A model determination method, device, computer readable storage medium and computer program product

Abstract

This application provides a model determination method, comprising: acquiring sample data and determining sample feature information of the sample data; processing the sample feature information to obtain sample quantum state data; and training an initial hybrid expert model with a target quantum gated network and an expert network based on the sample feature information and the sample quantum state data to obtain a target hybrid expert model, thereby solving the problem of low training efficiency. This application also provides a model determination device, a computer-readable storage medium, and a computer program product.

Description

Technical Field

[0001] This application relates to model determination technology in the field of computer technology, and more particularly to a model determination method, apparatus, computer-readable storage medium, and computer program product. Background Technology

[0002] Hybrid expert models, as an ensemble learning method, construct a powerful overall model by combining a gated neural network with multiple expert neural networks. Different experts focus on their respective domains, thereby improving the model's efficiency, performance, and scalability. Currently, related technologies for hybrid expert models include object detection methods based on multi-gated hybrid expert models and inverse synthesis analysis methods based on quantum recurrent neural networks. However, these solutions suffer from low training efficiency and poor adaptability in scenarios with large datasets or complex data results. Summary of the Invention

[0003] This application provides a model determination method, device, computer-readable storage medium, and computer program product, which solves the problem of low training efficiency in related technologies and improves the adaptability of hybrid expert models.

[0004] The technical solution of this application embodiment is implemented as follows: A model determination method, the method comprising: Acquire sample data and determine the sample feature information of the sample data; The sample feature information is processed to obtain sample quantum state data; Based on the sample feature information and the sample quantum state data, an initial hybrid expert model with a target quantum gating network and an expert network is trained to obtain the target hybrid expert model.

[0005] In the above scheme, training an initial hybrid expert model with a target quantum gating network and an expert network based on the sample feature information and the sample quantum state data to obtain the target hybrid expert model includes: The sample quantum state data is processed using the target quantum gated network to obtain the weight value of each expert network; The sample feature information is processed based on each weight value and each expert network to obtain the target output result; The parameters of the expert network and the parameters of the target quantum gated network are optimized based on the target output results and the target quantum algorithm. The target quantum gating network and the expert network with optimized parameters are trained based on the sample feature information and the sample quantum state data until the target hybrid expert model is obtained.

[0006] In the above scheme, the step of processing the sample quantum state data using the target quantum gating network to obtain the weight value of each expert network includes: The sample quantum state data is processed using each quantum circuit of the target quantum gated network to obtain multiple processing results; The multiple processing results are processed using the connection gates and target logic gates of the target quantum gated network to obtain the weight value of each expert network; wherein the number of qubits of the target quantum gated network, the number of quantum circuits, and the number of expert networks are the same.

[0007] In the above scheme, the step of processing the multiple processing results using the connection gates and target logic gates of the target quantum gated network to obtain the weight value of each expert network includes: The connection gate is used to connect other processing results among the multiple processing results to obtain a connection processing result; wherein, the other processing results are the processing results in the target quantum gated network other than the processing result corresponding to the last quantum circuit; The target logic gate is used to connect the connection processing result and the processing result corresponding to the last quantum circuit to obtain the weight value of each expert network; wherein the target logic gate is the same as the logic gate in the quantum circuit.

[0008] In the above scheme, the step of processing the sample feature information based on each weight value and each expert network to obtain the target output result includes: Each expert network is used to process the sample feature information to obtain multiple sub-output results; The target output is obtained based on each sub-output and the weight value corresponding to each expert network.

[0009] In the above scheme, optimizing the parameters of the expert network and the parameters of the target quantum gated network based on the target output result and the target quantum algorithm includes: The parameters of the target quantum gated network are optimized based on the target output results; Based on the target output, the target quantum optimization algorithm, and the target quantum annealing algorithm, the parameters of the expert network are optimized; wherein, the target quantum algorithm includes the target quantum optimization algorithm and the target quantum annealing algorithm.

[0010] In the above scheme, optimizing the parameters of the expert network based on the target output result, the target quantum optimization algorithm, and the target quantum annealing algorithm includes: Based on the target output and the target quantum optimization algorithm, the parameters of the expert network are processed to obtain the intermediate parameters of the expert network; The intermediate parameters are optimized based on the target quantum annealing algorithm.

[0011] A model determining apparatus, the apparatus comprising: A determining unit is used to acquire sample data and determine the sample feature information of the sample data; The first processing unit is used to process the sample feature information to obtain sample quantum state data; The second processing unit is used to train an initial hybrid expert model with a target quantum gated network and an expert network based on the sample feature information and the sample quantum state data, so as to obtain a target hybrid expert model.

[0012] A model determination device, the device comprising: a processor, a memory, and a communication bus; The communication bus is used to realize the communication connection between the processor and the memory; The processor is used to execute the model determination program in the memory to implement the steps of the model determination method described above.

[0013] A computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the steps of the model determination method described above.

[0014] A computer program product comprising a computer program that, when executed by a processor, implements the aforementioned model determination method.

[0015] The model determination method, device, computer-readable storage medium, and computer program product provided in this application embodiment can acquire sample data, determine the sample feature information of the sample data, process the sample feature information to obtain sample quantum state data, and train an initial hybrid expert model with a target quantum gating network and an expert network based on the sample feature information and the sample quantum state data to obtain a target hybrid expert model. In this way, the gating network of the initial hybrid expert model used to train the target hybrid expert model is a quantum gating network, and the initial hybrid expert model is trained using the corresponding sample quantum state data and sample feature information. Thus, the gating network in the obtained target hybrid expert model is a quantum gating network, not a traditional neural network, and does not require a large number of neurons for training. This solves the problem of low training efficiency in related technologies and improves the adaptability of the hybrid expert model. Attached Figure Description

[0016] Figure 1 This is a schematic flowchart of a model determination method provided in an embodiment of this application; Figure 2 This is a schematic diagram of a PQC gating network in a model determination method provided in an embodiment of this application; Figure 3 This is a schematic diagram of an RT-PQC gating network in a model determination method provided in an embodiment of this application; Figure 4 This is a schematic diagram of the parameterized quantum circuit of QAOA in a model determination method provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of a model determining device provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of a model determination device provided in an embodiment of this application. Detailed Implementation

[0017] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0018] It should be understood that the phrases "embodiments of this application" or "foreign embodiments" throughout the specification mean that a specific feature, structure, or characteristic related to an embodiment is included in at least one embodiment of this application. Therefore, "embodiments of this application" or "in the foreign embodiments" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0019] Unless otherwise specified, any step in the embodiments of this application performed by the electronic device may be executed by the processor of the electronic device. It is also worth noting that the embodiments of this application do not limit the order in which the electronic device performs the following steps. Furthermore, the methods used to process data in different embodiments may be the same or different methods. It should also be noted that any step in the embodiments of this application can be executed independently by the electronic device; that is, when the electronic device performs any step in the following embodiments, it may not depend on the execution of other steps.

[0020] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of this application.

[0021] This application provides a model determination method, which can be applied to a model determination device, as described above. Figure 1 As shown, the model determination method may include the following steps: Step 101: Obtain sample data and determine the sample feature information of the sample data.

[0022] Here, sample data can refer to the large amount of training data used to train the initial hybrid expert model. It should be noted that sample data can correspond to data from various scenarios that require the use of hybrid expert models; in one feasible implementation, sample data can include data from scenarios that require high-dimensional data processing, such as image recognition and recommendation systems.

[0023] Specifically, sample feature information can be obtained by processing sample data; it should be noted that sample feature information can refer to the sample feature vector obtained after feature extraction of sample data.

[0024] Step 102: Process the sample feature information to obtain sample quantum state data.

[0025] Specifically, the sample quantum state data can be obtained by processing the format of the sample feature information. In one feasible implementation, the sample feature information can be encoded to obtain the sample quantum state data; or, in another feasible implementation, the sample feature vector can be encoded using amplitude encoding, angle encoding, or other encoding methods to obtain the sample quantum state data.

[0026] Step 103: Train the initial hybrid expert model with target quantum gating network and expert network based on sample feature information and sample quantum state data to obtain the target hybrid expert model.

[0027] Specifically, the target hybrid expert model can be obtained by training an initial hybrid expert model with a target quantum gated network and an expert network using sample feature information and sample quantum state data.

[0028] It should be noted that the target quantum gated network can refer to a quantum state-gated network. In one feasible implementation, the target quantum gated network can refer to a parametric quantum circuit (PQC) gated network. Furthermore, the initial hybrid expert model can have multiple expert networks.

[0029] Based on the foregoing embodiments, in other embodiments of this application, step 103 can be implemented in the following ways: A1. The sample quantum state data is processed using a target quantum gated network to obtain the weight value of each expert network.

[0030] Specifically, sample quantum state data can be input into the target quantum gated network, and then the target quantum gated network performs a series of processes on the sample quantum state data to obtain the weight value of each expert network.

[0031] It should be noted that the target quantum gated network will classify the expert networks according to the number of expert networks in the initial hybrid expert model and calculate the corresponding weight value for each expert network. Furthermore, if the calculated weight value of an expert network is 0, it means that the expert network will not participate in this training or inference; if the calculated weight value of an expert network is not 0, it means that the expert network will participate in this training or inference.

[0032] A2. Process the sample feature information based on each weight value and each expert network to obtain the target output result.

[0033] Specifically, sample feature information can be input into each expert network for processing, and the processed results can be combined with the calculated weight values ​​of each expert network to obtain the target output result.

[0034] A3. Optimize the parameters of the expert network and the target quantum gated network based on the target output results and the target quantum algorithm.

[0035] Specifically, the parameters of the target quantum gated network can be optimized based on the obtained target output to obtain the parameter-optimized target quantum gated network. At the same time, the parameters of the expert network can be optimized based on the obtained target output and the target quantum algorithm to obtain the parameter-optimized expert network.

[0036] It should be noted that the target quantum algorithm can be a quantum optimization algorithm capable of optimizing the data.

[0037] A4. Train the target quantum gated network and the expert network with optimized parameters based on sample feature information and sample quantum state data until the target hybrid expert model is obtained.

[0038] Specifically, based on the implementation process of steps A1-A3 above, the target quantum gated network with optimized parameters is used to process the sample quantum state data, and the expert network with optimized parameters is used to process the sample feature information and the obtained processing results. Then, based on the obtained output results, the parameters of the target quantum gated network with optimized parameters and the parameters of the expert network with optimized parameters are further optimized. This process is repeated multiple times until the number of iterations reaches a predetermined value or the obtained loss value reaches the target loss value, thus obtaining the target hybrid expert model.

[0039] In other embodiments of this application, the above-described processing of sample quantum state data using a target quantum gating network to obtain the weight values ​​of each expert network includes: Each quantum circuit of the target quantum gated network is used to process the sample quantum state data, resulting in multiple processing results.

[0040] The target quantum gated network may include multiple quantum circuits, multiple connection gates, and target logic gates, with each quantum circuit corresponding to one qubit. It should be noted that the preferred target quantum gated network in this application may refer to a reconnection technique-parametric quantum circuit (RT-PQC) gated network.

[0041] Multiple processing results are processed using the connection gates and target logic gates of the target quantum gated network to obtain the weight value of each expert network.

[0042] In this case, the number of qubits, the number of quantum circuits, and the number of expert networks in the target quantum gated network are the same.

[0043] In this embodiment, the target quantum gated network includes multiple connection gates and one target logic gate. Specifically, the RT-PQC gated network can be obtained by replacing the connection gate corresponding to a quantum circuit in the PQC gated network with the target logic gate. It should be noted that the sum of the number of multiple connection gates and the number of the target logic gate is the same as the number of quantum circuits.

[0044] In other embodiments of this application, the connection gates and target logic gates of the target quantum gated network are used to process multiple processing results to obtain the weight value of each expert network, including: A connection gate is used to connect other processing results among multiple processing results to obtain a connection processing result.

[0045] Other processing results refer to the processing results in the target quantum gated network other than the processing result corresponding to the last quantum circuit.

[0046] The target logic gate is used to connect the connection processing result and the processing result corresponding to the last quantum circuit to obtain the weight value of each expert network.

[0047] The target logic gate is the same as the logic gate in a quantum circuit.

[0048] It should be noted that the connection gate corresponding to the last quantum circuit in the target quantum gated network is the target logic gate, while the connection gates corresponding to the other quantum circuits in the target quantum gated network (excluding the last one) are control-not (CNOT) gates. Furthermore, the target logic gate can be any logic gate in the quantum circuit. In one feasible implementation, such as... Figure 3 The target logic gate shown can be an Rz logic gate.

[0049] In this embodiment of the application, if the number of expert networks is four, the structure of the corresponding PQC gating network can be as follows: Figure 2 As shown, the number of qubits and quantum circuits in this PQC gated network are consistent with the number of expert networks; among them, Figure 2 q in i R represents the i-th qubit, and the following horizontal line represents its corresponding quantum circuit. The quantum circuit mainly indicates which logic gates the quantum qubit evolves through. Z Logic gates can finely adjust the interference modes between quantum states by controlling the phase of the quantum states, and are used for weight adjustment; R X Logic gates can be used to adjust the weights of quantum states, thereby affecting the interference between quantum states; ultimately, each quantum circuit is sequentially augmented with R... Z R XThe logic gates, followed by CNOT gates, connect two adjacent quantum circuits, forming the structure of the PQC gated network corresponding to the 4-expert network. It should be noted that R... Z The formula for a logic gate is: R X The formula for a logic gate is: The CNOT gate corresponds to In addition, R Z R X Both logic gates can be configured with variable parameters θ, representing the rotation angle between them. By repeatedly changing the value of the variable parameter θ and measuring the resulting quantum state, a single measurement result can be converted into a four-digit number represented by 0s and 1s. For example, if the measurement result is 0101, it indicates that expert networks 2 and 4 are selected respectively. By sampling the overall measurement results, the selected expert network can be determined.

[0050] In other embodiments of this application, if the number of expert networks is 4, then the structure of the corresponding RT-PQC gating network can be as follows: Figure 3 As shown. It should be noted that using RT-PQC as the gating network for the target hybrid expert model significantly improves the inference and training efficiency of the gating network. Simultaneously, optimizing the PQC gating network using reconnection techniques mitigates the impact of quantum noise to some extent. Through repeated training and parameter updates, the gating network exhibits better adaptability.

[0051] In other embodiments of this application, the above-described processing of sample feature information based on each weight value and each expert network to obtain the target output result includes: Each expert network processes the sample feature information to obtain multiple sub-output results.

[0052] Specifically, the sample feature information is used as input information and fed into each expert network. Each expert network then processes the sample feature information to obtain multiple sub-output results.

[0053] The target output is obtained based on each sub-output and the weight value corresponding to each expert network.

[0054] Specifically, the multiple sub-outputs of the expert network can be weighted and fused with the weight values ​​of each expert network to obtain the final target output.

[0055] In other embodiments of this application, the above-described optimization of the parameters of the expert network and the parameters of the target quantum gated network based on the target output result and the target quantum algorithm includes: The parameters of the target quantum gated network are optimized based on the target output.

[0056] Specifically, a loss function can be used to process the target output and the corresponding standard result to obtain a loss value, and the parameters of the target quantum gated network can be optimized based on the loss value.

[0057] The parameters of the expert network are optimized based on the target output, the target quantum optimization algorithm, and the target quantum annealing algorithm.

[0058] The target quantum algorithm includes the target quantum optimization algorithm and the target quantum annealing algorithm.

[0059] Specifically, a loss function can be used to process the target output and the corresponding standard result to obtain the loss value. The parameters of the expert network are then optimized based on the loss value, the target quantum optimization algorithm, and the target quantum annealing algorithm. It should be noted that the target quantum optimization algorithm can refer to the Quantum Approximation Optimization Algorithm (QAOA), and the target quantum annealing algorithm can refer to the quantum annealing algorithm.

[0060] In other embodiments of this application, the above-described optimization of the expert network parameters based on the target output result, the target quantum optimization algorithm, and the target quantum annealing algorithm includes: The parameters of the expert network are processed based on the target output and the target quantum optimization algorithm to obtain the intermediate parameters of the expert network.

[0061] Specifically, a parameterized quantum circuit can be constructed, and then a suitable driving Hamiltonian or problem Hamiltonian can be selected. The parameters in the quantum circuit are optimized using classical methods, and finally, by measuring the quantum states, it is determined that they fall near the optimal solution of the objective function with a high probability, thus obtaining a set of approximate optimal solutions (i.e., intermediate parameters). In the embodiments of this application, QAOA can flexibly adjust the quantum circuit structure according to the characteristics of the expert network. If the number of expert networks is four, then the parameterized quantum circuit structure of QAOA can be as follows: Figure 4 As shown; where the number of qubits, the number of quantum circuits, and the number of expert networks are consistent with the final output dimension. R Y Logic gates enhance the expressive power of quantum states; and U, as a flexible three-parameter logic gate, can perform parameterized encoding, quantum feature mapping, and fine-tuning of quantum states; finally, each quantum circuit is sequentially equipped with R... Y The U logic gate is then used, followed by a CNOT gate connecting two adjacent quantum circuits to form a QAOA circuit with a four-dimensional output. The formula for the newly added U logic gate is as follows: .

[0062] The intermediate parameters are optimized based on the target quantum annealing algorithm.

[0063] Specifically, the target quantum annealing algorithm can receive a set of approximate optimal solutions (i.e., intermediate parameters) output by QAOA, execute the annealing process through a quantum annealing device or quantum simulator, slowly adjust the Hamiltonian and parameters, read the state of the qubits through a measurement circuit, and decode the quantum state when the Hamiltonian approaches the global minimum to obtain the parameters of the optimized expert network, ultimately achieving the optimization of the expert network parameters.

[0064] It should be noted that the scheme in this application combines QAOA with quantum annealing, which can leverage the advantages of both algorithms: QAOA can find a set of approximate solutions relatively quickly, usually covering multiple local minima, increasing the chance of quantum annealing escaping local minima; while quantum annealing focuses more on global search, and can be further optimized based on a set of approximate solutions, increasing the probability of finding the global optimum.

[0065] The model determination method provided in the embodiments of this application can acquire sample data and determine the sample feature information of the sample data. The sample feature information is processed to obtain sample quantum state data. Based on the sample feature information and sample quantum state data, an initial hybrid expert model with a target quantum gating network and an expert network is trained to obtain a target hybrid expert model. In this way, the gating network of the initial hybrid expert model used to train the target hybrid expert model is a quantum gating network, and the initial hybrid expert model is trained with the corresponding sample quantum state data and sample feature information. Thus, the gating network in the obtained target hybrid expert model is a quantum gating network, which is not a traditional neural network and does not require a large number of neurons for training. This solves the problem of low training efficiency in related technologies and improves the adaptability of the hybrid expert model.

[0066] Based on the foregoing embodiments, embodiments of this application provide a model determination apparatus, which can be applied to... Figure 1 In the model determination method provided in the corresponding embodiment, refer to Figure 5 As shown, the model determining device 2 may include: a determining unit 21, a first processing unit 22, and a second processing unit 23, wherein: The determining unit 21 is used to acquire sample data and determine the sample feature information of the sample data; The first processing unit 22 is used to process the sample feature information to obtain sample quantum state data; The second processing unit 23 is used to train an initial hybrid expert model with a target quantum gating network and an expert network based on sample feature information and sample quantum state data, so as to obtain a target hybrid expert model.

[0067] In other embodiments of this application, the second processing unit 23 is further configured to perform the following steps: The sample quantum state data is processed using a target quantum gated network to obtain the weight values ​​of each expert network; The sample feature information is processed based on each weight value and each expert network to obtain the target output result; The parameters of the expert network and the target quantum gated network are optimized based on the target output results and the target quantum algorithm. The target quantum gated network and the expert network with optimized parameters are trained based on sample feature information and sample quantum state data until the target hybrid expert model is obtained.

[0068] In other embodiments of this application, the second processing unit 23 is further configured to perform the following steps: Each quantum circuit of the target quantum gated network is used to process the sample quantum state data, resulting in multiple processing results; Multiple processing results are processed using the connection gates and target logic gates of the target quantum gated network to obtain the weight value of each expert network; wherein the number of qubits, the number of quantum circuits, and the number of expert networks in the target quantum gated network are the same.

[0069] In other embodiments of this application, the second processing unit 23 is further configured to perform the following steps: A connection gate is used to connect other processing results among multiple processing results to obtain a connection processing result; wherein, the other processing results are the processing results in the target quantum gated network other than the processing result corresponding to the last quantum circuit; The target logic gate is used to connect the connection processing result and the processing result corresponding to the last quantum circuit to obtain the weight value of each expert network; wherein, the target logic gate is the same as the logic gate in the quantum circuit.

[0070] In other embodiments of this application, the second processing unit 23 is further configured to perform the following steps: Each expert network processes the sample feature information to obtain multiple sub-output results; The target output is obtained based on each sub-output and the weight value corresponding to each expert network.

[0071] In other embodiments of this application, the second processing unit 23 is further configured to perform the following steps: The parameters of the target quantum gated network are optimized based on the target output results; The parameters of the expert network are optimized based on the target output, the target quantum optimization algorithm, and the target quantum annealing algorithm; the target quantum algorithm includes the target quantum optimization algorithm and the target quantum annealing algorithm.

[0072] In other embodiments of this application, the second processing unit 23 is further configured to perform the following steps: The parameters of the expert network are processed based on the target output and the target quantum optimization algorithm to obtain the intermediate parameters of the expert network. The intermediate parameters are optimized based on the target quantum annealing algorithm.

[0073] It should be noted that the specific implementation process of the steps performed by each unit in the embodiments of this application can be referred to Figure 1 The implementation process of the model determination method provided in the corresponding embodiment will not be described in detail here.

[0074] The model determination apparatus provided in the embodiments of this application can acquire sample data and determine the sample feature information of the sample data. It processes the sample feature information to obtain sample quantum state data. Based on the sample feature information and the sample quantum state data, it trains an initial hybrid expert model with a target quantum gating network and an expert network to obtain a target hybrid expert model. Thus, the gating network of the initial hybrid expert model used to train the target hybrid expert model is a quantum gating network, and the initial hybrid expert model is trained using the corresponding sample quantum state data and sample feature information. Therefore, the gating network in the obtained target hybrid expert model is a quantum gating network, not a traditional neural network, and does not require a large number of neurons for training. This solves the problem of low training efficiency in related technologies and improves the adaptability of the hybrid expert model.

[0075] Based on the foregoing embodiments, embodiments of this application provide a model determination device, which can be applied to... Figure 1 In the model determination method provided in the corresponding embodiment, refer to Figure 6 As shown, the model determines that device 3 may include: a processor 31, a memory 32, and a communication bus 33, wherein: Communication bus 33 is used to realize the communication connection between processor 31 and memory 32; The memory 32 is used to store computer programs that can run on the processor 31; Processor 31 is used to run computer programs to perform the following steps: Acquire sample data and determine the sample feature information of the sample data; Sample quantum state data is obtained by processing sample feature information; The initial hybrid expert model with a target quantum gating network and an expert network is trained based on sample feature information and sample quantum state data to obtain the target hybrid expert model.

[0076] In other embodiments of this application, the processor 31 is used to run computer programs and can also perform the following steps: The sample quantum state data is processed using a target quantum gated network to obtain the weight values ​​of each expert network; The sample feature information is processed based on each weight value and each expert network to obtain the target output result; The parameters of the expert network and the target quantum gated network are optimized based on the target output results and the target quantum algorithm. The target quantum gated network and the expert network with optimized parameters are trained based on sample feature information and sample quantum state data until the target hybrid expert model is obtained.

[0077] In other embodiments of this application, the processor 31 is used to run computer programs and can also perform the following steps: Each quantum circuit of the target quantum gated network is used to process the sample quantum state data, resulting in multiple processing results; Multiple processing results are processed using the connection gates and target logic gates of the target quantum gated network to obtain the weight value of each expert network; wherein the number of qubits, the number of quantum circuits, and the number of expert networks in the target quantum gated network are the same.

[0078] In other embodiments of this application, the processor 31 is used to run computer programs and can also perform the following steps: A connection gate is used to connect other processing results among multiple processing results to obtain a connection processing result; wherein, the other processing results are the processing results in the target quantum gated network other than the processing result corresponding to the last quantum circuit; The target logic gate is used to connect the connection processing result and the processing result corresponding to the last quantum circuit to obtain the weight value of each expert network; wherein, the target logic gate is the same as the logic gate in the quantum circuit.

[0079] In other embodiments of this application, the processor 31 is used to run computer programs and can also perform the following steps: Each expert network processes the sample feature information to obtain multiple sub-output results; The target output is obtained based on each sub-output and the weight value corresponding to each expert network.

[0080] In other embodiments of this application, the processor 31 is used to run computer programs and can also perform the following steps: The parameters of the target quantum gated network are optimized based on the target output results; The parameters of the expert network are optimized based on the target output, the target quantum optimization algorithm, and the target quantum annealing algorithm; the target quantum algorithm includes the target quantum optimization algorithm and the target quantum annealing algorithm.

[0081] In other embodiments of this application, the processor 31 is used to run computer programs and can also perform the following steps: The parameters of the expert network are processed based on the target output and the target quantum optimization algorithm to obtain the intermediate parameters of the expert network. The intermediate parameters are optimized based on the target quantum annealing algorithm.

[0082] It should be noted that a detailed description of the steps performed by the processor can be found in [reference needed]. Figure 1 The model determination method provided in the corresponding embodiments will not be described in detail here.

[0083] The model determination device provided in the embodiments of this application can acquire sample data and determine the sample feature information of the sample data. It processes the sample feature information to obtain sample quantum state data. Based on the sample feature information and the sample quantum state data, it trains an initial hybrid expert model with a target quantum gating network and an expert network to obtain a target hybrid expert model. Thus, the gating network of the initial hybrid expert model used to train the target hybrid expert model is a quantum gating network, and it is trained using the corresponding sample quantum state data and sample feature information. Therefore, the gating network in the obtained target hybrid expert model is a quantum gating network, not a traditional neural network, and does not require a large number of neurons for training. This solves the problem of low training efficiency in related technologies and improves the adaptability of the hybrid expert model.

[0084] Based on the foregoing embodiments, embodiments of this application provide a computer-readable storage medium storing one or more programs, which can be executed by one or more processors 31 to implement... Figure 1 The steps of the model determination method provided in the corresponding embodiment.

[0085] Based on the foregoing embodiments, embodiments of this application provide a computer program product, including a computer program that can be executed by a processor 31 to perform... Figure 1 The steps of the model determination method provided in the corresponding embodiment.

[0086] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0087] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0088] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0089] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. ​ One or more processes and / or boxes ​ The steps of the function specified in one or more boxes.

[0090] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A model determination method, characterized in that, The method includes: Acquire sample data and determine the sample feature information of the sample data; The sample feature information is processed to obtain sample quantum state data; Based on the sample feature information and the sample quantum state data, an initial hybrid expert model with a target quantum gating network and an expert network is trained to obtain the target hybrid expert model.

2. The method according to claim 1, characterized in that, The process of training an initial hybrid expert model with a target quantum gated network and an expert network based on the sample feature information and the sample quantum state data to obtain a target hybrid expert model includes: The sample quantum state data is processed using the target quantum gated network to obtain the weight value of each expert network; The sample feature information is processed based on each weight value and each expert network to obtain the target output result; The parameters of the expert network and the parameters of the target quantum gated network are optimized based on the target output results and the target quantum algorithm. The target quantum gated network and the expert network with optimized parameters are trained based on the sample feature information and the sample quantum state data until the target hybrid expert model is obtained.

3. The method according to claim 2, characterized in that, The process of using the target quantum gated network to process the sample quantum state data to obtain the weight values ​​of each expert network includes: The sample quantum state data is processed using each quantum circuit of the target quantum gated network to obtain multiple processing results; The multiple processing results are processed using the connection gates and target logic gates of the target quantum gated network to obtain the weight value of each expert network; wherein the number of qubits of the target quantum gated network, the number of quantum circuits, and the number of expert networks are the same.

4. The method according to claim 3, characterized in that, The process of using the connection gates and target logic gates of the target quantum gated network to process the multiple processing results to obtain the weight value of each expert network includes: The connection gate is used to connect other processing results among the multiple processing results to obtain a connection processing result; wherein, the other processing results are the processing results in the target quantum gated network other than the processing result corresponding to the last quantum circuit; The target logic gate is used to connect the connection processing result and the processing result corresponding to the last quantum circuit to obtain the weight value of each expert network; wherein the target logic gate is the same as the logic gate in the quantum circuit.

5. The method according to claim 2, characterized in that, The process of processing the sample feature information based on each weight value and each expert network to obtain the target output result includes: Each expert network is used to process the sample feature information to obtain multiple sub-output results; The target output is obtained based on each sub-output and the weight value corresponding to each expert network.

6. The method according to claim 2, characterized in that, The optimization of the parameters of the expert network and the parameters of the target quantum gated network based on the target output result and the target quantum algorithm includes: The parameters of the target quantum gated network are optimized based on the target output results; Based on the target output, the target quantum optimization algorithm, and the target quantum annealing algorithm, the parameters of the expert network are optimized; wherein, the target quantum algorithm includes the target quantum optimization algorithm and the target quantum annealing algorithm.

7. The method according to claim 6, characterized in that, The optimization of the expert network parameters based on the target output, the target quantum optimization algorithm, and the target quantum annealing algorithm includes: Based on the target output and the target quantum optimization algorithm, the parameters of the expert network are processed to obtain the intermediate parameters of the expert network; The intermediate parameters are optimized based on the target quantum annealing algorithm.

8. A model determining device, characterized in that, The device includes: a processor, a memory, and a communication bus; The communication bus is used to realize the communication connection between the processor and the memory; The processor is used to execute a model determination program in memory to implement the steps of the model determination method as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs, which can be executed by one or more processors to implement the steps of the model determination method as described in any one of claims 1-7.

10. A computer program product, the computer program product comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the model determination method according to any one of claims 1-7.

Images