A method and system for autonomous control and optimization of quantum experiments

Abstract

The application relates to a method and system for autonomous control and optimization of quantum experiments. The method comprises: determining an experiment parameter set and an experiment strategy according to a target specification for a quantum experiment task; converting the experiment parameter set and the experiment strategy into hardware control instructions to perform at least one quantum experiment, and obtaining an observation value set containing quantum observation results and classical observation results; calculating a statistical estimator of a target function in the target specification and an uncertainty of the statistical estimator according to the observation value set; inputting the experiment parameter set, the observation value set, the statistical estimator and the uncertainty into an optimizer to update parameters, and obtaining an updated experiment parameter set and an updated experiment strategy; converting the updated experiment parameter set and the updated experiment strategy into hardware control instructions, and returning to perform at least one quantum experiment based on the hardware control instructions until a stopping criterion in the target specification is met. The method can improve the efficiency of quantum experiments.

Description

Technical Field

[0001] This application relates to the field of quantum information technology, and in particular to a method and system for autonomous control and optimization of quantum experiments. Background Technology

[0002] Quantum experiments are a fundamental means of developing quantum physics. They not only provide crucial support for a deeper understanding of the basic laws governing the microscopic world, but also lay the experimental foundation for cutting-edge technologies such as quantum computing, quantum communication, and quantum precision measurement. Typical quantum experimental systems (such as optical quantum computing platforms, superconducting quantum chips, and cold atom experimental systems) usually contain a large number of tunable components (such as lasers, waveform generators, displacement stages, and modulators). The success of the experiment highly depends on the precise setting and coordinated control of the parameters of these components.

[0003] However, the parameters in quantum experiments are strongly coupled, the system configuration of each parameter is difficult, and the quantum experiments are time-consuming, resulting in low efficiency. Summary of the Invention

[0004] Therefore, it is necessary to provide a method and system for autonomous control and optimization of quantum experiments that can improve the efficiency of quantum experiments, addressing the aforementioned technical problems.

[0005] Firstly, this application provides a method for autonomous control and optimization of quantum experiments, including:

[0006] Receive the target specifications for the quantum experiment mission, and determine the set of experimental parameters and experimental strategies based on the target specifications;

[0007] The experimental parameter set and experimental strategy are converted into hardware control instructions. At least one quantum experiment is carried out based on the hardware control instructions, and the set of observation values, including quantum observation results and classical observation results, is obtained from at least one quantum experiment.

[0008] Based on the set of observations, calculate the statistical estimator and uncertainty of the objective function in the target specification;

[0009] The experimental parameter set, observation set, statistical estimate, and uncertainty are input into the optimizer to update the parameters, resulting in the updated experimental parameter set and the updated experimental strategy.

[0010] The updated set of experimental parameters and the updated experimental strategy are converted into hardware control instructions, and then the system returns to perform at least one quantum experiment based on the hardware control instructions until the stopping criterion in the target specification is met.

[0011] Secondly, this application also provides a system for autonomous control and optimization of quantum experiments, including:

[0012] The task interface module is configured to receive target specifications for quantum experimental tasks.

[0013] The quantum experiment execution module is configured to determine the set of experimental parameters and experimental strategy according to the target specification; convert the set of experimental parameters and experimental strategy into hardware control instructions; perform at least one quantum experiment based on the hardware control instructions; and obtain a set of observation values ​​containing quantum observation results and classical observation results generated in at least one quantum experiment.

[0014] The data feedback module, which is connected in communication with the quantum experiment execution module, is configured to calculate the statistical estimate and uncertainty of the objective function in the target specification based on the set of observations.

[0015] The optimization module, which communicates with the quantum experiment execution module and the data feedback module, is configured to input the experimental parameter set, observation set, statistical estimate and uncertainty into the optimizer to update the parameters, obtain the updated experimental parameter set and the updated experimental strategy; convert the updated experimental parameter set and the updated experimental strategy into hardware control instructions, and return to perform at least one quantum experiment based on the hardware control instructions until the stopping criterion in the target specification is met.

[0016] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the method provided in the first aspect above.

[0017] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method provided in the first aspect above.

[0018] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps in the method provided in the first aspect above.

[0019] The aforementioned methods, systems, computer equipment, computer-readable storage media, and computer program products for autonomous control and optimization of quantum experiments, after obtaining a set of observations including both quantum and classical observations, calculate the statistical estimate and uncertainty of the objective function. This allows optimization decisions to be based on the statistical reliability of noisy observations, dynamically adjusting the balance between exploration and utilization based on uncertainty. Under sparse counting conditions, it intelligently avoids oversampling in spurious optimal regions caused by noise, thereby reducing the number of invalid experiments. The optimizer integrates the experimental parameter set, observation set, statistical estimate, and uncertainty to jointly update the experimental parameter set and experimental strategy, achieving integrated optimization of equipment control parameters and measurement configuration parameters. This overcomes the defect that fixed measurement configurations may lead to misjudgments of true performance and effectively reduces additional verification iterations caused by improper measurement configurations. The entire execution-computation-update process forms a closed loop and runs continuously until the stopping criterion is met, enabling online adaptive compensation for environmental drift and system state changes. This maintains the directionality and stability of the optimization search, avoiding the time-consuming manual recalibration required due to performance degradation, thus significantly reducing the total time of quantum experiments and improving their efficiency. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is an application environment diagram of a method for autonomous control and optimization of quantum experiments in one embodiment;

[0022] Figure 2 This is a flowchart illustrating a method for autonomous control and optimization of quantum experiments in one embodiment;

[0023] Figure 3 This is a flowchart illustrating the method for autonomous control and optimization of quantum experiments in yet another embodiment;

[0024] Figure 4 This is a flowchart illustrating the process of determining hardware control instructions in one embodiment;

[0025] Figure 5 This is a block diagram of a system for autonomous control and optimization of quantum experiments in one embodiment;

[0026] Figure 6 This is a structural block diagram of a two-dimensional Bell inequality violation verification experimental setup in one embodiment.

[0027] Figure 7 This is a schematic diagram illustrating experimental results showing the violation of the two-dimensional Bell inequality in one embodiment;

[0028] Figure 8 This is a block diagram of the system for autonomous control and optimization of quantum experiments in yet another embodiment;

[0029] Figure 9 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0031] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0032] Typical quantum experimental systems (such as optical quantum computing platforms, superconducting quantum chips, and cold atom experimental devices) usually contain a large number of tunable components (such as lasers, waveform generators, shift stages, and modulators). The success of the experiment highly depends on the precise setting and coordinated control of the parameters of these components. However, traditional experimental procedures have the following inherent bottlenecks:

[0033] 1. Parameter settings in quantum experiments are highly dependent on human experience and manual operation. This is specifically reflected in the following aspects: equipment control parameters that control the evolution of the quantum system (such as waveplate / polarizer angle, phase delay, pump power, and optical path displacement in optical platforms; microwave / RF pulse amplitude, frequency, phase, pulse width, and shaping parameters in superconducting / semiconductor qubit platforms; laser frequency detuning, light intensity / pulse duration, trap voltage, and magnetic field compensation in ion trap / neutral atom platforms), as well as measurement configuration parameters used to acquire observation data, such as the selection of the measurement basis / POVM (Positive Operator-Valued Measure), readout threshold, integration time / number of samplings, detector gate width, coincidence time window, measurement sequence order, and the number of measurement setup repetitions. There are numerous parameters that are interdependent, and the parameter tuning process often relies on the experience accumulated by the experimenters and repeated trial and error, making it difficult to obtain the globally optimal configuration within a reasonable time.

[0034] 2. Lack of real-time feedback and adaptive capability: The experimental environment (such as temperature and vibration) and the system's own state (such as component drift) will change dynamically. Traditional static parameter settings cannot compensate for this online, resulting in a decrease in system performance.

[0035] 3. Difficulty in system-level collaborative optimization: Quantum experimental systems are complex systems with strong coupling effects between parameters. Manually separating and optimizing individual parameters makes it difficult to find the global optimal solution, thus failing to fully utilize the system's performance.

[0036] 4. Calibration and verification are very expensive: To characterize the quality of quantum states and operational fidelity, complex measurement procedures such as quantum state tomography are often required, which consumes a lot of additional time and resources.

[0037] Traditional quantum experiments typically rely on manually setting control parameters, fixing measurement configurations, and offline data analysis. The optimization process is time-consuming and sensitive to environmental drift. While machine learning-based closed-loop optimization methods aim to accelerate parameter tuning, they generally suffer from the following problems:

[0038] 1. Only control parameters were optimized, without joint optimization of measurement configuration;

[0039] 2. Treat the computational target as a deterministic quantity and ignore the propagation of uncertainty under counting noise;

[0040] 3. It relies on platform-customized scripts and lacks cross-device hardware abstraction and compilation;

[0041] 4. Lack of transferability based on experimental memory / multi-fidelity models;

[0042] 5. The optimal sampling and stopping strategy under safety and resource constraints (power, temperature, number of samples) is imperfect.

[0043] Therefore, there is an urgent need for a general, autonomous solution applicable to multiple platforms and capable of achieving reliable optimization under sparse and noisy counting conditions. The method and system for autonomous control and optimization of quantum experiments provided in this application are applicable to scenarios in quantum computing, quantum simulation, and quantum precision measurement where multi-parameter joint and dynamic real-time optimization of quantum state preparation and quantum logic gate operations is required.

[0044] The method for autonomous control and optimization of quantum experiments provided in this application can be applied to, for example... Figure 1 In the application environment shown, the quantum platform 102 communicates with the server 104 via a network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated onto the server 104, or it can be located in the cloud or on another network server.

[0045] Server 104 can receive the target specifications for a quantum experiment task, determine the experimental parameter set and experimental strategy based on the target specifications, and convert the experimental parameter set and experimental strategy into hardware control instructions. Server 104 can send the hardware control instructions to quantum platform 102, so that quantum platform 102 can perform at least one quantum experiment based on the hardware control instructions. Server 104 can obtain the set of observations generated by quantum platform 102 in at least one quantum experiment, including quantum observation results and classical observation results. Server 104 can calculate the statistical estimate and uncertainty of the objective function in the target specifications based on the set of observations, and input the experimental parameter set, the set of observations, the statistical estimate, and the uncertainty into the optimizer to update the parameters, obtaining the updated experimental parameter set and the updated experimental strategy. Server 104 can convert the updated experimental parameter set and the updated experimental strategy into hardware control instructions, and send the hardware control instructions to quantum platform 102 again, so that quantum platform 102 can continue to perform at least one quantum experiment based on the hardware control instructions until the stopping criterion in the target specifications is met.

[0046] The quantum platform 102 may be, but is not limited to, at least one of multiphoton, superconducting, semiconductor, ion trap, and neutral atom array. The server 104 may be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server providing cloud computing services.

[0047] In one exemplary embodiment, such as Figure 2 As shown, a method for autonomous control and optimization of quantum experiments is provided. In the embodiments of this application, this method is applied to... Figure 1 Taking the server in the example, the explanation includes the following steps 202 to 210. Wherein:

[0048] Step 202: Receive the target specifications for the quantum experiment task, and determine the set of experimental parameters and experimental strategy based on the target specifications.

[0049] The quantum experiment task is the specific goal that the user expects to achieve through a quantum experiment. Its data form can be natural language or a structured task description file. For example, a quantum experiment task could be "preparing a maximally entangled state" or "calibrating a quantum logic gate to achieve a fidelity exceeding 99.9%". The quantum experiment task can serve as the starting point and final evaluation criterion for the entire autonomous optimization process. The goal specification is a concrete and executable definition of the quantum experiment task. For example, the goal specification can include definitions of experimental parameters, experimental strategies, objective functions, physical and resource constraints, and stopping criteria.

[0050] Experimental parameters are physical quantities that can be directly or indirectly controlled by hardware devices; they are decision variables in the optimization process. The data form of experimental parameters can be numerical (e.g., angle, voltage, time) or discrete category identifiers (e.g., measurement base number). For example, an experimental parameter can be the power of a laser, the phase of a microwave pulse, or the rotation angle of a waveplate. A set of experimental parameters is a group of experimental parameters that serve the same quantum experimental task and need to be jointly optimized. The set of experimental parameters can be a multidimensional vector, for example... ={parameter1, parameter2, ..., parameterN}, the data structure of the experimental parameter set represents the search space dimension of the optimization problem.

[0051] Experimental strategies can include control strategies and measurement strategies. Control strategies can include delays / logical sequences for issuing control commands to devices to save waiting time or for synchronous execution; different branches triggered based on different conditions, such as stopping the experiment, retrying, or skipping the current step; and the order or rules for using the same device when multiple tasks occupy it. Measurement strategies are measurement configurations and execution plans developed to acquire observations. An experimental strategy can be a sequence of instructions including sequence, configuration, and budget. Measurement strategies can at least include: under which measurement settings will observations be performed, the execution order of these settings, and the sampling resources invested in each setting (such as the number of measurements and integration time). Experimental strategies guide how to perform quantum experiments and how the data feedback module can efficiently and reliably acquire data to serve the computation of the objective function.

[0052] Optionally, the server can receive target specifications for a specific quantum experiment task submitted by the user through the task interface module. Based on the target specifications, the server can obtain the set of experimental parameters and experimental strategies. For example, the server can parse the target specifications and extract explicit definitions of experimental parameters and experimental strategies, thereby determining the set of variables (set of experimental parameters) and data acquisition scheme (experimental strategy) that need to be operated on in this optimization loop.

[0053] In some embodiments, the target specification may include the definition of experimental parameters and experimental strategies. The definition of experimental parameters. The definition of experimental parameters may include a set of experimental parameters. The dimensions, name of each parameter, type (continuous / discrete / categorical), value range / resolution / unit, and feasible region constraints are defined. The definition of experimental parameters may also include an initial set of experimental parameters. Experimental strategy The definition can include at least the set of measurement settings (such as measurement base / readout mode / tomography settings), the execution order of each measurement setting, and the measurement strategy such as the sampling budget (number of samples or integration time) corresponding to each measurement setting; the delay / logical order of control commands sent to the device to save waiting time for execution completion or to achieve synchronous execution; different branches triggered according to different conditions, such as stopping the experiment, retrying, or skipping the current step; and the control strategy such as the usage order or rules when multiple tasks occupy the same device. The definition can also include an initial strategy. The server can parse the received structured target specifications to obtain the initial set of experimental parameters and experimental strategies.

[0054] In some embodiments, the objective specification may further include physical and resource constraints, an objective function definition, and a stopping criterion. The physical and resource constraints may include equipment safety constraints (such as maximum power, maximum amplitude, temperature / current / voltage limits, mechanical travel limits, etc.) and resource budget constraints (such as total experiment time, total number of samples, computational resource limits, etc.); the objective function definition may be given by observation data. With parameters Calculate the objective function The method should specify whether it is maximization or minimization; for uniform expression, the maximization objective can be converted into a minimization form through symbol transformation or normalization; the stopping criteria may include the maximum number of iterations, the objective value improvement threshold, the objective uncertainty threshold (such as the confidence interval / confidence interval width), and optional robustness test rules, etc.

[0055] Step 204: Convert the experimental parameter set and experimental strategy into hardware control instructions, conduct at least one quantum experiment based on the hardware control instructions, and obtain a set of observation values ​​that include both quantum and classical observation results generated in at least one quantum experiment.

[0056] Hardware control instructions are those that can be directly recognized and executed by the underlying physical experimental equipment to drive it to complete specific actions or state settings. Hardware control instructions are used to achieve unambiguous "soft" to "hard" driving, ensuring that optimizer decisions can be accurately and safely translated into physical world operations. A quantum experiment is a process in which the server, based on a specific set of experimental parameters and experimental strategies, drives the hardware to complete a complete data acquisition cycle within the current iteration cycle. Each quantum experiment can generate a set of observational data based on the current parameter configuration, used to evaluate the performance of the current parameters and provide a basis for the next parameter update. For example, in a quantum state tomography task, after the server controls the equipment to prepare a quantum state, it sequentially performs projection measurements under multiple measurement bases such as "X-basis," "Y-basis," and "Z-basis" according to the experimental strategy. A quantum experiment refers to completing this entire set of preset data acquisition processes under different measurement bases, thereby obtaining sufficient statistical information to reconstruct the quantum state.

[0057] An observation set is a collection of measurement data collected from various detectors and sensors after one or more quantum experiments have been performed. The observation set can include both quantum and classical observation results. Quantum observation results are data that directly or indirectly reflect the state and properties of the quantum system itself, such as the timestamp of a detector "click" or the bits whose readout is "0" or "1". Quantum observation results are random results under the influence of quantum mechanical measurement axioms, and their statistical regularities carry information about the quantum state. Classical observation results are data that reflect the state of the experimental equipment itself, the actual values ​​of control parameters, or the macroscopic environmental conditions of the laboratory, such as continuous physical quantity measurements like temperature (K), voltage (V), current (A), power (W), mechanical displacement (mm), or the feedback value (deg) of an angle encoder. Classical observation results follow classical physical laws and are used to monitor system stability, calibrate execution errors, or explain drift in quantum observations.

[0058] In some embodiments, the set of observations It consists of measurement results from quantum state measurement units and / or classical parametric measurement units. The set of observations. This includes quantum observation results and / or classical observation results. Quantum observation results may include, but are not limited to: single-photon counting / coincidence counting, qubit readout strings, readout voltage / current sampling sequences, fluorescence counting, arrival time distribution, measurement statistics required for tomographic reconstruction, and survival probabilities at different sequence lengths in randomized benchmarking (RB). Classical observation results may include, but are not limited to: temperature, magnetic field, laser power, cavity length locking error signal, frequency drift, mechanical displacement feedback, and device state quantities (e.g., power amplifier output power, detector dark count rate).

[0059] For example, the server can convert the experimental parameter set and experimental strategy into hardware control instructions. The server can receive the experimental parameter set (such as a set of microwave pulse parameters) and experimental strategy (such as the measurement sequence and the number of repetitions for each setting) to be tested in the current iteration cycle from its internal AI (Artificial Intelligence) optimization module, and derive hardware control instructions based on these parameters. The server can then distribute these hardware control instructions to various hardware devices within the quantum platform to conduct at least one quantum experiment and obtain a set of observations for that experiment. This set of observations can include both quantum and classical observations. For instance, the server can distribute the hardware control instructions to hardware devices within the quantum platform, such as an AWG (Arbitrary Waveform Generator), oscilloscope, and motor controller. Upon receiving the hardware control instructions, these devices execute them synchronously to jointly complete at least one quantum experiment. For example, the AWG generates microwave pulses to manipulate qubits, and then at a set time, the acquisition card begins recording and reading the signals from the resonant cavity.

[0060] Step 206: Based on the set of observations, calculate the statistical estimator and uncertainty of the objective function in the target specification.

[0061] The objective function is used to quantify the success or failure of a quantum experimental task, and it may include a predefined mathematical expression or calculation rule. In some embodiments, the objective function... The evaluation function is a specific quantum experimental task and is pre-defined by the task interface module in the target specification. It can be a single indicator or a combination of multiple indicators, and constraint penalties can be introduced to handle security / resource constraints. For example, Can come from a set of observations The calculated results include interference visibility, entanglement metric, nonlocality index, gate fidelity, readout fidelity, signal-to-noise ratio, and / or quantum Fisher information; for the purpose of unified optimization, This can be defined as a scalar to be minimized (e.g., constructing a minimization form by negating the "maximization index" or by its difference from the target value). This application is not limited to a specific index itself, but rather aims to minimize the scalar under sparse and noisy observation conditions. The statistical estimates and uncertainties are modeled, and the closed-loop updates of parameters and measurement strategies are driven accordingly, thereby reducing experimental iterations and improving the confidence and robustness of the results.

[0062] A statistical estimator is an approximate calculation of the true value of a target function. Due to the inherent randomness and noise in quantum observations (such as single-photon counting), the set of observations obtained from a single or finite number of experiments cannot directly and accurately reflect the true value of the target function. It is necessary to statistically summarize and infer from the observation data in the original, noisy set of observations to calculate a reliable numerical estimate representing the level of the target function—the statistical estimator. A statistical estimator is not a definitive value, but rather the best estimate based on the current finite data sample. The uncertainty of a statistical estimator is a quantitative characterization of its reliability or error range. A statistical estimator can be a scalar (such as standard deviation or confidence interval width) or an interval (such as a 95% confidence interval). A statistical estimator reflects the estimation fluctuations caused by finite sampling and observational noise. The greater the uncertainty, the less reliable the estimate based on the finite data; the smaller the uncertainty, the more reliable the estimate.

[0063] Optionally, the server can determine the objective function from the target specification and calculate the statistical estimate and uncertainty of the objective function based on the set of observations. For example, the server can calculate the statistical estimate and uncertainty of the objective function based on the predefined objective function calculation method and the set of observations in the target specification.

[0064] Step 208: Input the experimental parameter set, observation set, statistical estimate and uncertainty into the optimizer to update the parameters, and obtain the updated experimental parameter set and the updated experimental strategy.

[0065] The optimizer automatically searches for and determines the parameters and strategies to be tried in the next round of experiments based on the current experimental results, driving the entire system to evolve towards optimizing the objective function. The optimizer is a configurable decision-making and update unit capable of selecting and executing any of the following update modes or combinations thereof based on factors such as the differentiability of the objective function, evaluation noise level, parameter dimensionality, and evaluation cost, to improve the search efficiency and convergence robustness of optimal parameters in quantum experiments. Optimizer update modes include, but are not limited to, gradient-free update modes (such as random search, coordinate / direction search, evolutionary strategies and covariance adaptation, genetic algorithms, and cross-entropy methods), gradient-assisted update modes, surrogate model-based update modes (such as Gaussian process regression models, radial basis function networks / multinomial response surface models, neural network models, and hybrid models based on physical priors or constraint embeddings), or mode switching and hybrid update modes. The optimizer can be a software module or algorithm instance. The input data received by the optimizer includes: a set of experimental parameters. Observation set and based on The calculated objective function The statistical estimators and their uncertainties; the optimizer output includes the updated set of experimental parameters. and the updated experimental strategy .

[0066] For example, the server can input the set of experimental parameters, the set of observations, statistical estimates, and uncertainties into the optimizer, so that the optimizer can execute a dynamic optimization search process. This optimization search process is essentially about finding points that can optimize the objective function in a potentially high-dimensional and complex space composed of device parameters and measurement strategies. The optimizer's decision-making not only considers how to change the system state (updating experimental parameters) but also simultaneously considers how to observe the system more effectively (updating experimental strategies), thereby achieving an improvement in global efficiency.

[0067] In some embodiments, the server can use an optimizer to fuse current input data with historical data to update its internal surrogate model (such as a Gaussian process model), which is a set of experimental parameters. +Experimental Strategy → An approximate mathematical model of the mapping relationship between the statistical estimate of the objective function and its uncertainty; then, based on this updated surrogate model, the optimizer uses acquisition functions, such as Expected Improvement (EI) and Upper Confidence Bound (UCB), to weigh "exploration" (exploring regions of high uncertainty) and "utilization" (deepening the current best-predicted region), thereby searching and recommending the next set of updated experimental parameters that are most promising for improving performance in the vast joint space of parameters and policies. and the updated experimental strategy During this process, the hardware abstraction layer in the server can also interact with the optimizer to pre-validate candidate parameters and ensure they do not exceed device safety constraints.

[0068] Step 210: Convert the updated set of experimental parameters and the updated experimental strategy into hardware control instructions, and return to perform at least one quantum experiment based on the hardware control instructions until the stopping criterion in the target specification is met.

[0069] The stopping criterion is used to determine when to terminate the iterative process for a quantum experimental task, ensuring that sufficiently reliable results are obtained while avoiding meaningless infinite loops. The stopping criterion may include at least one of various conditions such as reaching a set length of iteration steps, the width of the confidence interval being less than a threshold, or the resource budget being exhausted (the total experimental time or the total number of samples reaching the upper limit).

[0070] Optionally, the server can convert the updated set of experimental parameters and the updated experimental strategy into hardware control instructions, and return the processing based on the hardware control instructions to perform at least one quantum experiment, so as to carry out the next quantum experiment iteration loop until the stopping criterion in the target specification is met, at which point the quantum experiment can be terminated and the final optimized result can be obtained.

[0071] In some embodiments, such as Figure 3 As shown, the server receives the digitized target specification submitted by the user through the task interface module. The server parses the target specification and extracts information such as the feasible domain of parameters, initial parameter suggestions, and strategy templates. The server can directly adopt the initial set of experimental parameters and experimental strategies provided by the user based on the target specification. In addition, the server can also query the experimental memory bank to find historical task data similar to the current quantum experiment task and generate the initial set of experimental parameters through transfer learning. and experimental strategies The server can be based on a defined ( , The system receives hardware control instructions and performs at least one quantum experiment using these instructions, resulting in a set of observations that include both quantum and classical observations. The server calculates the objective function F according to the formula defined in the target specification, based on the set of observations. The statistical estimate and uncertainty of the objective function are obtained. The server can input the current iteration's (X, Y, statistical estimate, uncertainty) into the optimizer in the AI ​​optimization module. The optimizer integrates all information to make a decision and obtain an updated set of experimental parameters. and the updated experimental strategy The server will ( , As input for the next iteration, the process jumps back to the process of performing at least one quantum experiment based on hardware control instructions. For example, at the beginning or end of each iteration, the server checks the iteration counter, the improvement of the latest objective function estimate, and its uncertainty, comparing them with the preset stopping conditions in the objective specification. As soon as any stopping condition is triggered, the entire closed-loop cycle terminates, and the server outputs the final optimized parameter set and the corresponding best performance index.

[0072] The method for autonomous control and optimization of quantum experiments provided in this application significantly reduces experimental iterations and total time under sparse counting and environmental drift conditions through adaptive sampling and uncertainty-driven decision-making. It also reduces reliance on the experience of experimenters, lowering the operational threshold for quantum experiments. Simultaneously, the efficient optimization process reduces the number of experiments required to obtain optimal parameters, saving valuable equipment time and computational resources.

[0073] In the aforementioned method for autonomous control and optimization of quantum experiments, after obtaining a set of observations including both quantum and classical observations, the statistical estimate and uncertainty of the objective function are calculated. This allows optimization decisions to be based on the statistical reliability of noisy observations, dynamically adjusting the balance between exploration and utilization based on uncertainty. Under sparse counting conditions, it intelligently avoids oversampling in noisy spurious optimal regions, thereby reducing the number of invalid experiments. The optimizer integrates the experimental parameter set, observation set, statistical estimate, and uncertainty to jointly update the experimental parameter set and experimental strategy, achieving integrated optimization of equipment control parameters and measurement configuration parameters. This overcomes the defect that fixed measurement configurations may lead to misjudgments of true performance and effectively reduces additional verification iterations caused by improper measurement configurations. The entire execution-computation-update process forms a closed loop and runs continuously until the stopping criterion is met, enabling online adaptive compensation for environmental drift and system state changes. This maintains the directionality and stability of the optimization search, avoiding the time-consuming manual recalibration required due to performance degradation. Consequently, it significantly reduces the total time of quantum experiments and improves their efficiency.

[0074] In an exemplary embodiment, the experimental parameter set includes device control parameters and measurement configuration parameters; the device control parameters are used to control the preparation and evolution of the quantum system corresponding to the quantum experiment; the measurement configuration parameters are used to define the measurement method for the quantum system.

[0075] Among them, the set of experimental parameters Includes: equipment control parameters ( ): Parameters used to control the preparation and evolution of quantum systems in quantum experiments, such as pulse amplitude / frequency / phase / duration, bias voltage / current, laser detuning and intensity, optical element angles, displacement, etc.; measurement configuration parameters ( ): Used to define parameters for the measurement methods and readout settings of the quantum system corresponding to a quantum experiment, such as measurement basis selection and angle, readout threshold, integration window, coincidence time window, detector gate width, tomography settings, etc. A quantum system is the physical entity that carries quantum effects (such as superposition and entanglement) in a quantum experiment and its directly related control environment. The quantum system is the object of action of device control parameters and the object of observation of measurement configuration parameters.

[0076] For example, the optimizer is configured to work with a set of experimental parameters, including device control parameters and measurement configuration parameters. Optimization is performed to achieve joint control-measurement optimization. This joint control-measurement optimization is manifested in the optimizer's adjustment of the current set of experimental parameters in each iteration. (Including equipment control parameters and measurement configuration parameters) Update the set of valid iterative experimental parameters. The system is compiled and executed by the hardware abstraction layer, thereby jointly optimizing two types of parameters, namely "system evolution" and "observation acquisition", within the same closed loop to reduce misjudgments and additional iterations caused by improper measurement configuration.

[0077] In this embodiment, the equipment control parameters and measurement configuration parameters are incorporated into the same set of experimental parameters for joint optimization. This avoids the low information extraction efficiency or misjudgment caused by fixed and suboptimal measurement methods. As a result, the optimizer can evaluate the system performance based on more accurate and lower noise observation data, thereby finding the globally optimal equipment control parameters more quickly and improving optimization efficiency. This reduces the number of experimental iterations and the total time consumed.

[0078] In one exemplary embodiment, such as Figure 4 As shown, the processing of hardware control instructions is determined, that is, the experimental parameter set and experimental strategy are converted into hardware control instructions, including steps 402 to 406. Wherein:

[0079] Step 402: Obtain device description information of the hardware devices in the quantum platform, which is used to perform quantum experimental tasks.

[0080] A quantum platform is an integrated physical system comprised of a series of controllable hardware devices, constructed to perform specific quantum physics experiments. A quantum platform can include, but is not limited to, at least one of multiphoton arrays, superconducting qubits, semiconductor qubits, ion traps, and neutral atom arrays. Hardware devices are independent physical instruments or components with digital control interfaces that constitute the quantum platform. Device description information can include a set of structured metadata used to define the control and communication attributes of the hardware devices within the quantum platform. Device description information can include, but is not limited to, device identifiers (IDs), device types (e.g., “motor”, “AWG”, “detector”), a list of controllable channels (e.g., “channel1.angle”), parameter types (continuous / discrete) for each channel, value ranges and units (e.g., [0, 360] degrees), physical security constraints (e.g., maximum rotational speed, maximum power), communication interface type and address, device-specific timing characteristics, or instruction set templates.

[0081] Optionally, when the server initializes for a specific quantum experiment task, it can load the device description information corresponding to that quantum platform based on the quantum platform identifier specified by the task. For example, for an electrically driven rotary stage in the quantum platform, its device description information may include: {"type": "rotation_stage", "id": "HWP_Alice", "channels": {"angle": {"min": 0, "max": 360, "unit": "deg", "max_speed": 10}}, "interface": "GPIB", "address": "12"}.

[0082] Step 404: Based on the device description information, the experimental parameter set and experimental strategy are parsed into an intermediate representation that is independent of the hardware device.

[0083] Intermediate representations are standardized instruction abstractions used within the hardware abstraction layer, independent of specific hardware devices. An intermediate representation can include a series of structured instruction objects or data frames describing "what to do," rather than "how to do it specifically." For example, an intermediate representation could include "set the angle channel value of device HWP_Alice to 45.5 degrees" or "conduct continuous counting and acquisition of detector SNSPD1 for 10 seconds and trigger the synchronization signal Trig_Start." Based on intermediate representations, optimization logic and hardware drivers can be decoupled, allowing the same optimization algorithms and strategies to be applied to different quantum platforms without modification; only the device description information library and underlying compiler corresponding to the platform need to be changed.

[0084] For example, the server can parse the experimental parameter set and experimental strategy based on device description information to obtain an intermediate representation independent of the hardware device. For instance, the server can understand the meaning of each parameter and strategy item based on the device description information, and translate the experimental parameter set and experimental strategy into an intermediate representation within the hardware abstraction layer according to the device description information. The parsing process for the experimental parameter set and experimental strategy relies on the device description information for mapping, but the generated intermediate representation itself does not contain any driver details of any specific hardware device.

[0085] For example, in the experimental parameter set and experimental strategy, it is necessary to set the "Alice-side measurement base angle" to 30 degrees. The corresponding intermediate representation may include: {action: "SET", device: "HWP_Alice", channel: "angle", value: 30, unit: "deg"}. This instruction does not care whether HWP_Alice is a Newport or Thorlabs motor, nor does it care whether the instruction is sent via GPIB (General Purpose Interface Bus) or USB (Universal Serial Bus).

[0086] Step 406: If the intermediate representation passes the constraint verification, compile the intermediate representation into hardware control instructions for the hardware device.

[0087] Constraint verification is a process performed by the hardware abstraction layer (HAL) to audit the security and feasibility of the operations described in the intermediate representations before they are compiled into final hardware instructions. Its purpose is to prevent illegal or dangerous instructions from being issued to the hardware, protecting device security and ensuring the correctness of experimental logic. Constraint verification can include static constraints defined in the device description information (such as angle range and power limit) as well as dynamic and global constraints defined in the experimental task objective specifications (such as timing synchronization requirements between multiple devices and total experimental time budget).

[0088] Optionally, the server can perform constraint verification on the intermediate representation. For example, it can apply safety constraints to the parameter ranges, power / temperature / stroke, etc., of candidate parameters in the intermediate representation, and verify the timing constraints for multi-device synchronization, thereby obtaining constraint verification results. If the constraint verification result is successful, the server can compile the intermediate representation to obtain hardware control instructions for each hardware device. For instance, the server can send the compiled hardware control instructions, such as SCPI (Standard Commands for Programmable Instruments) strings and binary waveform files, to the corresponding hardware devices via physical interfaces such as GPIB, USB, and Ethernet. The hardware devices receive and execute these hardware control instructions, thereby changing the physical state or starting data acquisition.

[0089] In some embodiments, the server can read device description information (including controllable channels, communication interfaces, feasible domains, and timing constraints) through the hardware abstraction layer, and then (… , The intermediate representation is parsed and is independent of specific hardware devices (e.g., "Set channel - amplitude / frequency / phase", "Set rotary table angle", "Set readout gate width / threshold", "Set sampling count / integration time", "Trigger and synchronization", etc.). The server can perform constraint verification on the intermediate representation, such as verifying the parameter range of candidate parameters, safety constraints such as power / temperature / stroke, and multi-device synchronization timing constraints. When the intermediate representation passes the constraint verification, the server can compile the intermediate representation into hardware control instructions executable by each hardware device (e.g., SCPI instructions, waveform data, TTL (Transistor-Transistor Logic Trigger Sequence) trigger sequences, etc.) and send them to the executing hardware devices via interfaces such as GPIB / USB / Ethernet. Among them, parameters and strategies directly related to hardware device control and measurement configuration are included in the compilation and sending; parameters related to data post-processing (e.g., objective function weights) are used by the data feedback module and are not sent to the hardware devices.

[0090] In this embodiment, based on the device description information of the hardware devices in the quantum platform, the experimental parameter set and experimental strategy are parsed into intermediate representations independent of the hardware devices. When the intermediate representation passes the constraint verification, it is compiled into hardware control instructions for the hardware devices. This realizes a layered and secure hardware abstraction and compilation system. Only the device description information library and the underlying driver need to be replaced to seamlessly deploy to different quantum platforms. This greatly expands the applicability of the method, reduces the cost of redeveloping and optimizing software for each new experimental system, and achieves cross-quantum platform universality and portability.

[0091] In an exemplary embodiment, the method for autonomous control and optimization of quantum experiments further includes: updating the candidate parameters in the intermediate representation when the intermediate representation fails the constraint verification to obtain an updated intermediate representation; compiling the updated intermediate representation into hardware control instructions for the hardware device; and returning a constraint violation flag to the optimizer to guide the optimizer to update the parameters.

[0092] Here, candidate parameters are the specific instructions to be sent to the hardware device for execution in the intermediate representation, and the candidate parameters are derived from the experimental parameter set. and experimental strategies Constraint violation flags are used for closed-loop feedback to explicitly indicate to the optimizer that the parameters suggested in the previous iteration were not adopted because they did not comply with physical safety or system constraints. This serves as a learning signal to guide the optimizer to avoid similar infeasible regions in future iterations.

[0093] Optionally, the server can perform constraint verification on candidate parameters in the intermediate representation through the hardware abstraction layer. This may include checking whether each candidate parameter is within the range allowed by the hardware device, and whether the combination of multiple candidate parameters satisfies system-level synchronization timing or resource budget constraints. If the constraint verification fails, the server can perform feasibility correction on the candidate parameters in the intermediate representation, such as performing update processing like projection / pruning / resampling, to obtain an updated intermediate representation. The parameters in the updated intermediate representation satisfy the hardware's immediate safety constraints. For example, for an out-of-range amplitude value, the server automatically sets it to the maximum allowed value for that channel; for an invalid enumerated measurement basis selection, the server resets it to the default safe value. The server can compile the updated intermediate representation into specific hardware control instructions and issue them for execution, ensuring that the quantum experiment can continue without damaging the hardware device. The server can also return constraint violation flags to the optimizer. These flags can carry the type of violation and possible specific information (such as which parameter violated the rule, the difference between the suggested value and the boundary value) to guide the optimizer in updating parameters. For example, in subsequent iterations, the optimizer can adjust its search strategy based on constraint violation flags (e.g., updating the constraint priors of the surrogate model in Bayesian optimization), thereby exploring the feasible parameter space more intelligently, avoiding repeatedly proposing solutions that would lead to violations, and significantly improving the overall optimization efficiency.

[0094] In this embodiment, when the intermediate representation fails the constraint check, the candidate parameters are updated. This avoids equipment overload, damage, or even safety accidents caused by the optimization algorithm exploring infeasible regions, allowing the fully automated closed-loop optimization to run reliably for a long time. Furthermore, by returning constraint violation flags to the optimizer, an information feedback channel is established, enabling the optimizer to incorporate infeasible regions as prior knowledge into its model. In subsequent iterations, the optimizer can proactively avoid sampling similar infeasible regions, focusing more on exploring feasible parameter spaces, thus converging to the vicinity of the global optimum faster and shortening the overall optimization task time.

[0095] In an exemplary embodiment, the calculation of the statistical estimate and uncertainty of the objective function in the target specification based on the set of observations includes: determining a statistical summarization and post-processing method that matches the statistical characteristics of the original observation data in the set of observations; mapping the original observation data to intermediate statistics according to the statistical summarization and post-processing method; constructing a likelihood function based on the intermediate statistics through a preset observation model and parameterized model, and obtaining parameter estimates based on the likelihood function; and obtaining the statistical estimate and uncertainty of the objective function based on the parameter estimates and the objective function in the target specification.

[0096] Raw observation data refers to observation data directly generated by measurement units in quantum experiments (such as single-photon detectors, oscilloscopes, and temperature sensors) without any aggregation or inference. Statistical characteristics refer to the inherent probability distribution law followed by the raw observation data during its generation, describing the source, type, and amplitude of data noise. Statistical aggregation and post-processing methods are processing methods used to compress and refine massive, high-dimensional, and noisy raw observation data into a low-dimensional, stable, and more information-concentrated representation form based on the statistical characteristics of the raw observation data. Statistical aggregation and post-processing methods are used for data dimensionality reduction and feature extraction. Intermediate statistics are transitional statistical results obtained after statistical aggregation and post-processing of the raw observation data, which have not yet been directly substituted into the final objective function calculation. The data form of intermediate statistics can be scalars (such as mean), vectors (such as probability-frequency vectors), matrices (such as covariance matrices), or more complex structures (such as power spectra).

[0097] An observation model describes how intermediate statistics depend on the physical parameters to be estimated and random noise in the measurement process. An observation model can be a mathematical equation that relates the physical parameters, measurement settings, and the statistical expectation of the observed data. For example, an observation model can be expressed as "in the case of parameters..." Under these conditions, the probability of observing result k is P(k| A parameterized model is a mathematical description of a quantum system itself or its evolution, containing a series of parameters to be determined. Examples include the parameterized form of the density matrix of a quantum state and the parameterized form of the process matrix of a quantum gate. The likelihood function is constructed based on observational models and intermediate statistics obtained from actual observations, expressing the likelihood of the parameters to be estimated. The function of likelihood. The function value L( This indicates that when the parameter takes the value of Under the given assumptions, what is the probability of observing this set of data? Parameter estimates are numerical solutions to unknown parameters in a parameterized model, obtained by processing the likelihood function using optimization algorithms. Parameter estimates are an inference of the true state of the system and are the core output of the original observation data after a series of statistical inferences.

[0098] Optionally, the set of observations may include the raw observation data generated during this experimental iteration. The server can determine the statistical characteristics of the raw observation data and, based on these characteristics, determine a statistical summarization and post-processing method that matches them. For example, if it is determined to be a Poisson count, the "binning count - frequency calculation" process is selected; if it is determined to be Gaussian noise readings, the "moving average - mean and variance calculation" process is selected. The server can map the raw observation data according to the statistical summarization and post-processing method to obtain intermediate statistics. For example, in the photon counting example, the server can perform time binning to convert the timestamp sequence into a histogram; in the voltage reading example, the server can remove obvious outliers and then calculate the mean and standard error.

[0099] The server can invoke preset observation models and parameterized models, which can be loaded according to the target specifications during task initialization. The server can use intermediate statistics as input to the "observed data" and the estimated variables (denoted as vectors) defined in the parameterized model as input. Assuming that each observation is independent, the overall likelihood function L( ) is taken as an unknown, and the observation model is combined with the statistical characteristics of the original observation data to write out the probability distribution of each component of the intermediate statistic. ) is the product of all these probabilities, and the likelihood function L( Essentially, it's about the parameter to be estimated. A function whose value represents the value of a parameter taking a specific value. At that time, the server determines the likelihood of observing the current set of actual data. The server uses numerical optimization algorithms, such as gradient descent, Newton's method, or specialized maximum likelihood estimation algorithms, to find the likelihood function L( The parameter vector that reaches its maximum value is the parameter estimate, which is the parameter value that is statistically most likely to produce the current observation data. The server can calculate the uncertainty (such as standard deviation and confidence interval) of the parameter estimate based on the likelihood function curvature (Fischer information) around the parameter estimate or by using a resampling method (such as Bootstrap). The server can also calculate the statistical estimate of the objective function by substituting the parameter estimate into the formula of the objective function defined in the objective specification, and then calculate the uncertainty of the statistical estimate using the law of error propagation.

[0100] In some embodiments, the server can perform statistical summarization and post-processing on the raw observation data in the observation set to map high-frequency, noisy, or large-scale raw data into a low-dimensional statistical representation that can be used to solve the target problem, i.e., to obtain intermediate statistics. The server can use data post-processing methods that match the observation statistics to generate statistical estimates for solving the target problem.

[0101] The raw observation data may include, but is not limited to: event counts and coincidence counts, single / multiple measurement results, timestamp sequences, waveform sampling sequences, spectral data and their annotation information, etc. The statistical summarization and post-processing methods are matched to the statistical characteristics of the raw observation data, which may include, but are not limited to, Poisson statistics, binomial / multinomial statistics, Gaussian noise models, mixed distribution models and their extended models with drift or outliers. Correspondingly, the statistical summarization and post-processing may include, but is not limited to: bin counting, frequency / proportion calculation, mean / variance / covariance calculation, correlation function or cross-correlation function calculation, spectral estimation and peak extraction, peak position and width estimation, drift trend estimation, outlier removal, robust reweighting and denoising filtering, etc., to obtain intermediate statistics (e.g., histograms, probability-frequency vectors, moment estimators, covariance matrices, correlation spectra or power spectral densities, etc.).

[0102] Based on the physical principles and prior knowledge of the experiment, the server can match the inherent statistical model to different types of observation data. Based on the matched statistical characteristics, the server can aggregate and clean the raw observation data, mapping it to intermediate statistics that characterize the core features of the data. For example, for timestamp sequences, the server can perform bin counting to generate a histogram of event counts; for repeated measurement bit strings, the server can calculate the frequency / proportion of different results to obtain a probability frequency vector; for simulated waveform sampling, the server can calculate its mean / variance / covariance as moment estimators; for signals with time correlations, the server can calculate their correlation function or power spectral density.

[0103] The server can construct statistical estimates that match the target problem based on intermediate statistics. For example, the server can invoke observation models (theoretical models describing how parameters affect observations) and parameterized models (mathematical descriptions of unknown physical parameters) pre-defined according to the target specification. For instance, in quantum state tomography, the observation model is the probability formula of quantum states under the measurement basis, and the parameterized model is the parametric representation of the density matrix. The server can construct a likelihood function based on the observation model and the parameterized model, and obtain parameter estimates using methods such as maximum likelihood estimation, maximum a posteriori estimation, least squares estimation, robust estimation, or Bayesian a posteriori inference. Based on the parameter estimates and the objective function in the target specification, the server obtains the statistical estimate of the objective function and the uncertainty of the statistical estimate.

[0104] In this embodiment, by determining a statistical summarization and post-processing method that matches the statistical characteristics, the systematic bias or information loss introduced by improper data processing can be reduced; by constructing a likelihood function based on intermediate statistics and obtaining parameter estimates based on the likelihood function, more accurate parameter estimates with better statistical properties can be obtained under the same amount of data, thereby improving the accuracy and confidence of parameter and target estimates.

[0105] In one exemplary embodiment, performing at least one quantum experiment based on hardware control instructions includes: determining a scheduling decision for the hardware control instructions; performing at least one quantum experiment based on the hardware control instructions via hardware devices in a quantum platform if the scheduling decision is a hardware evaluation; and performing at least one simulated quantum experiment based on the hardware control instructions if the scheduling decision is a simulation evaluation.

[0106] In each optimization iteration, scheduling decisions are crucial. After generating a set of candidate experimental parameters and strategies, the optimizer must decide how to evaluate the actual effectiveness of these candidate schemes. Scheduling decisions can be either a binary choice or a proportional allocation, indicating whether the current iteration should perform hardware evaluation or simulation evaluation. Hardware evaluation involves controlling a real quantum experimental hardware system (such as a superconducting quantum bit chip or an optical interferometer) and making scheduling decisions to actually run one or more quantum experiments based on compiled hardware control instructions. The hardware evaluation process generates realistic observational data containing physical noise and system drift; this data is high-fidelity, high-cost, and directly reflects the real response in the physical world. Simulation evaluation, on the other hand, inputs the compiled hardware control instructions into a quantum system simulation model (surrogate model) to simulate the evolution and measurement process of the quantum system in the digital world and outputs scheduling decisions based on simulated observational data. The simulated observational data is low-fidelity and low-cost, and its accuracy depends on how closely the simulation model approximates the real physical system.

[0107] Optionally, the server can determine scheduling decisions for hardware control instructions. This can be based on a preset fixed strategy (e.g., "interspersing one simulation evaluation with every three hardware evaluations") or dynamic rules (e.g., "using simulation evaluation when the surrogate model's prediction uncertainty is below a threshold"). When the scheduling decision is hardware evaluation, the server can further compile the hardware control instructions into final instructions recognizable by specific hardware devices (e.g., arbitrary waveform generators, electric displacement stage controllers), such as SCPI commands and waveform arrays, and distribute them to the actual hardware devices of the quantum platform via communication interfaces (e.g., USB, Ethernet). The hardware devices execute these instructions to complete the preparation, evolution, and measurement of quantum states. The server can acquire raw observation sets Y (e.g., photon counts, voltage waveforms) from the real physical world through data acquisition cards, detectors, and other devices.

[0108] When scheduling decisions are made for simulation evaluation, the server can pass hardware control commands as input parameters to a pre-built simulation model. This simulation model runs in a digital environment, simulating the physical processes of the quantum system and generating a simulated set of observations Y in a computational virtual environment. (For example, generating simulated coincidence counts based on a Poisson distribution, or generating simulated readout signals based on a noise model).

[0109] In this embodiment, through intelligent scheduling decisions, simulation is used as much as possible to replace real hardware experiments while ensuring that the optimization direction is correct. This enables the intelligent mixing of high-cost, high-fidelity hardware experiments with low-cost, low-fidelity simulation calculations, significantly reducing hardware consumption and total experiment time, and accelerating the optimization process.

[0110] In one exemplary embodiment, the method for autonomous control and optimization of quantum experiments further includes at least one of the experimental parameter set and experimental strategy, which is obtained based on historical task data that matches the quantum experimental task.

[0111] Historical task data comprises the set of experimental parameters used in each iteration of past quantum experiments, the experimental strategies employed, the resulting set of experimental observations, and the statistical estimates and uncertainties of the calculated objective function. It may also include metadata describing the task background and experimental environment. Historical task data can provide prior knowledge and experiential references for new and unknown quantum experiments. By analyzing this historical task data, the server can understand the influence of different parameters on experimental results, thus avoiding completely random exploration starting from a "zero-knowledge" state in new tasks. Historical task data matching the quantum experiment task can be selected from a massive amount of historical task data, choosing those similar to the current quantum experiment task, such as those with similarities in task description and parameter feasible regions.

[0112] Optionally, the server can determine historical task data matching the quantum experiment task. For example, the server can access an experimental memory and, based on the description information of the quantum experiment task (such as the type of objective function and physical platform) and the feasible domain of parameters, perform similarity calculation and retrieval with the historical task data stored in the experimental memory to obtain historical task data matching the quantum experiment task. In some embodiments, the experimental memory is used for structured storage, version management, retrieval, and reuse of multi-source historical task data. Multi-source historical task data includes, but is not limited to: historical hardware experiment data, simulation experiment data, calibration data under different equipment / batch / environmental conditions, optimization iteration trajectory data, candidate parameter and evaluation result pairs, anomaly and failure mode annotations, equipment status and environmental status (including temperature, drift index, calibration version, noise level, etc.), and task identifiers and task feature descriptions. In some embodiments, the experimental memory supports retrieval by task similarity, equipment status similarity, parameter neighborhood, or time window to output candidate initialization, prior distribution, feasible domain boundary, policy template, or parameter constraint for subsequent tasks, thereby achieving cross-task migration and cold start acceleration. The server can determine at least one of the experimental parameter set and experimental strategy based on historical mission data that matches the quantum experimental mission.

[0113] In this embodiment, at least one of the experimental parameter set and experimental strategy is determined based on historical task data that matches the quantum experimental task. This avoids completely random exploration starting from a "zero-knowledge" state in a new task and reduces the number of real hardware experiment iterations required to achieve the same optimization goal, thereby significantly shortening the completion time of the entire optimization task.

[0114] In an exemplary embodiment, the method for autonomous control and optimization of quantum experiments further includes: the optimizer includes a surrogate model for parameter updates, wherein at least one of the hyperparameters or prior distribution parameters of the surrogate model is initialized based on historical task data that matches the quantum experiment task.

[0115] In this context, the surrogate model is a mathematical model used by optimizers to quickly predict and compare the performance of a large number of candidate parameters, thereby guiding parameter updates. The surrogate model allows the optimizer to perform rapid, low-cost virtual testing and comparison of thousands of candidate parameters without conducting real experiments, intelligently guessing which parameters are more promising and making update decisions accordingly, resulting in an updated set of experimental parameters and an updated experimental strategy. Surrogate models can include various specific machine learning or statistical model structures. For example, a surrogate model can include a Gaussian process regression model, whose data structure includes a mean function and a covariance function (kernel function), capable of simultaneously providing the predicted value and its uncertainty (variance). Another example is a neural network model, whose structure consists of multiple layers of neurons, weights, and activation functions, learning the nonlinear relationship between input and output through training. Yet another example is the multinomial response surface model, which uses a multinomial function to fit the relationship between parameters and the target.

[0116] Hyperparameters are a set of parameters that need to be pre-defined manually or by an algorithm before training a surrogate model begins. The specific content of hyperparameters varies depending on the surrogate model. For example, for a Gaussian process model, hyperparameters may include the type of kernel function, such as the radial basis function (RB), length scale, signal variance, etc., which determine the smoothness and range of variation of the function space. For neural networks, hyperparameters may include the number of network layers, the number of neurons per layer, the learning rate, regularization coefficients, etc. Prior distribution parameters describe the initial beliefs or assumptions about the unknown parameters of the surrogate model (or the objective function itself) before any new experimental data is observed; prior distribution parameters define the shape and characteristics of the prior distribution.

[0117] For example, the optimizer may include a surrogate model with prior knowledge. When the optimizer needs to generate a new set of experimental parameters and experimental strategies for the next iteration, it can rely on the surrogate model for prediction and decision-making. For instance, the optimizer can generate a large number of candidate points in the parameter space using optimization algorithms (such as Bayesian optimization). For each candidate point, the optimizer can query the surrogate model to obtain the predicted statistical estimate and the predicted uncertainty. The optimizer can then synthesize the predicted statistical estimates and uncertainties of all candidate points to determine the updated set of experimental parameters and the updated experimental strategy. At least one of the hyperparameters and prior distribution parameters of the surrogate model is initialized based on historical task data that matches the quantum experimental task. For example, the surrogate model can learn from historical task data to construct a mapping from the set of experimental parameters to the predicted statistical estimates and their uncertainties, serving as a rapid simulator for real physics experiments. Since the surrogate model has already been preliminarily shaped by historical data, its predictions of the new task parameter space will be more accurate, thus guiding the optimizer to perform more effective parameter updates and reducing blind exploration.

[0118] In this embodiment, the hyperparameters or prior distribution of the proxy model are initialized based on similar historical task data. The model has a relatively reasonable initial approximation of the relationship between the parameter space and the objective function at the beginning of optimization, which makes the candidate parameters generated by the optimizer of higher quality and more likely to be close to the optimal region. This avoids a large number of hardware experiment iterations used to explore ineffective regions, achieves cold start acceleration, significantly improves the optimization convergence speed, and reduces the number of hardware experiment iterations.

[0119] In one exemplary embodiment, the method for autonomous control and optimization of quantum experiments further includes: the optimizer includes a surrogate model for parameter updates, the surrogate model being trained based on historical task data that matches the quantum experiment task.

[0120] Optionally, the surrogate model is trained based on historical task data that matches the quantum experimental task. The surrogate model allows the optimizer to perform fast and low-cost virtual testing and comparison of thousands of candidate parameters without performing real experiments, thereby intelligently guessing which parameters are more promising and making updated decisions accordingly, resulting in an updated set of experimental parameters and an updated experimental strategy.

[0121] In this embodiment, the proxy model is trained based on similar historical task data and can perform rapid virtual evaluation, enabling the optimizer to accurately nominate more promising candidate parameters, significantly filter out inefficient or invalid parameter attempts, achieve cold start acceleration, significantly improve optimization convergence speed, and reduce the number of hardware experiment iterations.

[0122] In an exemplary embodiment, the method for autonomous control and optimization of quantum experiments further includes: the optimizer includes a proxy model for parameter updates, and after obtaining current task data through at least one quantum experiment, the proxy model is incrementally trained based on the current task data.

[0123] The current task data refers to the set of experimental data directly related to the current task objective, generated and collected in real-time by the closed-loop system during the execution of this quantum experiment optimization task. The current task data can be a complete snapshot of the current experimental state. For example, the current task data may include a "parameter-observation-evaluation" triple, such as the set of experimental parameters output by the optimizer in this iteration, the set of experimental observations obtained after executing these parameters on hardware, and the calculated statistical estimate of the objective function and its uncertainty. Incremental training is the process of locally adjusting and optimizing the surrogate model based on the existing surrogate model using the newly acquired current task data.

[0124] For example, after the closed-loop system completes a full "execution-measurement-computation" cycle, the server can obtain the current task data and perform incremental training on the surrogate model based on this data. After training, the updated surrogate model parameters are saved and immediately published to the optimizer. When the optimizer needs to generate a new set of experimental parameters and experimental strategies, it can call the updated surrogate model to predict the performance of different candidate parameters or calculate the uncertainty of the predictions, thereby making more informed exploration and utilization decisions.

[0125] In this embodiment, the surrogate model can quickly fit the real response of the current experimental system through incremental training, which helps to reduce invalid or suboptimal explorations caused by model errors, significantly reduces the number of hardware iterations, and improves optimization efficiency.

[0126] In an exemplary embodiment, the method for autonomous control and optimization of quantum experiments further includes: the optimizer includes a proxy model for parameter updates, simulation data and real hardware experimental data of the quantum experimental task are acquired, and the proxy model is jointly trained based on the simulation data and real hardware experimental data.

[0127] Simulation data refers to data generated by running theoretical or computational models of quantum systems in a digital environment to simulate experimental processes. Simulation data does not originate from real physical devices but rather from predictions derived based on physical laws (such as the Schrödinger equation), device characteristic models (such as noise and distortion models), and the logical calculations of the experimental process. Simulation data can serve as a low-cost, high-speed "low-fidelity" information source for preliminary exploration of the parameter space and pre-training of surrogate models, thereby reducing reliance on expensive and time-consuming real hardware experiments. Real hardware experimental data consists of raw data and derived statistics directly collected and fed back by measuring instruments after controlling actual quantum experimental devices (such as lasers, arbitrary waveform generators, and detectors) to perform physical experiments. Real hardware experimental data can serve as a high-cost, high-reliability "high-fidelity" information source and is the ultimate basis for optimization objectives. Real hardware experimental data can be used to correct deviations in simulation models, make crucial corrections to surrogate models, and verify the true performance of optimization results.

[0128] Optionally, the server can acquire simulation data and real hardware experimental data for quantum experimental tasks. Simulation data originates from its internal quantum system simulation program or external simulation services, while real hardware experimental data comes from real data in closed-loop control. The server's AI optimization module can maintain a surrogate model (e.g., a Gaussian process model or a neural network model), whose goal is to learn the mapping relationship from experimental parameters X to the objective function F (or observed values ​​Y). Joint training refers to the server using both the acquired simulation data and real hardware experimental data as a training dataset to update the same surrogate model. During training, the server can employ a hierarchical modeling approach, assigning different noise levels or confidence weights to data of different fidelities; alternatively, it can use a weighted fusion approach, allowing high-fidelity real data to have a greater impact on model parameters during model updates, while low-fidelity simulation data is used to shape the overall trend of the model in unexplored regions. The trained surrogate model can be used for parameter update decisions; when the optimizer in the server needs to determine the next set of experimental parameters, it can query this surrogate model. The surrogate model can predict statistical estimates and uncertainties for different sets of experimental parameters and experimental strategies based on learned mapping relationships. The optimizer can use this predictive information (such as selecting points with expected good performance or high uncertainty) to generate the set of experimental parameters and experimental strategies for the next iteration, thereby achieving intelligent guidance for hardware experiments.

[0129] In this embodiment, low-cost generated simulation data is used for extensive pre-training and exploration of the surrogate model, enabling the surrogate model to possess a certain performance prediction capability before guiding real experiments. During real hardware iterations, the server can more intelligently select parameter points that are more promising or contain the most information for experiments, avoiding numerous invalid or inefficient attempts, significantly reducing the number of hardware iterations, and saving experimental resources and time.

[0130] In one exemplary embodiment, the optimizer can utilize experimental memory and / or a multi-fidelity proxy model to migrate between historical experiments and simulations to initialize parameters and reduce the number of hardware trials. In some embodiments, the experimental memory is used to store sample data and metadata of historical tasks. When a new task is received, the optimizer retrieves a set of samples of similar historical tasks from the experimental memory based on the task description and parameter feasible domain, for:

[0131] (1) Generate the initial set of experimental parameters and / or experimental strategies;

[0132] (2) Initialize the hyperparameters / prior mean of the proxy model;

[0133] (3) Use historical samples as low-cost prior data and retrain them after new task data arrives.

[0134] In some embodiments, the multi-fidelity proxy model uses simulation data as low-fidelity data and real hardware experimental data as high-fidelity data, and updates them jointly through hierarchical modeling or weighted fusion, thereby reducing the number of costly hardware iterations while maintaining accuracy. For example, in repetitive calibration tasks with the same equipment, several candidate points can be generated within the feasible region based on historical optimal parameters and simulation predictions, and then verified on the hardware with fewer iterations and converged quickly.

[0135] In some embodiments, the surrogate model can be trained, updated, or fine-tuned based on historical and / or real-time data in the experimental memory to form an approximate representation of the mapping relationship between the experimental system and the objective function. The trained surrogate model (including model parameters, hyperparameters, and / or uncertainty calibration parameters) is then published to the optimizer for use in the surrogate model-based update mode. Training methods include, but are not limited to, transfer learning, incremental learning, online updates, and joint training or phased training using simulation data and hardware data. Preferably, the surrogate model training module supports online updates of the surrogate model according to a preset period, by the number of iterations, by drift detection triggering, or by uncertainty threshold triggering, to maintain the effectiveness of the surrogate model for the current experimental state.

[0136] The method for autonomous control and optimization of quantum experiments provided in this application, based on an AI-powered optimizer, is independent of the mathematical structure or prior knowledge of any specific experimental task. By changing the objective function F, the same system and method can be widely applied to various scenarios such as quantum state preparation, quantum gate calibration, and qubit readout. It is applicable to different physical platforms, including optics, superconductivity, and atomic physics, demonstrating strong versatility, universality, and transferability. Moreover, the optimizer can handle complex nonlinear coupling relationships between parameters and find the global optimum, thereby fully realizing the performance potential of the quantum experimental system. Simultaneously, this closed-loop system can respond to environmental disturbances and system drift in real time, automatically compensating for parameter deviations, significantly enhancing the system's stability and robustness, and improving its performance and robustness.

[0137] This application also provides an application scenario in which the above-described method for autonomous control and optimization of quantum experiments is applied. Specifically, this method is applied to the autonomous optimization of experiments violating Bell's inequality, and its application in this scenario is as follows:

[0138] like Figure 5 As shown, this application provides an autonomous two-dimensional Bell inequality violation verification experimental device. In this embodiment, the system for autonomous control and optimization of the quantum experiment specifically includes: a quantum experiment execution module, a task interface module, a data feedback module, and an optimization module, which together form a closed-loop control circuit.

[0139] (I) Quantum Experiment System and Quantum Experiment Execution Module

[0140] like Figure 6 As shown, the quantum experimental system includes a first laser source 1, a first polarization beam splitter 2, a first half-wave plate 3 with an electrically driven rotating displacement stage, a first beam deflector 4, a second half-wave plate 5, a third half-wave plate 6, a first parametric down-conversion crystal 7, a fourth half-wave plate 8, a fifth half-wave plate 9, a second beam deflector 10, a sixth half-wave plate 11 with an electrically driven rotating displacement stage, a second polarization beam splitter 12, a first single-photon detector 13, a second single-photon detector 14, a seventh half-wave plate 15, an eighth half-wave plate 16, a third beam deflector 17, a ninth half-wave plate 18 with an electrically driven rotating displacement stage, a third polarization beam splitter 19, a third single-photon detector 20, and a fourth single-photon detector 21. The electrically driven rotating displacement stage controllers of the first half-wave plate 3, the sixth half-wave plate 11, and the ninth half-wave plate 18 serve as quantum experimental devices for executing experimental parameter sets, constituting a quantum experimental execution module.

[0141] The first laser source 1 is used to generate continuous pump light with a center wavelength of 775 nm.

[0142] The first polarization beam splitter 2 is used to polarize the pump light to generate linearly polarized pump light with a fixed polarization direction.

[0143] The first half-wave plate 3 with an electrically driven rotary displacement stage is used to change the polarization direction of the linearly polarized pump light;

[0144] The first beam deflector 4 is used to split the linearly deflected pump light into two spatially separated parallel pump lights in the horizontal and vertical directions.

[0145] The second half-wave plate 5 and the third half-wave plate 6 are used to adjust the polarization direction of the two pump beams so that their beam polarization direction is consistent with the phase matching angle direction of the first parametric down-conversion crystal 7, thus satisfying the non-collinear type II phase matching condition.

[0146] After the first parametric downconversion crystal 7 is pumped, each 775nm pump light generates a pair of correlated photons with a center wavelength of 1550nm through a spontaneous parametric downconversion process. The two pump lights generate two sets of correlated photons, in which the signal photons are deflected in the same direction and remain parallel, while the idler photons are deflected in the other direction and remain parallel.

[0147] The fourth half-wave plate 8 and the fifth half-wave plate 9 are used to modulate the polarization direction of the signal photons, so that the paths of the two signal photons are merged into one beam after passing through the second beam deflector 10.

[0148] The combination of the sixth half-wave plate 11 with an electric rotary displacement stage, the second polarization beam splitter 12, the first single-photon detector 13, and the second single-photon detector 14 serves as a signal photon state projection measurement, with the projection measurement basis determined by the direction of the sixth half-wave plate 11 with an electric rotary displacement stage.

[0149] The seventh half-wave plate 15 and the eighth half-wave plate 16 respectively modulate the polarization direction of the intermediate frequency photons of the two sets of correlated photon pairs, so that the paths of the two intermediate frequency photons are merged into one beam after passing through the third beam deflector 17.

[0150] The combination of the ninth half-wave plate 18 with an electric rotary displacement stage, the third polarization beam splitter 19, the third single-photon detector 20, and the fourth single-photon detector 21 serves as an idler photon state projection measurement, with the projection measurement basis determined by the direction of the ninth half-wave plate 18 with an electric rotary displacement stage.

[0151] (ii) Task Interface Module

[0152] The task interface module is computer 24, which is used to receive the definition of the objective function, the rotation angle range of the electric rotary displacement stage, the iteration upper limit, and optimization parameters.

[0153] (III) Data Feedback Module

[0154] The data acquisition module includes a quantum state measurement unit and a classical parameter measurement unit.

[0155] Among them, the quantum state measurement unit is the first time-to-digital converter 22, which is used to record the response results of the first single-photon detector 13, the second single-photon detector 14, the third single-photon detector 20, and the fourth single-photon detector 21 in the quantum experimental system, and to calculate the coincidence count.

[0156] The classic parametric measurement unit records the actual angle values ​​of the first half-wave plate 3 with an electric rotary displacement stage, the sixth half-wave plate 11 with an electric rotary displacement stage, and the ninth half-wave plate 18 with an electric rotary displacement stage after the execution of the rotation command, which are used to calibrate the equipment parameter set X.

[0157] The optimization module is computer 23, which communicates with the electric rotary displacement stage and data feedback module in the quantum experiment execution module via a GPIB / USB interface.

[0158] The autonomous experimental steps for the method of autonomous control and optimization in quantum experiments are as follows:

[0159] (Initialization): Defines the target specifications for this experiment, including:

[0160] Five-dimensional experimental parameter set X={ , , , , The elements in the set correspond sequentially to the first half-wave plate 3 and the sixth half-wave plate 11 with an electrically driven rotary displacement stage. / Measurement base), Ninth half-wave plate 18 ( / The angle value of the measuring base is specified as the rotation angle range of the electric rotary displacement stage as [-360°, 360°].

[0161] The objective function is F = -(P(00|00) - P(00|01) - P(00|10) - P(11|11));

[0162] The stopping criterion is 400 iterations;

[0163] The optimization module (computer 23) generates a set of five-dimensional initial equipment parameters X0={ (0), (0), (0), (0), (0)} and experimental strategies The specific experimental strategy is as follows: Measurement basis combinations are set sequentially. Measure the photon coincidence counts Ca0b0, Ca0b1, Ca1b0, Ca1b1 under different bases, and then calculate the probability P(ab|xy) (e.g., when the measurement base is...). At that time, P(a0b0|x0y0)=Ca0b0 / (Ca0b0+Ca0b1+Ca1b0+Ca1b1)).

[0164] (Experiment Execution): The quantum experiment execution module receives the experimental parameter set X and controls three electrically driven rotary displacement stages to rotate to a specified angle. Subsequently, the system combines four measurement bases ( Single-photon coincidence count measurements were performed for 10 seconds each.

[0165] (Acquisition and Calculation): The data feedback module (first-time digital converter 22) acquires the coincidence count results and calculates the objective function F based on these results. After the calculation is completed, the value of F is fed back to the optimization module.

[0166] (Optimization Decision): The optimization module receives the current set of equipment parameters X, experimental observations (symbol counts), and the objective function value F. Its built-in parameter-based optimizer uses this data as input to generate a new set of optimized experimental parameters X'={...} that is expected to minimize the value of F. When generating new parameters, the algorithm must satisfy physical constraints to ensure that all waveplate angles... ∈[0°,360°].

[0167] (Iterative Update): The optimization module takes X' as the new set of experimental parameters X and repeats steps S2 to S4.

[0168] (Convergence Criterion): After multiple iterations (usually 150-300 times), when the measured value of the objective function F stabilizes around -0.19 (the classical limit is F=0), i.e., when the Bell inequality violation is maximized, the system is considered to have converged, and the optimization process ends. Figure 7 The figure shown is a schematic diagram of the experimental results of the autonomous two-dimensional Bell inequality violation.

[0169] Specific component selection includes:

[0170] Laser source 1 is a 775nm continuous-pump laser.

[0171] The parametric downconversion crystal 7 is a barium β-borate crystal with an AR (Anti-Reflective Coating) coating (R<0.5%@775 / 1550nm).

[0172] The electric rotary stage is an integrated electric rotary stage (Newport, CONEX-PR50CC, 360° travel range, accuracy ±25mdeg, bidirectional repeatability ±75mdeg).

[0173] Single-photon detectors 13, 14, 20, and 21 are superconducting nanowire single-photon detectors (detection efficiency ≥98%@1550nm, dark count <50Hz).

[0174] The single iteration cycle time T of the closed-loop control loop is approximately 2 seconds, satisfying the requirement of T≤10 seconds. Specifically, the delay in data acquisition and objective function calculation is less than 0.1 seconds, and the AI ​​optimization calculation delay is less than 0.5 seconds.

[0175] Optionally, the β-barium borate crystal 7 can be replaced by parametric downconversion crystals such as lithium triborate or periodically polarized potassium titanyl phosphate.

[0176] Optionally, the first laser source 1 can be selected from different working wavelengths, such as solid-state laser source, gas laser source, semiconductor laser source or dye laser source, according to the working wavelength of the parametric downconversion crystal.

[0177] Alternatively, single-photon detectors 13, 14, 20, and 21 can be replaced by semiconductor avalanche detectors or superconducting transition edge single-photon detectors.

[0178] Optionally, the algorithm used in the optimization module can be replaced with a neural network-based model or a reinforcement learning algorithm.

[0179] Optionally, the objective function F can also be set to other performance indicators such as quantum state fidelity and interference visibility to suit different quantum experimental tasks.

[0180] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0181] Based on the same inventive concept, this application also provides a system for implementing the above-described method for autonomous control and optimization of quantum experiments. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more system embodiments for autonomous control and optimization of quantum experiments provided below can be found in the limitations of the method for autonomous control and optimization of quantum experiments described above, and will not be repeated here.

[0182] In one exemplary embodiment, such as Figure 8 As shown, a system 800 for autonomous control and optimization of quantum experiments is provided, including: a task interface module 802, a quantum experiment execution module 804, a data feedback module 806, and an optimization module 808, wherein:

[0183] The task interface module 802 is configured to receive target specifications for quantum experiment tasks.

[0184] The quantum experiment execution module 804 is configured to determine the set of experimental parameters and experimental strategy according to the target specification; convert the set of experimental parameters and experimental strategy into hardware control instructions; perform at least one quantum experiment based on the hardware control instructions; and obtain a set of observation values ​​containing quantum observation results and classical observation results generated in at least one quantum experiment.

[0185] The data feedback module 806 is communicatively connected to the quantum experiment execution module 804 and is configured to calculate the statistical estimate and uncertainty of the objective function in the objective specification based on the set of observations.

[0186] The optimization module 808, which is communicatively connected to the quantum experiment execution module 804 and the data feedback module 806, is configured to input the experimental parameter set, observation set, statistical estimate and uncertainty into the optimizer to update the parameters, obtain the updated experimental parameter set and the updated experimental strategy; convert the updated experimental parameter set and the updated experimental strategy into hardware control instructions, and return to perform at least one quantum experiment based on the hardware control instructions until the stopping criterion in the target specification is met.

[0187] The task interface module 802, quantum experiment execution module 804, data feedback module 806, and optimization module 808 are interconnected and work together to complete the iterative process of autonomous control and optimization of the quantum experiment. The optimization module 808, quantum experiment execution module 804, and data feedback module 806 form a closed-loop control circuit.

[0188] The quantum experiment execution module 804 includes a hardware abstraction layer that compiles the experimental parameter set X into legal timing control signals that satisfy physical safety and resource budget constraints, enabling synchronous control and feedback across multiple devices. This hardware abstraction layer is not simply a communication encapsulation but rather part of a closed-loop optimization system. Upon receiving candidate parameters and measurement strategies from the optimizer, it performs parameter feasible region verification, resource budget verification, and multi-device synchronous timing compilation to generate legal control sequences that satisfy physical safety and resource constraints. When constraint conflicts occur, it feeds back constraint violation information to the optimizer to trigger parameter projection / resampling, thereby achieving closed-loop automation of "optimization decision-safe and executable instructions-experimental feedback."

[0189] In some embodiments, the quantum experiment execution module 804 includes a digital control instrument selected from an electrically rotating lens holder, an arbitrary waveform generator, a laser, or a piezoelectric controller.

[0190] In some embodiments, the quantum experiment execution module 804 has the ability to maintain device performance descriptions and device communication interfaces, and supports the expansion and / or replacement of hardware devices.

[0191] In some embodiments, the quantum experiment execution module 804 includes a quantum state measurement unit and a classical parameter measurement unit, wherein: the quantum state measurement unit is used to acquire quantum state information, and the quantum state measurement unit includes a single-photon detector or a quantum state tomography device; the classical parameter measurement unit is used to acquire system state parameters or environmental parameters, and the classical parameter measurement unit includes an optical power meter or a temperature sensor.

[0192] In some embodiments, the optimization module 808 can be an AI optimization module, which may include an optimizer, an experimental memory, and a surrogate model training module. The optimizer can be an AI optimizer, which maps simulation evaluation results and / or hardware evaluation results to a consistent evaluation format, and allocates the proportion and order of simulation evaluation and hardware evaluation according to a preset objective function or adaptive rules during the optimization process; the experimental memory and surrogate model training module are used to realize hybrid simulation-hardware co-optimization.

[0193] The experimental memory is used for structured storage, version management, retrieval, and reuse of multi-source experimental data. This multi-source experimental data includes, but is not limited to: historical hardware experimental data, simulation experimental data, calibration data from different devices / batches / environmental conditions, optimization iteration trajectory data, candidate parameter and evaluation result pairs, outlier and failure mode annotations, device and environmental states (including temperature, drift indicators, calibration version, noise level, etc.), and task identifiers and task feature descriptions. Preferably, the experimental memory supports retrieval based on task similarity, device state similarity, parameter neighborhood, or time window to output candidate initializations, prior distributions, feasible domain boundaries, policy templates, or parameter constraints for subsequent tasks, thereby achieving cross-task migration and cold start acceleration.

[0194] The surrogate model training module is used to train, update, or fine-tune the surrogate model based on historical and / or real-time data from the experimental memory in surrogate model mode. This forms an approximate representation of the mapping relationship between the experimental system and the objective function. The trained surrogate model (including model parameters, hyperparameters, and / or uncertainty calibration parameters) is then published to the optimizer for use in the surrogate model-based update mode. Training methods include, but are not limited to: transfer learning, incremental learning, online updates, and joint training or phased training using simulation data and hardware data. Preferably, the surrogate model training module supports online updates of the surrogate model according to a preset period, by the number of iterations, by drift detection triggering, or by uncertainty threshold triggering, to maintain the effectiveness of the surrogate model for the current experimental state.

[0195] In some embodiments, the experimental parameter set includes device control parameters and measurement configuration parameters; the device control parameters are used to control the preparation and evolution of the quantum system corresponding to the quantum experiment; the measurement configuration parameters are used to define the measurement method for the quantum system.

[0196] In some embodiments, the quantum experiment execution module 804 is further configured to acquire device description information of hardware devices in a quantum platform, the quantum platform being used to execute quantum experiment tasks; based on the device description information, parse the experimental parameter set and experimental strategy into intermediate representations independent of the hardware devices; and, if the intermediate representations pass constraint verification, compile the intermediate representations into hardware control instructions for the hardware devices.

[0197] In some embodiments, the quantum experiment execution module 804 is further configured to update the candidate parameters in the intermediate representation if the intermediate representation fails the constraint verification, to obtain an updated intermediate representation; compile the updated intermediate representation into hardware control instructions for the hardware device; and return a constraint violation flag to the optimizer to guide the optimizer to update the parameters.

[0198] In some embodiments, the data feedback module 806 is further configured to determine a statistical summarization and post-processing method that matches the statistical characteristics of the original observation data in the observation set; map the original observation data into intermediate statistics according to the statistical summarization and post-processing method; construct a likelihood function based on the intermediate statistics through a preset observation model and parameterized model, and obtain parameter estimates based on the likelihood function; and obtain the statistical estimate of the objective function and the uncertainty of the statistical estimate based on the parameter estimates and the objective function in the objective specification.

[0199] In some embodiments, the quantum experiment execution module 804 is further configured to determine a scheduling decision for hardware control instructions; if the scheduling decision is hardware evaluation, perform at least one quantum experiment based on hardware control instructions using hardware devices in the quantum platform; and if the scheduling decision is simulation evaluation, perform at least one simulated quantum experiment based on hardware control instructions.

[0200] In some embodiments, at least one of the experimental parameter set and experimental strategy is obtained based on historical mission data that matches the quantum experimental mission.

[0201] In some embodiments, the optimizer includes a surrogate model for parameter updates, wherein at least one of the hyperparameters or prior distribution parameters of the surrogate model is initialized based on historical task data that matches the quantum experimental task.

[0202] In some embodiments, the optimizer includes a surrogate model for parameter updates, which is trained based on historical task data that matches the quantum experimental task.

[0203] In some embodiments, the optimizer includes a surrogate model for parameter updates, and the optimization module 808 is further configured to perform incremental training on the surrogate model based on the current task data after obtaining current task data through at least one quantum experiment.

[0204] In some embodiments, the optimizer includes a surrogate model for parameter updates, and the optimization module 808 is further configured to acquire simulation data and real hardware experimental data of the quantum experimental task, and to jointly train the surrogate model based on the simulation data and real hardware experimental data.

[0205] The modules in the aforementioned quantum experimental autonomous control and optimization system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0206] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 9 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores various data involved in methods for autonomous control and optimization in quantum experiments. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for autonomous control and optimization in quantum experiments.

[0207] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0208] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0209] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0210] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0211] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0212] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0213] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0214] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for autonomous control and optimization of quantum experiments, characterized in that, The method includes: Receive the target specifications for the quantum experiment task, and determine the set of experimental parameters and experimental strategies based on the target specifications; The experimental parameter set and the experimental strategy are converted into hardware control instructions. At least one quantum experiment is performed based on the hardware control instructions, and a set of observation values ​​containing quantum observation results and classical observation results is obtained from the at least one quantum experiment. Based on the set of observations, calculate the statistical estimate of the objective function in the target specification and the uncertainty of the statistical estimate; The experimental parameter set, the observation set, the statistical estimator, and the uncertainty input optimizer are updated to obtain the updated experimental parameter set and the updated experimental strategy. The updated set of experimental parameters and the updated experimental strategy are converted into hardware control instructions, and the process is repeated to perform at least one quantum experiment based on the hardware control instructions until the stopping criterion in the target specification is met.

2. The method according to claim 1, characterized in that, The experimental parameter set includes equipment control parameters and measurement configuration parameters; the equipment control parameters are used to control the preparation and evolution of the quantum system corresponding to the quantum experiment; the measurement configuration parameters are used to define the measurement method for the quantum system.

3. The method according to claim 1, characterized in that, The step of converting the experimental parameter set and the experimental strategy into hardware control instructions includes: Obtain device description information of hardware devices in a quantum platform, the quantum platform being used to perform the quantum experiment task; Based on the device description information, the experimental parameter set and the experimental strategy are parsed into intermediate representations that are unrelated to the hardware device; If the intermediate representation passes the constraint verification, the intermediate representation is compiled into hardware control instructions for the hardware device.

4. The method according to claim 3, characterized in that, The method further includes: If the intermediate representation fails the constraint verification, the candidate parameters in the intermediate representation are updated to obtain the updated intermediate representation. The updated intermediate representation is compiled into hardware control instructions for the hardware device, and a constraint violation flag is returned to the optimizer to guide the optimizer in updating parameters.

5. The method according to claim 1, characterized in that, The step of calculating the statistical estimate of the objective function in the target specification and the uncertainty of the statistical estimate based on the set of observations includes: Based on the statistical characteristics of the original observation data in the observation set, a statistical summarization and post-processing method matching the statistical characteristics is determined; According to the statistical summarization and post-processing method described above, the original observation data are mapped into intermediate statistics; Using a pre-defined observation model and a parameterized model, a likelihood function is constructed based on the intermediate statistics, and parameter estimates are obtained based on the likelihood function. Based on the parameter estimates and the objective function in the target specification, the statistical estimate of the objective function and the uncertainty of the statistical estimate are obtained.

6. The method according to claim 1, characterized in that, The process of conducting at least one quantum experiment based on the hardware control instructions includes: Determine the scheduling decision for the hardware control instructions; When the scheduling decision is a hardware evaluation, at least one quantum experiment is performed by the hardware device in the quantum platform based on the hardware control instructions. When the scheduling decision is a simulation evaluation, at least one simulated quantum experiment is performed based on the hardware control instructions.

7. The method according to any one of claims 1 to 6, characterized in that, The method further includes at least one of the following: At least one of the experimental parameter set and the experimental strategy is obtained based on historical mission data that matches the quantum experimental mission. The optimizer includes a surrogate model for parameter updates, wherein at least one of the hyperparameters or prior distribution parameters of the surrogate model is initialized based on historical task data that matches the quantum experimental task. The optimizer includes a surrogate model for parameter updates, which is trained based on historical task data that matches the quantum experiment task. The optimizer includes a proxy model for parameter updates. After obtaining current task data through at least one quantum experiment, the proxy model is incrementally trained based on the current task data. The optimizer includes a proxy model for parameter updates, acquires simulation data and real hardware experimental data of the quantum experiment task, and performs joint training on the proxy model based on the simulation data and the real hardware experimental data.

8. A system for autonomous control and optimization of quantum experiments, characterized in that, The system includes: The task interface module is configured to receive target specifications for quantum experimental tasks. A quantum experiment execution module is configured to determine an experimental parameter set and an experimental strategy according to the target specification; convert the experimental parameter set and the experimental strategy into hardware control instructions; perform at least one quantum experiment based on the hardware control instructions; and obtain a set of observation values ​​including quantum observation results and classical observation results generated in the at least one quantum experiment. The data feedback module, which is communicatively connected to the quantum experiment execution module, is configured to calculate the statistical estimate of the objective function in the target specification and the uncertainty of the statistical estimate based on the set of observations. The optimization module, communicatively connected to the quantum experiment execution module and the data feedback module, is configured to update the experimental parameter set, the observation set, the statistical estimate, and the uncertainty input optimizer to obtain an updated experimental parameter set and an updated experimental strategy; convert the updated experimental parameter set and the updated experimental strategy into hardware control instructions, and return to perform at least one quantum experiment based on the hardware control instructions until the stopping criterion in the target specification is met.

9. The system according to claim 8, characterized in that, The quantum experiment execution module includes at least one digital control instrument selected from the following: an electrically rotating lens holder, an arbitrary waveform generator, a laser, or a piezoelectric controller.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

Images