided in an embodiment of this application.

[0022] Figure 2 This is a schematic diagram of an SLM+DMD optoelectronic hybrid system structure in an embodiment of this application.

[0023] Figure 3 This is a conceptual diagram illustrating the candidate state reliability pre-selection step in the embodiments of this application.

[0024] Figure 4 This is a conceptual diagram illustrating the correction acceptance determination step in the embodiments of this application.

[0025] Figure 5 This is a schematic diagram of a coherent optical Ising system in an embodiment of this application.

[0026] Figure labeling: 10-Control computer; 20-Light source; 30-Spatial light modulator; 40-Digital micromirror device; 50-Optical system; 60-Single-pixel photodetector; 120-Laser; 160-Area array camera; S101-Initialization; S102-Generate candidate schemes; S103-Pre-selection; S104-Optical measurement; S105-Correction determination; S106-Update status; S107-Stability monitoring; S108-Freeze output; C1…CN-Candidate schemes; R1…RN-Reliability index; E_theory-Theoretical response curve; E_measured-Measured value curve; N_floor-Noise floor. Detailed Implementation

[0027] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0028] Example 1 This embodiment provides a noise-resistant correction update method for optical ising machines. This method is implemented on an optoelectronic hybrid system based on spatial light modulators (SLM) and digital micromirror devices (DMD), aiming to solve the problem of noise and uncertainty in the physical system interfering with the solution process.

[0029] It is important to emphasize that the hardware architecture of the SLM+DMD optoelectronic hybrid system and the core polarization correction algorithm in this embodiment constitute the main innovation of this application. In this architecture, fixed optical modulation information and state optical modulation information are carried collaboratively by SLM and DMD, respectively, thus forming a hardware implementation architecture suitable for performing optical measurement feedback updates. This provides the hardware foundation for the following algorithm steps of pre-selection, measurement, polarization correction, and freezing. The core polarization correction algorithm relies on this architecture to achieve technical effects against noise and uncertainty. The two work together to improve the solving performance of the optical ising machine.

[0030] Please see Figure 1 This illustrates the overall flow of the noise-resistant correction and update method provided in this application embodiment. The method searches for the optimal solution to the problem by repeatedly executing an iterative process including pre-selection, measurement, correction determination, and stability monitoring. Please also refer to... Figure 2 This is a schematic diagram of the optoelectronic hybrid optical Ising system used in this embodiment. The system mainly includes: a control computer 10, which serves as the core of the system control and is responsible for executing the algorithm logic of this embodiment; a light source 20, such as a light-emitting diode or broadband light source, for providing the illumination beam; two key optical modulation devices for loading optical information—a spatial light modulator (SLM) 30 and a digital micromirror device (DMD) 40; an optical system 50, composed of lenses and beam splitters, for guiding and converging the light path; and a single-pixel photodetector 60 for measuring the total light intensity. The control computer 10 is responsible for data processing, generating modulation information, controlling the hardware, and analyzing the measurement results. The optical system 50 is responsible for guiding, shaping, and combining the light path, ensuring that the beam can correctly interact with the spatial light modulator 30 and the digital micromirror device 40, and ultimately converging the resulting light field onto the detector. The single-pixel photodetector 60 converts the measured analog light intensity signal into an electrical signal and feeds it back to the control computer 10 for further processing. From the perspective of hardware division of labor, the fixed pattern loaded on the spatial light modulator 30 remains unchanged throughout the solution process, while the binary pattern loaded on the digital micromirror device 40 can be refreshed at high speed with the iteration process. The two perform their respective functions and work together to carry optical information, so that the problem mapping and state update are decoupled at the hardware level. This is an important aspect of the system architecture design of this embodiment.

[0031] Specifically, the method steps in this embodiment are as follows: First, in initialization step S101, the control computer 10 receives the problem to be optimized. In this embodiment, the problem is given in the form of a quadratic unconstrained bivariate optimization problem, which is typically expressed mathematically as an N×N coupling matrix J, where N is the number of spin variables in the problem. The Hamiltonian energy can be expressed as... ,in, It is the subscript index of the spin variable. Let be the spin variable, representing the first... The state of a spin particle It is the first One spin, Representing the The first spin and the second The coupling strength between the spins. The control computer 10 generates corresponding fixed optical modulation information based on the coupling matrix J. This information is encoded as a grayscale image in this embodiment, with pixel values ​​associated with the element values ​​of matrix J. Subsequently, the control computer 10 loads this grayscale image onto the spatial light modulator 30, and this pattern remains unchanged throughout the solution process. Simultaneously, the control computer 10 initializes a random spin state vector σ, representing an initial solution to the problem. Based on this initial spin state vector σ, the control computer 10 generates corresponding state optical modulation information, which is encoded as a binary image in this embodiment. Sites with a spin of +1 correspond to the "on" state of the micromirror, while sites with a spin of -1 correspond to the "off" state of the micromirror. This binary image is loaded onto the digital micromirror device 40, and its pattern dynamically changes with state updates during the iteration process. At this point, the initial state setting of the system is complete.

[0032] After initialization, the system enters an iterative loop, which repeats until a preset freeze condition is met. In each iteration, the following steps are performed: Step S102: Generate multiple candidate schemes. Unlike traditional methods that randomly select a spin to flip, in this step, the control computer 10 generates multiple, for example, 10, candidate spin-flipping schemes. Each scheme represents a spin flipping scheme for a specific spin in the current spin state vector σ. The spins are flipped, either from +1 to -1 or from -1 to +1. It should be noted that these candidate spins can be generated randomly or according to a specific strategy, such as selecting spins that have not been recently flipped.

[0033] Step S103: Calculate reliability indicators and perform pre-selection. This step is one of the core aspects of this application, and its concept can be found in [reference needed]. Figure 3 For each candidate spin-flip scheme generated in step S102, such as Figure 3 For candidate schemes C1…CN, the control computer 10 will construct a candidate state reliability index for them, such as… Figure 3 The reliability index R1…RN is used to assess the probability that a candidate state is reliably identified as superior to the current state under the current physical conditions of the system. In this embodiment, the reliability index R can be constructed based on the ratio of theoretical energy change to the system noise level, for example, using the following calculation formula: ,in, This represents the theoretical Hamiltonian energy change caused by the candidate flipping scheme. For the flipped... One spin, It can be quickly calculated by the control computer 10 by consulting the coupling matrix J and the current spin state σ, without the need for optical measurement. The system's estimated average noise level represents the typical fluctuation range of the measured value of a single-pixel photodetector 60. This value can be obtained by calibrating the system before the solution begins or by dynamically estimating it during iteration. Therefore, a higher R value indicates that the candidate flip theoretically causes a larger energy change, and the corresponding optical signal change is also larger, making it more likely to be clearly detected in a noisy background. After calculating the reliability index R of all candidate schemes, the control computer 10 sorts them and selects the candidate scheme with the largest R value as the sole "candidate state to be tested" for this round of iteration, such as... Figure 3 The middle arrow points to the pre-selection result. Through this pre-selection mechanism, the system can concentrate valuable physical measurement resources on candidate states that are most likely to produce effective signals, thereby avoiding invalid measurements on a large number of low signal-to-noise ratio states.

[0034] Step S104: Perform optical measurement. After determining the candidate state to be measured, the control computer 10 generates new state optical modulation information corresponding to the candidate state, that is, updates the binary pattern on the digital micromirror device 40, flips the state of the corresponding micromirror, and loads it onto the digital micromirror device 40. At this time, the light beam emitted by the light source 20 passes through the optical system 50 and interacts with the fixed pattern representing the problem matrix on the spatial light modulator 30 and the new pattern representing the candidate state to be measured on the digital micromirror device 40, respectively. The modulated total light field is finally converged by the optical system 50 onto the photosensitive surface of the single-pixel photodetector 60. The photodetector 60 measures the total light intensity at this moment and converts it into a measured value. The value is then transmitted back to the control computer 10.

[0035] Step S105 involves determining whether to accept the correction. This step is another core aspect of this application, and its principle can be found in [reference needed]. Figure 4 A conceptual diagram. The control computer 10 obtains the measured values. Instead of using it directly for decision-making, it is compared with the light intensity measurement value before the flip to obtain the measured light intensity change. At the same time, the control computer 10 is aware of the theoretical energy change of the candidate state to be tested. It is understandable that in an ideal optical ising machine, changes in light intensity are inversely proportional to changes in energy, i.e. .therefore, The sign of the symbol determines the trend of the theoretical optical response, such as... Figure 4 The theoretical response curve E_theory is represented here by energy, with the light intensity trend being opposite. The core of the correction acceptance judgment lies in simultaneously comparing the "theoretical expectation" and the "physical measurement". In this embodiment, the specific logic of the judgment includes: First, the measured change direction is consistent with the theoretical trend, that is... In other words, if theoretically the energy should decrease, that is... Therefore, the measured light intensity should increase, that is... This corresponds to Figure 4 In scenario a), the measured value curve E_measured shows a consistent trend with the theoretical response curve E_theory. If the directions are opposite, then as shown in scenario b), the flip is directly rejected. Secondly, the measured change amplitude is large enough to be distinguished from noise, i.e. ,in This is a threshold factor greater than 1, whose value can be adjusted according to the system noise level, for example, it can be set to 1.5 or other empirical values. This ensures that the observed changes are real signals, rather than random noise, such as... Figure 4 The random fluctuations caused by the noise floor N_floor in the data. This corresponds to... Figure 4 In scenario c), although the direction may be correct, the change is minor and still within the noise range, and should also be rejected. Only when both of the above conditions are met will the control computer 10 determine "acceptance".

[0036] Step S106: Update the state or perform a rejection operation. If the determination result of step S105 is "accept", the control computer 10 will formally update the current spin state vector σ, making it the accepted candidate state to be tested. Correspondingly, if the determination result is "disapproval", the flip is rejected, and the current spin state vector σ remains unchanged, that is, it is restored to the state before the start of this iteration.

[0037] Step S107: Perform stability monitoring. After each iteration, regardless of whether the state is updated, the control computer 10 monitors one or more metrics characterizing the stability of the solution process. In this embodiment, the stability metric is defined as the "state acceptance rate of the most recent 100 iterations". The control computer 10 maintains a window of length 100 to record the number of "acceptance" decisions in the most recent 100 iterations. As the system gradually converges to the vicinity of the optimal solution, most candidate flips will lead to an increase in energy, and therefore will be rejected by the "correction acceptance decision" step, resulting in a significant decrease in the state acceptance rate.

[0038] Finally, it is determined whether the freezing condition is met. The control computer 10 continuously checks the stability index calculated in step S107. The freezing condition in this embodiment can be constructed based on the stability index. For example, it can be set as follows: if the state acceptance rate is lower than a preset low threshold for several consecutive checks, the system is considered to have converged to a stable low-energy region, thus meeting the freezing condition. The number of checks, the check interval, and the threshold can all be adjusted according to the specific problem size and convergence characteristics.

[0039] Step S108: Freeze and output the results. Once the freezing condition is met, the iteration loop terminates. The control computer 10 will output the final solution. As a preferred option, the system will also record and update a "historical preferred state" throughout the iteration process, which is the spin state with the lowest Hamiltonian energy value encountered so far, corresponding to the problem to be optimized. During output, the system can simultaneously output the current final spin state and this historical preferred state for the user to choose from.

[0040] Through the closed-loop control process that integrates "pre-selection-measurement-correction-freezing" described above, the method in this embodiment can effectively combat physical noise and system drift, avoid wasting measurement resources on low-quality candidate solutions, correct erroneous measurement decisions caused by noise, and finally reliably lock and output high-quality solutions when the solution process tends to stabilize, thereby significantly improving the accuracy, stability and repeatability of the solution.

[0041] Example 2 As an optional implementation, this embodiment improves upon Embodiment 1, aiming to further optimize the construction method of the candidate state reliability index to adapt to optimization problems with specific structural characteristics, such as sparse graphs. The hardware system used in this embodiment is exactly the same as that in Embodiment 1, i.e. Figure 2 The overall process of the photoelectric hybrid system shown also follows... Figure 1 The framework shown. The main difference between this embodiment and Embodiment 1 lies in the specific method for calculating the candidate state reliability index in step S103.

[0042] In Example 1, the reliability index R only considers the ratio of the absolute value of the theoretical energy change to the noise level. However, in some problems, flipping a "core" node with many connections to other spins (i.e., a node with many non-zero terms in the corresponding row / column of the coupling matrix) may produce the same theoretical energy change as flipping a sparsely connected "edge" node. However, the former may have a more severe and complex impact on the entire system, and its non-ideal effects on optical response may be stronger. To provide more refined guidance for the search process, this embodiment introduces node degree information to construct a composite reliability index.

[0043] Specifically, in step S103 of this embodiment, for each candidate spin-flipping scheme, it is assumed that the flipped one is the first... For each spin, the reliability index R of its candidate state can be constructed based on theoretical energy changes, system noise levels, and node degree information. For example, it can be calculated using the following composite formula: in: and The definitions are the same as in Example 1, representing the absolute value of the theoretical energy change and the system noise level, respectively; It is the flipped spin The "degree" of the spin in the coupling matrix J is the degree of the spin. The relevant non-zero coupling terms, i.e. or The number of spins reflects how tightly connected the spins are in the problem graph structure; and These are two preset weighting coefficients used to balance the importance of the energy change term and the degree term, and For example, you can set Or other empirical values. It should be noted that these two weights can be adjusted according to the type of problem.

[0044] By adopting this composite reliability index, the control computer 10 has a richer decision-making basis when making preliminary selections. The system not only still prefers to select candidate flips with large theoretical energy changes and high signal-to-noise ratios, but also... This reflects, while also giving a lower degree of flipping, that is Smaller Larger spins have a certain priority.

[0045] The difference in the working process is reflected in the following: When the control computer 10 evaluates multiple candidate schemes C1…CN in step S103, it calculates the aforementioned composite index R for each scheme. For example, candidate scheme 1 is to flip a node with high energy change and high connectivity, with a high score in the first aspect and a low score in the second aspect; while candidate scheme 2 is to flip a node with medium energy change but very low connectivity, with a medium score in the first aspect but a high score in the second aspect. The final R value will combine these two aspects. After selecting the candidate scheme with the highest R value as the candidate state to be tested, the system proceeds to step S104 to perform optical measurement. The subsequent steps such as correction acceptance determination S105, state update S106, stability monitoring S107, and output freeze S108 are exactly the same as in Example 1.

[0046] The beneficial effect of this embodiment is that by introducing node degree as a consideration, the solution process can be guided along a smoother path. In the early stages of the solution process, prioritizing the flipping of "edge" nodes with smaller impact ranges for "fine-tuning" avoids drastic system state oscillations caused by prematurely flipping critical core nodes, making the energy reduction process more stable. For certain graph problems with specific structures, such as sparse graphs or graphs with community structures, this strategy may help the algorithm converge to a high-quality solution region more quickly. This method of constructing the candidate state reliability index is an effective extension of the basic idea in Embodiment 1, demonstrating that the index can be derived from... Related theoretical optical response change information and candidate flips corresponding to Technical features are composed of a set of related affected items.

[0047] Example 3 Alternatively, in another embodiment, the correction and acceptance decision step S105 in Embodiment 1 is improved. Embodiment 1 employs a deterministic "hard decision" logic, meaning that any decision is rejected if the preset conditions are not met. This embodiment introduces a probabilistic "soft decision" mechanism, making the acceptance or rejection decision somewhat random. Its idea is partly borrowed from the simulated annealing algorithm, aiming to enhance the algorithm's ability to escape local optima. The hardware system and overall process of this embodiment are consistent with Embodiment 1; the main difference lies in the specific implementation of step S105.

[0048] In Example 1, the boundaries of acceptance judgment are clear. However, in this example, the control computer 10 performs the following operations in step S105: First, the measured light intensity change is obtained in step S104. Then, the control computer 10 calculates a "theoretical-measured agreement parameter" C. This parameter aims to quantify the degree of agreement between the measurement results and theoretical expectations. Its calculation formula can be constructed based on the product of the measured light intensity change and the theoretical expectation, after noise normalization. For example, it can be defined as follows: Here, This represents the expected direction and magnitude of the theoretical light intensity change. Analysis of this formula shows that when the measured light intensity change... Direction and theoretical expectations When the signals are consistent, the product is positive, the consistency parameter C is positive, and the stronger the measured signal and the greater the theoretically expected change, the larger the C value. When the direction of the measured light intensity change is opposite to the theoretical expectation, the product is negative, C is negative, representing a serious discrepancy. The denominator contains... This serves as a normalization function, making the C value a dimensionless number related to the signal-to-noise ratio.

[0049] Next, based on the compliance parameter C, the control computer 10 calculates an acceptance probability. This probability can be constructed based on the conformity parameter C, for example, it can take the form of a logistic function: Where k is a positive proportionality coefficient used to adjust the steepness of the probability curve. From this formula, it can be seen that when the conformity parameter C is very large, Approaching 0, the probability of acceptance Approaching 1; when the conformity parameter C is 0, the acceptance probability is... The value is 0.5; when the compliance parameter C is negative, If it is a large positive number, the probability of acceptance is... It approaches 0. Thus, the probability of acceptance depends smoothly on the degree of consistency between theory and practice.

[0050] Then, the control computer 10 generates a random number r that is uniformly distributed in the interval [0, 1], and compares r with... To make the final decision: if If yes, it is judged as "accepted"; otherwise, it is judged as "disapproved".

[0051] Finally, for cases deemed "unacceptable," this embodiment can introduce a more complex processing strategy. Unlike the direct rejection in Embodiment 1, this embodiment can be configured such that, when deemed unacceptable, a rejection operation can be performed with a certain probability, i.e., the current state remains unchanged, while triggering a "retest" with a certain probability. The aforementioned probability ratio can be adjusted according to the system noise characteristics and the solution stage. Triggering a retest means that the control computer 10 will instruct the optical system to perform the optical measurement of step S104 again on the same candidate state to be tested, and re-perform the probabilistic determination of step S105 based on the new measurement results. This retest mechanism provides an opportunity for states with acceptable compliance that may have been wrongly rejected due to a single extreme noise interference.

[0052] The working process of this embodiment can be summarized as follows: After a measurement is completed, the system no longer simply answers "yes" or "no," but calculates an "acceptance probability" based on the degree to which the measurement result matches the theoretical expectation. Even if the measurement result does not perfectly match the expectation, for example, the signal amplitude is less than the noise threshold, but the direction is correct, as long as the deviation is not large and the C value is a small positive number, the system still has a certain probability of accepting the state. This "flexible" decision-making mechanism, especially in the early stages of solving the problem, helps the system to explore more extensively, allowing it to occasionally accept some states that "don't look good" but may actually lead to a better solution region, thereby effectively reducing the risk of prematurely falling into local optima.

[0053] Compared to the deterministic decision-making used in Example 1, the probabilistic correction decision-making strategy in this embodiment may find a better global solution when solving complex problems. It embodies the technical feature that the correction acceptance decision can be a probabilistic decision, and the acceptance probability depends on the degree of consistency between the theoretical optical response trend and the measured value. Simultaneously, the retesting process for non-acceptance cases also provides explicit support for the technical feature of "performing a retest" in the claims.

[0054] Example 4 The core noise-resistant correction and update method proposed in this application is not only applicable to the SLM+DMD optoelectronic hybrid incoherent architecture described in the foregoing embodiments, but can also be applied to coherent optical ising machines with different physical principles and hardware structures. A brief description is given below using a coherent optical ising machine as an example to demonstrate the universality of the method in this application.

[0055] Please see Figure 5This is a schematic diagram of a coherent optical ising system used in this embodiment. The system mainly includes a control computer 110, a laser 120 as a coherent light source, a single spatial light modulator 130, an optical system 150 composed of Fourier lenses and imaging lenses, and an area array camera 160, such as a CMOS or CCD camera. In this system, information is typically encoded in the phase and / or amplitude of light, while the calculation results are represented by the interference of the light field.

[0056] This embodiment also follows Figure 1 The overall method flow is shown, but the specific implementation of each step is adjusted to adapt to the physical characteristics of the coherent system.

[0057] In initialization step S101, the control computer 110 receives the coupling matrix J and the initial spin state σ. Unlike in Embodiment 1, here it is necessary to use a specific encoding scheme to uniformly generate a complex modulation pattern containing amplitude and phase information, which is then loaded onto a single spatial light modulator 130.

[0058] In the iterative loop, the core "pre-selection-measurement-correction" logic still exists, but the objects of operation have changed: Step S102: Generate candidate schemes. Similar to Example 1, the control computer 110 generates N candidate single spin-flip schemes.

[0059] Step S103, pre-selection. For each candidate flipping scheme, the control computer 110 can theoretically calculate and estimate the significance of the corresponding optical response change, and construct a candidate state reliability index accordingly; then, the control computer 110 also selects the candidate scheme with the highest reliability index as the candidate state to be tested.

[0060] In step S104, optical measurement is performed. The control computer 110 updates the complex modulation pattern on the spatial light modulator 130 according to the candidate state to be measured. The collimated coherent beam emitted by the laser 120 is incident perpendicularly on the spatial light modulator 130. The modulated beam passes through the optical system 150 and forms a "measured interference pattern" on the target surface of the area array camera 160. The area array camera 160 captures this complete two-dimensional image and transmits it back to the control computer 110 as the measurement result.

[0061] Step S105, correction acceptance determination. The control computer 110 obtains the measured optical response change based on the measured results before and after the flip, and compares it with the theoretical optical response trend of the candidate state to be tested; if the two are consistent in key features, it is determined to be "accepted", otherwise it is determined to be "unaccepted".

[0062] The logic of steps S106, S107, and S108 is basically the same as in Example 1. If a flip is accepted, the spin state σ is updated; otherwise, it remains unchanged. The stability monitoring index can be based on the state acceptance rate or on the recent interferogram energy, i.e., the rate of change of total light intensity. When the freezing condition is met, the current solution and / or the historical best solution are also output.

[0063] The description of this embodiment shows that although the details of the physical implementation have changed significantly—from measuring total light intensity to capturing interferograms, and from comparing numerical values ​​to comparing images—the core ideas of "pre-selecting high-quality candidate states by calculating reliability indices" and "making correction judgments by comparing theoretical expectations with physical measurements" proposed in this application have been fully preserved. Therefore, the method of this application has good universality and platform independence, and can improve the solution accuracy and stability for optical ising machines with various architectures.

[0064] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0065] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A noise-resistant correction and update method for an optical ising machine, characterized in that, The optical ising machine includes at least one spatial light modulator (SLM) and a digital micromirror device (DMD). The at least one spatial light modulator (SLM) is used to load fixed optical modulation information corresponding to the problem to be optimized, and the digital micromirror device (DMD) is used to load state optical modulation information corresponding to the current spin state. The method includes: Repeat the iterative steps until the preset freezing condition is met; the iterative steps include: Multiple candidate spin-flip schemes are generated, and a candidate state reliability index is constructed for each candidate spin-flip scheme. Based on the candidate state reliability index, multiple candidate states corresponding to the multiple candidate spin-flipping schemes are pre-selected to determine at least one candidate state to be tested. The optical modulation information of the state loaded on the digital micromirror device (DMD) is updated according to the candidate state to be tested, and optical measurement is performed on the candidate state to be tested to obtain the measurement value. The theoretical optical response trend of the candidate state to be tested is determined, and a correction acceptance decision is made by combining the theoretical optical response trend with the measured value; wherein, if the decision is to accept, the spin state is updated based on the candidate state to be tested; if the decision is to reject, at least one of the following processes is performed: The following options are available: reject the candidate state to be tested, restore the spin state to the state before the pre-selection in the current iteration, reduce the priority of the candidate state to be pre-selected in subsequent iteration steps, or perform a retest on the candidate state to be tested. During the iteration process, stability indicators are monitored, and the spin state with the lowest Hamiltonian energy value corresponding to the problem to be optimized is recorded and updated as the historical preferred state, wherein the preset freezing condition is determined based on the stability indicators. When the freezing condition is met, output the current spin state and / or the historical preferred state.

2. The method according to claim 1, characterized in that, The candidate state reliability index is constructed based on one or more of the following information: theoretical light intensity variation information, theoretical optical response variation information, affected item information, historical measurement statistics, noise characterization information, and device / optical path stability information.

3. The method according to claim 1, characterized in that, The correction acceptance decision considers at least one of the following: the consistency between the theoretical optical response trend and the measured value, and the distinguishability of the measured value relative to the noise background.

4. The method according to claim 1, characterized in that, The stability index includes at least one of the following: state acceptance rate within a predetermined window, rate of change of measured value, and state flip frequency.

5. The method according to claim 3, characterized in that, The correction acceptance decision is a probabilistic decision, and its acceptance probability depends on the degree of consistency between the theoretical optical response trend and the measured value.

6. A noise-resistant correction and update method for an optical ising machine, characterized in that, Applied to coherent optical ising machines, the method includes: Based on the problem to be optimized, fixed optical modulation information is generated, and state optical modulation information is generated based on the current spin state. The fixed optical modulation information and the state optical modulation information are jointly loaded onto at least one spatial light modulator. The iterative steps are repeated until the preset freezing condition is met.

7. The method according to claim 6, characterized in that, The fixed optical modulation information and the state optical modulation information are jointly loaded onto a single spatial optical modulator through time-division multiplexing, space-division multiplexing, code superposition, or partition loading.

8. A noise-resistant correction and update system for an optical ising machine, characterized in that, Includes a control computer, a light source, at least one spatial light modulator (SLM), a digital micromirror device (DMD), an optical system, and a photodetector; The at least one spatial light modulator (SLM) is used to load fixed optical modulation information corresponding to the problem to be optimized; the digital micromirror device (DMD) is used to load state optical modulation information corresponding to the current spin state; the light beam emitted by the light source interacts with the fixed optical modulation information and the state optical modulation information through the optical system, and is then detected by the photodetector to obtain a measurement value; the control computer includes: The pre-selection unit is configured to generate multiple candidate spin-flip schemes in each iteration, construct a candidate state reliability index for each candidate spin-flip scheme, and perform pre-selection based on the candidate state reliability index to determine at least one candidate state to be tested. The measurement control unit is configured to update the state optical modulation information loaded on the digital micromirror device (DMD) according to the candidate state to be measured, and to acquire the measurement value measured by the photodetector. The determination and update unit is configured to determine the theoretical optical response trend of the candidate state to be tested, and combine the theoretical optical response trend with the measured value to perform a correction acceptance determination. When the determination is accepted, the spin state is updated. When the determination is not accepted, at least one of the following processes is performed: rejecting the candidate state to be tested, restoring the spin state to the state before the pre-selection unit determined the at least one candidate state to be tested, reducing the priority of the candidate state to be tested being pre-selected in subsequent iterations, or instructing the measurement control unit to perform a retest on the candidate state to be tested. The monitoring and output unit is configured to monitor stability indices during the iteration process, record and update the spin state with the lowest Hamiltonian energy value corresponding to the problem to be optimized as the historical preferred state, and control the output of the current spin state and / or the historical preferred state when the stability indices meet the preset freezing conditions.

9. The system according to claim 8, characterized in that, The fixed optical modulation information loaded onto the at least one spatial light modulator (SLM) remains unchanged during the solution process, while the state optical modulation information loaded onto the digital micromirror device (DMD) changes dynamically with the spin state update during the iteration process; the photodetector is a single-pixel photodetector.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the noise-resistant correction and update method for the optical Ising machine as described in any one of claims 1 to 7.

Images