Computing power task analysis adaptation method, control system, storage medium and electronic equipment

Abstract

The invention discloses a computing power task analysis adaptation method, a control system, a storage medium and electronic equipment, and the method comprises the steps: giving a method for constructing neural primitives and dynamic neuron data of semantic information for the semantic information and node types; the semantic information-oriented hybrid computing power task analysis adaptation is realized through the links of calibrating dynamic neuron classification labels, scanning and extracting dynamic semantic information, performing similar training on the dynamic neuron to obtain similar neural primitives, training the similar neural primitives, performing semantic adaptation relative entropy judgment feedback, performing error correction and reconstruction on the semantic information and the like. By determining the segmentation granularity of the semantic information, the purposes of segmentation and target detection of the semantic information are achieved; a method for obtaining a globally optimal solution is given through a given semantic information-oriented computing power task analysis adaptation building real distribution probability and prediction distribution probability prediction model, a weighted cross entropy model and a semantic information adaptation relative entropy model, and computing power task analysis adaptation is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of quantum - electronic hybrid computing power networks, and particularly to a computing power task parsing and adaptation method, a control system, a storage medium, and an electronic device. Background Art

[0002] With the application of quantum computing technology and quantum cloud platforms (quantum simulators), the main focus of quantum - electronic hybrid computing power applications is on aspects such as biopharmaceuticals, financial security, and logistics optimization. Among them, for application scenarios with computing power requirements where the semantic information is in the form of pictures or meanings and is dynamic, the request node is for quantum - electronic hybrid computing power task parsing. The hybrid computing power task parsing is a process of parsing, decomposing, classifying, adapting, and learning and judging the computing power task based on the quantum - electronic hybrid computing power requirements. For the computing power task requirements of different application scenarios, especially for application scenarios under dynamic conditions where the semantic information is in the form of pictures and meanings, it is a problem of hybrid computing power task parsing to achieve task parsing adaptation, dynamic calculation, and coupling with the parsing system in each link such as computing power task parsing, decomposition, classification, adaptation, learning, and judgment.

[0003] Semantic information considers the meaning of information and focuses on the accuracy of the meaning. In the actual application process of computing power services, when the application scenario of computing power requirements is semantic information in the form of pictures or meanings, there is a problem of how to accurately adapt the computing power task under semantic information; when the computing power request node is a quantum node or an electronic node, there is a problem of how to obtain computing power task parsing adaptation and couple with the hybrid computing power system. Summary of the Invention

[0004] (1) Technical Problems to be Solved

[0005] Aiming at the deficiencies of the prior art, the present invention provides a computing power task parsing and adaptation method, a control system, a storage medium, and an electronic device, which solves the problem of performing hybrid operations on computing power task parsing and adaptation under semantic information.

[0006] (2) Technical Solutions

[0007] To achieve the above - mentioned purposes, the present invention is realized through the following technical solutions:

[0008] A computing power task parsing and adaptation method includes:

[0009] S1: Establish a data set of neural primitives of the semantic information according to the type of the semantic information;

[0010] S2: Identify the dynamic neurons of the semantic information to obtain a request node set of the computing power task, and establish a neural network in combination with the computing power node set corresponding to the request node set;

[0011] S3: Classify the dynamic neurons according to the type of the semantic information, and calibrate the request node set according to the service demand application scenario;

[0012] S4: Identify and analyze the dynamic neurons according to the type of the semantic information to determine the data set of the dynamic neurons;

[0013] S5: Identify and extract the data set of the dynamic neurons according to the data set of the neural primitives, determine the segmentation granularity of the semantic information of the dynamic neurons, train through the neural network according to the request node set, update the state set of the dynamic neurons, obtain the similar neural primitives of the dynamic neurons through the computing power task parsing algorithm, and train according to the similar neural primitives to obtain the state set of the similar neural primitives;

[0014] S6: Use the trained neural network, input the state set of the similar neural primitives for training, and calculate the semantic information adaptation relative entropy data set between the computing power node and the type of the semantic information;

[0015] S7: Calculate the global optimal solution according to the semantic information adaptation relative entropy data set, and determine whether the global optimal solution exists. If it exists, output the global optimal solution.

[0016] As a further description of the above technical solution: In step S1, it includes:

[0017] S11: Divide the semantic information into picture class and meaning class information, identify the neural primitives of the semantic information, and respectively obtain the feature information and computing power demand ability information of the picture class neural primitives and the meaning class neural primitives;

[0018] S12: Establish the feature information data set and the computing power demand ability measure set of the picture class neural primitives and the meaning class neural primitives, and find the union to obtain the data set of the neural primitives.

[0019] As a further description of the above technical solution: In step S2, it includes:

[0020] Identify the dynamic neurons of the semantic information to obtain quantum request nodes or electronic request nodes, determine the corresponding quantum computing power nodes or electronic computing power nodes according to the quantum request nodes or electronic request nodes, and establish a neural network according to the quantum request nodes, electronic request nodes, quantum computing power nodes, electronic computing power nodes and the information transmission network.

[0021] As a further description of the above technical solution: In step S3, it includes:

[0022] Identify the request nodes of the sub-network in the neural network at any moment according to the type of semantic information, calibrate the request nodes that meet the service demand application scenario, and obtain the calibrated request node set.

[0023] As a further description of the above technical solution: In step S4, it includes:

[0024] S41: Scan the dynamic neurons, select the calibrated request nodes, obtain the data set of the dynamic neurons under this request node, and obtain the feature information data set of the dynamic neurons;

[0025] S42: According to the service demand application scenario, determine the feature information data set of the dynamic neurons, analyze it with the feature information data set of the neural primitives, and obtain the state set of the dynamic neurons.

[0026] As a further description of the above technical solution: In step S5, it includes:

[0027] S51: Segment and perform target detection on the semantic information according to the feature information data set, and determine the segmentation granularity of the semantic information;

[0028] S52: Take the neural primitives as the true distribution and the dynamic neurons as the predicted distribution, establish a prediction model of the true distribution probability and the predicted distribution probability, and obtain the state set of the neural primitives adapted to it according to the state set of the dynamic neurons;

[0029] S53: Take the feature information data set of the dynamic neurons as the predicted distribution, perform probability distribution training, obtain the predicted distribution probability, and perform normalization processing;

[0030] Exponentiate the true distribution of the neural primitives adapted to the dynamic neurons, calculate the true distribution probability, and perform normalization processing;

[0031] S54: Use weighted cross-entropy to measure the difference between the true distribution probability of the neural primitives and the predicted distribution probability of the dynamic neurons, take the weighted cross-entropy as the neural network training data, train to obtain the state set of similar neural primitives, and select the minimum weighted cross-entropy among them as the optimal similar neural primitives.

[0032] As a further description of the above technical solution: In step S6, it includes:

[0033] S61: Extract the state set, feature information data set, and computing power demand ability measure set of the neural primitives corresponding to the optimal similar neural primitives as training data, and perform secondary training on the neural network;

[0034] S62: Determine the computing power capability set corresponding to the computing power node set, and perform relative entropy calculation on the semantic information adaptation according to the service demand application scenario and the type of semantic information, so as to obtain the semantic information adaptation relative entropy data set.

[0035] As a further description of the above technical solution: In step S7, it includes:

[0036] S71: Establish a relative entropy aggregation non-linear programming equation according to the computing power capability set and the set weight coefficient, substitute the semantic information adaptation relative entropy data set of the computing power node set into the relative entropy aggregation non-linear programming equation, calculate the global optimal solution, and determine whether the global optimal solution exists. If it exists, output the global optimal solution of the computing power task parsing adaptation;

[0037] S72: If not, after updating the state set of the similar neural elements, perform error correction and reconstruction training, and return to step S5 to obtain a new state set.

[0038] As a further description of the above technical solution: In step S72, it includes:

[0039] S721: Identify the semantic information of the dynamic neurons, obtain a new feature information data set, calculate the semantic information entropy of the two identifications according to the data set obtained in step S5, and calculate the average value of the semantic information entropy as the conditional entropy;

[0040] S722: According to the service demand application scenario, calculate the minimum conditional entropy of the computing power nodes under different types of semantic information as the data set of the new dynamic neurons, and return to step S5.

[0041] It also includes a computing power task parsing adaptation control system, and the control system is applicable to the parsing adaptation method described in any one of the above technical solutions, including:

[0042] A data storage module, which is used to store the feature information data set, computing power demand capability measure set and state set corresponding to the neural elements, dynamic neurons and similar neural elements;

[0043] A neural network training module, which performs neural network training iteration according to the data in the data storage module to determine the state set of the similar neural elements;

[0044] A calibration module, which classifies and labels the types of semantic information of the dynamic neurons according to the service demand application scenario;

[0045] A semantic information extraction module, which is used to classify according to the computing power request of the service demand application scenario according to the type of semantic information, identify the feature information data set and the computing power demand capability measure set, and store them in the data storage module;

[0046] A judgment feedback module for solving and training the relative entropy of semantic information adaptation with respect to the global optimal solution processing value provided by a quantum node or an electronic node for computing power.

[0047] An error correction and reconstruction module for updating and identifying feature information, performing error correction and reconstruction training according to the application scenario of service requirements, and obtaining a new state set.

[0048] It also includes a computer-readable storage medium storing a computer program for parsing and adapting methods, wherein the computer program causes a computer to execute the parsing and adapting method described in any one of the above technical solutions.

[0049] It also includes an electronic device, including:

[0050] One or more processors; a memory; and

[0051] One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the programs include those for executing the parsing and adapting method described in any one of the above technical solutions.

[0052] (III) Beneficial effects

[0053] The above technical solutions have the following advantages or beneficial effects:

[0054] 1. A hybrid computing power task parsing and adaptation method based on semantic information is designed. When the semantic information is picture type, meaning type, and the nodes are quantum nodes or electronic nodes, methods for constructing neural primitives and dynamic neuron data of semantic information are given; from a full-process and engineering perspective, a method for constructing a training and inference neural network is given; through processes such as calibrating dynamic neuron classification labels, scanning and extracting dynamic semantic information, obtaining similar neural primitives through similar training of dynamic neurons, performing inference training on similar neural primitives, judging and feedback of semantic adaptation relative entropy, and error correction and reconstruction of semantic information, the parsing and adaptation of hybrid computing power tasks for semantic information are realized.

[0055] 2. By determining the segmentation granularity of semantic information, the purpose of segmenting semantic information and object detection is achieved; by giving a prediction model for establishing the true distribution probability and prediction distribution probability of computing power task parsing and adaptation for semantic information, a weighted cross-entropy model, and selecting the fuzzy minimum weighted cross-entropy for target evaluation as the optimal similar neural primitives for data of secondary training.

[0056] 3. By extracting the computing power capability set corresponding to the computing power node set, a calculation method for semantic information adaptation relative entropy is given, realizing the mapping between the data set of dynamic neurons and the data set of neural primitives.

[0057] 4. Use the state set of similar neural elements as input for secondary training of the trained neural network, establish a one-to-one mapping relationship between the request nodes and the computing power capacity sets of the computing power nodes, and obtain the optimal solution for computing power adaptation that can acquire unknown semantic information through known semantic information.

[0058] 5. Present an algorithm for semantic information adaptation relative entropy. Establish a semantic information adaptation relative entropy model for computing power task parsing, and give a method for obtaining the global optimal solution of the semantic adaptation entropy value for the relative entropy aggregation non-linear programming equation for computing power task parsing.

[0059] 6. A method for error correction and reconstruction of semantic information. Perform a secondary scan on the semantic information of the service demand application scenario, obtain a new feature information data set of dynamic neurons for error correction and reconstruction training, obtain the state set of the reconstructed new dynamic neurons, and further carry out computing power task parsing adaptation, improving the timeliness, accuracy, reliability, and overall efficiency of computing power services. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0061] Figure 1 It is a flowchart of the parsing adaptation method proposed by the present invention;

[0062] Figure 2 It is a flowchart of determining the data set of neural elements in the present invention;

[0063] Figure 3 It is a flowchart of determining the data set of dynamic neurons in the present invention;

[0064] Figure 4 It is a flowchart of obtaining the state set of similar neural elements in the present invention;

[0065] Figure 5 It is a flowchart of obtaining the semantic information adaptation relative entropy data set in the present invention;

[0066] Figure 6 It is a flowchart of calculating the global optimal solution in the present invention;

[0067] Figure 7 It is a flowchart of performing error correction and reconstruction training in the present invention;

[0068] Figure 8 It is a schematic structural diagram of the control system proposed by the present invention.

[0069] Legend Explanation:

[0070] 1. Data storage module; 2. Neural network training module; 3. Calibration module; 4. Semantic information extraction module; 5. Judgment feedback module; 6. Error correction and reconstruction module. Specific Embodiment

[0071] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present invention.

[0072] Refer to Figure 1 , to solve the problem of parsing and adapting computing power tasks and performing hybrid operations under semantic information, the present invention provides an embodiment:

[0073] A method for parsing and adapting computing power tasks includes:

[0074] S1: Establish a data set of neural primitives of semantic information according to the type of semantic information;

[0075] S2: Identify the dynamic neurons of semantic information to obtain a request node set of computing power tasks, and establish a neural network in combination with the computing power node set corresponding to the request node set;

[0076] S3: Classify the dynamic neurons according to the type of semantic information, and calibrate the request node set according to the application scenario of service requirements;

[0077] S4: Identify and analyze the dynamic neurons according to the type of semantic information to determine the data set of dynamic neurons;

[0078] S5: According to the data set of neural primitives, identify and extract the data set of dynamic neurons, determine the segmentation granularity of the semantic information of the dynamic neurons, train the neural network according to the request node set, update the state set of the dynamic neurons, obtain the similar neural primitives of the dynamic neurons through the computing power task parsing algorithm, and train according to the similar neural primitives to obtain the state set of the similar neural primitives;

[0079] S6: Use the trained neural network, input the state set of similar neural primitives for training, and calculate the semantic information adaptation relative entropy data set between the computing power node and the type of semantic information;

[0080] S7: Calculate the global optimal solution according to the semantic information adaptation relative entropy data set, and judge whether the global optimal solution exists. If it exists, output the global optimal solution.

[0081] In this embodiment, a virtual network (active network) topology structure and mapping relationship corresponding to the entity network are constructed, the connection relationship between computing power nodes is determined, and label calibration is established for the service demand application scenario state with semantic information picture type and meaning type classification; the mapping relationship between dynamic neurons of dynamic semantic information and neural primitives of semantic information is established, and the mapping network is stored in the semantic information source standard library to realize the virtual resource control, platform operation, state execution, decision-making and other functions of the virtual object (such as virtual machines, software images, etc.) by the service provision layer and network transmission layer of the virtual space; a neural network composed of neural primitives, dynamic neurons, and hybrid computing power network is constructed; according to the computing power request of the service demand application scenario by the neural network, based on the semantic information being picture type or meaning type, through AI, the dynamic neurons of the service demand application scenario are deeply learned, scanned, recognized, and analyzed to extract dynamic semantic information; the AI deep learning conditional entropy algorithm is used to train the similarity of dynamic neurons to obtain similar neural primitives; using the state set of similar neural primitives of the trained neural network with computing power nodes being quantum or electronic as the input, the AI deep learning weighted minimum conditional entropy algorithm is used to perform secondary training search adaptation on the similar neural primitives to calculate the computing power capability set corresponding to the computing power node set; global optimization adaptation service is performed according to the relative entropy non-linear programming optimal solution model; this method realizes the real-time generation of the computing power task parsing strategy for semantic information, realizes software-defined network, and plays the role of replacing hardware with software.

[0082] In the computing power resource layer, for the s-th sub-network, the i-th computing power providing node is a quantum computing power node set and an electronic computing power node set where Q represents quantum bits, B represents electronic bits, s is the sub-network ordinal number, i is the node ordinal number, and the requesting node is a quantum node set or an electronic node set and the computing network fusion system composed of the quantum-electronic hybrid computing power network {N}, defined as the quantum-electronic hybrid computing power network, denoted as It should be noted that the hybrid computing power network here generally refers to the network and platform that support quantum-electronic hybrid computing power services.

[0083] First, for any quantum-electronic hybrid computing power network in which the computing power providing node is a quantum computing power node set or an electronic computing power node set in the s-th sub-network, at any time t, the requesting node set of the service demand application scenario to be connected to the network When the quantum requesting node is denoted as or the electronic requesting node is denoted as For the computing power network of the computing power node set Or All are tagged and undergo network-wide identity security authentication. The specific network-wide security authentication algorithms for the tags and identities are existing technologies and will not be elaborated here.

[0084] Refer to Figure 2 , in step S1, it includes:

[0085] S11: Divide the semantic information into picture-type and meaning-type information, identify the neural primitives of the semantic information, and respectively obtain the feature information and computing power demand ability information of the picture-type neural primitives and meaning-type neural primitives;

[0086] S12: Establish a dataset of feature information and a measure set of computing power demand ability for the picture-type neural primitives and meaning-type neural primitives, and find the union to obtain the dataset of neural primitives.

[0087] In this embodiment, the semantic information considers the meaning of the information and focuses on the accuracy of the meaning. At the computing power resource layer, construct the "basic elements" of the semantic information source, that is, construct the semantic information units of the smallest particles (basic units), which are defined as the "basic" units of the semantic information source and are called neural primitives. According to categories such as picture-type and meaning-type of the semantic information, establish datasets for the feature information and computing power demand ability values of the neural primitives respectively. The feature information dataset is denoted as The measure set of computing power demand ability is denoted as Find the union,

[0088] Obtain the dataset of neural primitives as the neural primitive standard library of the semantic information source to provide support for subsequent training.

[0089] Among them, T represents the feature information, S represents the computing power demand ability information; 1 represents the picture-type of the semantic information, and 2 represents the meaning-type of the semantic information.

[0090] (1) Determine that the set of feature information parameters of the picture-type neural primitives of the semantic information source is denoted as

[0091] The set of computing power demand ability is denoted as

[0092]

[0093] (2) Determine that the set of feature information parameters of the meaning-type neural primitives of the semantic information source is denoted as

[0094] The set of computing power demand ability is denoted as

[0095]

[0096] Feature information dataset of neural elements Computing power demand capacity set The standard library of neural elements {Γ that constitutes the semantic information source (0)}(s = 1, 2,..., m; i = 1, 2,..., n; l = 1, 2,...).

[0097] Specifically, in step S2, it includes:

[0098] Identify the dynamic neurons of semantic information to obtain quantum request nodes or electronic request nodes, determine the corresponding quantum computing power nodes or electronic computing power nodes according to the quantum request nodes or electronic request nodes, and establish a neural network according to the quantum request nodes, electronic request nodes, quantum computing power nodes, electronic computing power nodes and information transmission network.

[0099] In this embodiment, in the computing power service layer, when the request node of the service demand application scenario is semantic information, define the unit that can identify the smallest particle of semantic information, which is called a dynamic neuron. For the s-th sub-network, at any time t, the request node set is the quantum request node Electronic request node Since the change of the semantic information of the service demand application scenario is a dynamic change, it is called dynamic semantic information.

[0100] In the network connection layer, build a deep neural network system and carry out computing power task parsing and adaptation deep learning based on the topological structure of the system.

[0101] Consisting of neural elements, at any time t, the i-th computing power request node of the s-th sub-network is the quantum request node Or electronic request node The dynamic neuron of the l-th category where the semantic information of the service application scenario is picture type or meaning type Or And the quantum computing power node that provides the computing power node Or electronic computing power node The neural network composed of the quantum-electronic hybrid computing power network {N}:

[0102]

[0103] Defined as the deep neural network system {DNN}. The deep neural network system {DNN} provides support for subsequent AI to train and infer dynamic neurons.

[0104] Specifically, in step S3, it includes:

[0105] Identify the request nodes of the sub-networks in the neural network at any moment according to the type of semantic information, calibrate the request nodes that meet the service demand application scenarios, and obtain the calibrated request node set.

[0106] In this embodiment, the dynamic neurons are classified according to service demand application scenarios such as semantic information picture type, meaning type, etc. For the s-th sub-network, at any moment t, calibrate the computing power request node set: when the request node i is a quantum request node or an electronic request node the l-th type of classification label, and classify them according to the type of semantic information, into picture type and meaning type, denoted respectively as or the semantic information picture type of the data set of the corresponding dynamic neuron is denoted as or the semantic information meaning type is denoted as or

[0107] Refer to Figure 3 , in step S4, it includes:

[0108] S41: Scan the dynamic neurons, select the calibrated request nodes, obtain the data set of the dynamic neurons under this request node, and obtain the feature information data set of the dynamic neurons;

[0109] S42: According to the service demand application scenario, determine the feature information data set of the dynamic neurons, analyze it with the feature information data set of the neural primitives, and obtain the state set of the dynamic neurons.

[0110] In this embodiment, obtain the feature information data set of the dynamic neurons

[0111] For the s-th sub-network, at any moment t, the service demand application scenario is classified according to semantic information sources into categories such as semantic information picture type or information meaning classification, etc. The AI scans and extracts dynamic semantic information from the dynamic neurons of the l-th type of service demand application scenario, and determines that the request node is a quantum node or an electronic node the data set of the dynamic neurons when or or

[0112] or

[0113] or

[0114] From the data set of the dynamic neurons

[0115] The set composed of is called the dynamic semantic information data set, denoted as {Γ (t)}:

[0116]

[0117] or or respectively correspond to the quantum nodes or electronic nodes of the semantic information picture class and meaning class, and are called the feature vectors of dynamic semantic information. Each component is called the dynamic semantic information measure value, which changes dynamically with different service demand application scenarios and satisfies the following conditions:

[0118]

[0119] The dynamic semantic information data set composed of dynamic neurons

[0120] The database formed is called the semantic information dynamic neuron database. {Γ (t)} can be dynamically and randomly stored and called in the semantic information dynamic neuron database.

[0121] AI deep learning scans and analyzes to determine the dynamic neuron feature information data set;

[0122] (1) For the semantic information picture class, at any time t, the s-th subnet requests the i-th node to calibrate the quantum or electronic node according to the computing power of the service demand application scenario as or Classification label; Based on the classification AI, for the dynamic neuron data set of the l-th service demand application scenario or Determine the feature information data set of the j-th dynamic neuron quantum node or electronic node according to the following process.

[0123] When the requested node is a quantum node or an electronic node, use the AI software to obtain the data set of dynamic neurons as or Perform deep learning scanning, analysis, recognition, and classification on it to obtain the dynamic semantic information feature information data set as or

[0124]

[0125] (2) For the semantic information meaning class, at any time t, the s-th subnet requests the i-th node to calibrate the quantum or electronic node according to the computing power of the service demand application scenario as or Classification label; Based on the classification AI, for the dynamic neuron data set of the l-th service demand application scenario Or Determine the characteristic information dataset of the j-th dynamic neuron quantum node or electronic node according to the following process.

[0126] When the requesting node is a quantum node or an electronic node, use AI software to obtain the dataset of dynamic neurons as Or Perform deep learning scanning, analysis, recognition, and classification on it to obtain the characteristic information dataset of dynamic semantic information as Or

[0127]

[0128] Extract the characteristic parameter set of the neuron element dataset in the semantic information source neuron element standard library {Γ (0)} Computing power demand capacity set Store the dataset of the characteristic information of dynamic neurons Or Prepare training data for subsequent determination of similar neuron elements.

[0129] The methods for obtaining the state set of dynamic neurons include:

[0130] (1) For semantic information picture types, at any time t, the s-th subnet requests the characteristic information dataset of the l-th service demand application scenario of the i-th node as a quantum node or an electronic node according to the computing power of the service demand application scenario. Or Perform AI deep learning feature analysis on it with the semantic information source neuron element standard library {Γ (0)} (such as performing scene analysis from the well-known ImageNet dataset: making the existing clear category standards of the ImageNet dataset into a picture dictionary and finding the neuron elements of the l-th subset Or ), and obtain the state set of the characteristic information of the dynamic neurons corresponding to the similar neuron elements Or

[0131]

[0132] (2) For semantic information meaning types, at any time t, the s-th subnet requests the characteristic information dataset of the i-th node as a quantum node or an electronic node according to the computing power of the service demand application scenario. Or Perform AI deep learning feature analysis on it with the dataset of semantic information neuron elements {Γ (0)} to obtain similar neuron elements The state set of the corresponding feature information

[0133] or

[0134]

[0135] (l = 1, 2,...);

[0136] Store the dynamic semantic information feature information data set and the feature information state set, and respectively or or or or or or Store them in the semantic information dynamic neuron database for subsequent calls.

[0137] Refer to Figure 4 , in step S5, it includes:

[0138] S51: Segment and perform target detection on the semantic information according to the feature information data set to determine the segmentation granularity of the semantic information;

[0139] S52: Use the neural primitives as the true distribution and the dynamic neurons as the predicted distribution to establish a prediction model of the true distribution probability and the predicted distribution probability, and obtain the state set of the neural primitives adapted to it according to the state set of the dynamic neurons;

[0140] S53: Use the feature information data set of the dynamic neurons as the predicted distribution, perform probability distribution training to obtain the predicted distribution probability, and perform normalization processing;

[0141] Exponentiate the true distribution of the neural primitives adapted to the dynamic neurons, calculate the true distribution probability, and perform normalization processing;

[0142] S54: Use weighted cross - entropy to measure the difference between the true distribution probability of the neural primitives and the predicted distribution probability of the dynamic neurons, use the weighted cross - entropy as the neural network training data, train to obtain the state set of similar neural primitives, and select the minimum weighted cross - entropy among them as the optimal similar neural primitives.

[0143] In this embodiment, the AI performs similar training on the dynamic neurons based on the feature information data set of the neural primitives in the semantic information library or The AI extracts the quantum nodes or electronic nodes of the semantic information dynamic neuron database, and the feature information data sets of the computing power requirements for semantic information picture classes and meaning classes are respectively or or Determine the semantic information segmentation granularity of the application scenario picture class, semantic information meaning class dynamic neurons, perform object detection, and obtain the feature information state set according to the request node for computing power learning and training or or Evaluate and calculate the similar neural primitives obtained from the target dynamic neurons, and define them as similar dynamic neurons; determine the computing power capacity requirement value of the similar neural primitives, and use the feature information state set of the similar neural primitives and the computing power capacity data set as the feature information state set and computing power capacity requirement data set of the dynamic neurons, and store them in the semantic information dynamic neuron database to provide data support for the subsequent secondary training of the dynamic neurons

[0144] The dynamic neuron training algorithm based on AI deep learning has the following steps, including:

[0145] (1) Determine the semantic information segmentation granularity, and object detection to obtain the feature information state set of the dynamic neurons. There are many uncertain factors in object detection, such as the uncertainty of the application scenario target types and numbers in semantic information. Regard these possible uncertain factors as feature information parameter identifiers to form the feature information state set of the dynamic neurons, and eliminate the uncertainty by calculating the cross entropy subsequently

[0146] When the application scenario is the semantic information picture class and the computing power node is a quantum node or an electronic node, use the feature information data set or Segment and perform object detection on the semantic information image class. First, determine the segmentation granularity of the semantic information image class, and use the feature information measurement unit to determine the segmentation granularity. Let's divide it into l (l ∈ N 1 one-dimensional integer set) classes; due to the uncertainty of the object detection feature information data set, obtain the feature information state set of the dynamic neurons

[0147]

[0148] or

[0149] When the application scenario is the semantic information meaning class and the computing power node is a quantum node or an electronic node, use the feature information data set or Segment and perform object detection on the semantic information meaning class. First, determine the segmentation granularity of the semantic information meaning class, and use the feature information measurement unit to determine the fine granularity. Let's divide it into l (l ∈ N 1 one-dimensional integer set) classes; due to the uncertainty of the object detection feature information data set, obtain the feature information state set of the dynamic neurons of the quantum node or the electronic node

[0150]

[0151] or Store the characteristic information state set of the above dynamic neurons into the semantic information dynamic neuron database.

[0152] (2) Obtain the characteristic information state set of the nerve primitives adapted to the dynamic neurons. For AI probability distribution training, regard the nerve primitives as the true distribution and the dynamic neurons as the predicted distribution, obtain a prediction model for establishing the true distribution probability and the predicted distribution probability, and obtain the characteristic information state set of the nerve primitives adapted to them according to the characteristic information state set of the dynamic neurons.

[0153] When the application scenario is semantic information picture class, at any time t, for any According to the characteristic information state set of the dynamic neurons of the quantum node or the electronic node

[0154]

[0155] or

[0156] The nerve primitives corresponding to the AI deep learning and calibrated dynamic neurons can find out the corresponding optimal characteristic information state set of the nerve primitives,

[0157]

[0158] When the application scenario is semantic information meaning class, at any time t, for any According to the characteristic information state set of the dynamic neurons of the quantum node or the electronic node

[0159]

[0160] or

[0161] The nerve primitives corresponding to the AI deep learning and calibrated dynamic neurons can find out the corresponding optimal characteristic information state data set of the nerve primitives,

[0162]

[0163] Store the state set of the above characteristic information into the semantic information source nerve primitive standard library.

[0164] (3) Calculate the prediction probability of dynamic neurons. Perform a normalization transformation on the scoring function of dynamic neurons. Extract the dynamic neuron feature information state dataset, and calculate the prediction probability of dynamic neurons by obtaining the evaluation score probability value of the l-th category at any time t.

[0165] When the service application scenario requested by the user is a semantic information picture class and the request node is a quantum node or an electronic node, the dynamic neuron feature information dataset or is regarded as a predictive distribution, and probability distribution training is carried out to obtain the predictive distribution probability or

[0166] Calculate the number of unnormalized probability values or Define the predictive distribution probability

[0167] or is the evaluation score value of the l-th category, and the normalized probability value is obtained:

[0168] or

[0169] When the service application scenario requested by the user is a semantic information meaning class and the request node is a quantum node or an electronic node, the dynamic neuron feature information dataset or is regarded as a predictive distribution, and probability distribution training is carried out to obtain the predictive distribution probability or

[0170] Calculate the number of unnormalized probability values or Define the predictive distribution probability or is the evaluation score value of the l-th category, and the normalized probability value is obtained:

[0171] or

[0172] (4) Obtain the neural basis elements and true probabilities adapted to the dynamic neurons. At any time t, for any Regard the neural basis elements adapted to the dynamic neurons as the true distribution, perform the exponential processing exp, and calculate the true distribution probability.

[0173] When the computing power demand application scenario is a semantic information picture class and the request node is a quantum node or an electronic node, at any time t, for any (l is the category type of the picture class), the neural basis elements or is regarded as the true distribution, and after exponential processing exp, the unnormalized probability value is obtained or Define the probability of the true distribution:

[0174]

[0175] or

[0176] represents the score value of the l-th type of picture category, and the normalized probability value is obtained:

[0177] or

[0178] When the computing power demand application scenario is the semantic information meaning category and the requesting node is a quantum node or an electronic node, the unnormalized probability value is obtained or Define the probability of the true distribution:

[0179]

[0180] or

[0181] where l is the category of the semantic information meaning category or represents the parameter value of the l-th semantic information meaning category.

[0182] (5) Calculate the fuzzy minimum weighted cross entropy for the target evaluation. The minimum weighted cross entropy is used to measure the difference between the two probability distributions of the true distribution probability p of the neural element and the predicted distribution probability q of the dynamic neuron. The smaller the weighted cross entropy value, the closer the two distributions are, that is, the more similar the dynamic neuron is to the neural element; the closer the predicted distribution is to the probability of correct classification being 1, the closer the two distributions are, that is, the more similar the meaning class dynamic neuron is to the neural element, and the similar neural element for the target evaluation is obtained. The closer the predicted distribution of the model is to the probability of correct classification being 1, the closer the value of the cross entropy H(p, q) loss function is to 0, indicating that the model performs better. Therefore, the minimum weighted cross entropy is taken as the training data for the similar neural element.

[0183] When scanning the semantic information picture category and the requesting node is a quantum node or an electronic node, at any time t, for any the weighted cross entropy H(p, q) of the dynamic neuron q adapted to the true distribution feature information measure set p of the neural element is obtained Q(1) or H(p, q) B(1) :

[0184]

[0185] or

[0186]

[0187] where correspond to the l-th picture category distribution respectively or the weight coefficients;

[0188] When scanning the semantic information meaning class, define the request node as a quantum node or an electronic node at any time t, and obtain the weighted cross-entropy H(p, q) of the dynamic neuron q that is adapted to the measure set p of the true distribution characteristics of the neural primitives Q(2) or H(p, q) B(2) :

[0189]

[0190] (l = 1, 2,...)

[0191] where correspond to the l-th picture category distribution respectively or the weight coefficients;

[0192] that is, use the weighted cross-entropy H(p, q) Q(1) or H(p, q) B(1) 、H(p, q) Q(2) or H(p, q) B(2) is to measure the true distribution probability or and the predicted distribution probability or true distribution probability or and the predicted distribution probability or the score value of the difference between these two probability distributions, and obtain the normalized probability value:

[0193] or

[0194] (6) Find the minimum weighted cross-entropy as the similar neuron. Conduct the weighted cross-entropy training analysis of the AI technology deep learning multi-logic regression algorithm, and calculate to obtain the similar neural primitives:

[0195] When scanning the semantic information picture class and the request node is a quantum node or an electronic node, define the cross-entropy H(p, q) at any time t 1Q(t) or H(p, q) 1B(t) ,

[0196] Take

[0197] or

[0198] corresponding or That is, the optimal similar neuron corresponding to the semantic information picture class is denoted as or

[0199] When the scan is of the semantic information meaning class and the request node is a quantum node or an electronic node, at any time t, the request node is defined as the cross-entropy H(p, q) 2Q(t) or H(p, q) 2B(t) ,

[0200] Take

[0201] or

[0202] corresponding or That is, the optimal similar neuron corresponding to the semantic information meaning class is denoted as or Store the similar neural primitives in the semantic information dynamic neuron database.

[0203] (7) Obtain the computing power requirement capability dataset of the similar neural primitive feature state dataset. The corresponding computing power capability requirement value can be obtained from the optimized similar neural primitives, and this is used as the dynamic neuron computing power capability requirement value. Store the dynamic neuron computing power capability requirement value in the semantic information dynamic neuron database.

[0204] The optimal similar neuron corresponding to the semantic information picture class or The corresponding neural primitive computing power requirement capability dataset is denoted as or

[0205] The optimal similar neuron corresponding to the semantic information meaning class or The corresponding neural primitive computing power requirement capability dataset is denoted as or

[0206] Refer to Figure 5 , in step S6, it includes:

[0207] S61: Extract the state set, feature information dataset, and computing power requirement capability measure set of the neural primitive corresponding to the optimal similar neural primitive as training data, and perform secondary training on the neural network;

[0208] S62: Determine the computing power capability set corresponding to the computing power node set, and perform relative entropy calculation for semantic information adaptation according to the service demand application scenario and the type of semantic information to obtain a semantic information adaptation relative entropy data set.

[0209] In this embodiment, for the inference training of similar neural primitives by AI, the state set of the similar neural primitive feature information of the neural network after training regarding the computing power nodes as quantum or electronic is used as the input for secondary training, that is, determine the s-th subnet, the i-th computing power node, and the l-th type of semantic information:

[0210] (1) Call the optimal similar neuron of the semantic information picture class Or The corresponding neural primitive computing power demand capability data set Or

[0211] (2) Call the optimal similar neuron of the semantic information meaning class Or The corresponding neural primitive computing power demand capability data set Or

[0212] (3) Call the computing power providing node set of the identification module For quantum nodes Or electronic nodes Of the computing power capability data set

[0213] (4) Calculate the semantic information adaptation relative entropy data set between the two types of capability sets Or Obtain the computing power capability relative entropy and perform semantic adaptation relative entropy judgment feedback.

[0214] The algorithm flow steps of the inference training of similar neural primitives by AI are as follows, including:

[0215] (1) Dynamic training data extraction and call. Call the neural primitive feature information state data set and the corresponding neural primitive computing power demand capability data set to participate in secondary inference and calculation.

[0216] When the service demand application scenario is the semantic information picture class and the computing power request node is a quantum node or an electronic node, extract the neural primitive feature information state data set corresponding to the optimal similar neuron And the corresponding neural primitive computing power demand capability data set Or

[0217] When the service demand application scenario is the semantic information meaning class and the computing power request node is a quantum node or an electronic node, extract the neural primitive feature information state data set corresponding to the optimal similar neuron and the corresponding computing power requirement ability dataset of neural elements or where s is the subnet ordinal number, i is the computing power node ordinal number, j is the component of the computing power requirement ability evaluation vector, and l is the classification where the semantic information is located.

[0218] (2) Extract the computing power providing node set Computing power value. Call the "computing power identification security authentication module" of the system for the computing power providing node set in the identification information library identification coding information to obtain the computing power ability dataset:

[0219] When the computing power providing node is a quantum node determine the s-th subnet and the i-th quantum computing power node The quantum computing power ability set is defined as For example, take the quantum computing power ability of the quantum computing power node in the identification information: as the quantum computing power ability to participate in the relative entropy calculation of semantic information adaptation;

[0220] When the computing power providing node is an electronic node determine the s-th subnet and the i-th electronic computing power node The computing power ability set is defined as For example, take the electronic computing power ability of the electronic computing power node in the identification information for the relative entropy calculation of the adaptation of the electronic computing power ability and the semantic information of the electronic computing power ability.

[0221]

[0222] (3) Calculate the relative entropy of the adaptation of the semantic information for the required application scenario. The relative entropy can be used to measure the degree of conformity between the two. The smaller the relative entropy, the higher the degree of conformity, and vice versa.

[0223] Calculate the relative entropy of the adaptation of the semantic information when the required application scenario is a semantic information picture type.

[0224] When the computing power providing node is a quantum computing power node the similar neural element computing power requirement ability and the quantum computing power ability relative entropy of semantic information adaptation: for the similar neural element computing power requirement ability is:

[0225]

[0226] Normalize:

[0227]

[0228] Obviously,

[0229] Define the computing power requirement ability of similar neural elements and the quantum computing power ability The semantic information adaptation relative entropy is denoted as

[0230]

[0231] When the provided computing power node is an electronic computing power node the computing power requirement ability of similar neural elements and the electronic computing power ability The semantic information adaptation relative entropy: the ability requirement for similar neural elements

[0232] Normalization processing:

[0233] Obviously,

[0234] Define the computing power requirement ability of similar neural elements and the electronic computing power ability The semantic information adaptation relative entropy is denoted as

[0235]

[0236]

[0237] When the computing application scenario is of the semantic meaning information type, the semantic information adaptation relative entropy value:

[0238] When the provided computing power node is a quantum node the computing power requirement ability of similar neural elements and the quantum computing power ability The semantic information adaptation relative entropy: the computing power requirement ability for similar neural elements is

[0239] Normalization processing:

[0240] Obviously,

[0241] Define the requirement ability of similar neural elements and the quantum computing power ability The semantic information adaptation relative entropy is denoted as

[0242]

[0243] When the provided computing power node is an electronic computing power node the computing power demand capacity of similar neural elements and the electronic computing power capacity Semantic information adaptation relative entropy: For the computing power demand capacity of similar neural elements is

[0244] Normalization processing:

[0245] Obviously

[0246] Define the computing power demand capacity of similar neural elements and the electronic computing power capacity The semantic information adaptation relative entropy is denoted as

[0247]

[0248] (4) Store the semantic information adaptation relative entropy or into the semantic information dynamic neuron database

[0249] Refer to Figure 6 , in step S7, it includes:

[0250] S71: According to the computing power capacity set and the set weight coefficient, establish a relative entropy aggregation non-linear programming equation, substitute the semantic information adaptation relative entropy data set of the computing power node set into the relative entropy aggregation non-linear programming equation, calculate the global optimal solution, and determine whether the global optimal solution exists. If it exists, output the global optimal solution of the computing power task parsing adaptation;

[0251] S72: If not, perform an update on the state set of similar neural elements; after that, perform error correction and reconstruction training, and return to step S5 to obtain a new state set

[0252] In this embodiment, semantic adaptation relative entropy judgment feedback. Obtain the global optimal solution processing value of the semantic adaptation relative entropy value with respect to the provided computing power as the quantum computing power node set or the electronic computing power node set If there is an optimal solution, it indicates that the computing power providing node corresponding to the global optimal solution processing value is the optimal output result, which can be stored as the computing power task parsing adaptation information computing power node information for subsequent output; if there is no optimal solution, perform semantic information error correction and reconstruction

[0253] The steps of the semantic adaptation relative entropy judgment feedback algorithm are as follows, including:

[0254] (1) Retrieve the semantic information adaptation relative entropy in the semantic information dynamic neuron database

[0255] When the demand service application scenario is semantic information picture type and the request node is a quantum node or an electronic node, the computing power demand capabilities of the similar neural element quantum computing power nodes in the semantic information dynamic neuron database are retrieved respectively and the computing power capabilities for semantic information adaptation relative entropy or the computing power demand capabilities of the electronic computing power nodes and the computing power capabilities for semantic information adaptation relative entropy are retrieved respectively When the demand service application scenario is semantic information meaning type and the request node is a quantum node or an electronic node, the computing power demand capabilities of the similar neural element quantum computing power nodes in the semantic information dynamic neuron database are retrieved respectively

[0256] and the computing power capabilities for semantic information adaptation relative entropy or the computing power demand capabilities of the electronic computing power nodes and the computing power capabilities for semantic information adaptation relative entropy are retrieved respectively are retrieved respectively for semantic information adaptation relative entropy

[0257] (2) Establish a relative entropy aggregation non-linear programming equation to find the global optimal solution of the semantic adaptation entropy value. The semantic information picture type adaptation relative entropy or

[0258] and the semantic information meaning type adaptation relative entropy or are respectively substituted into the corresponding relative entropy aggregation non-linear programming to find the global optimal solution

[0259] When the demand service application scenario is semantic information picture type and the request node is a quantum node or an electronic node, establish a relative entropy aggregation non-linear programming (XP) equation to find the global optimal solution of the adaptation relative entropy value

[0260] (Ⅰ) Substitute the quantum node adaptation relative entropy into the relative entropy aggregation non-linear programming (XPQ1) to find the optimal adaptation solution of the quantum computing power node set The optimal adaptation solution

[0261]

[0262] w s is the evaluation value of the quantum node computing efficiency ability and the weight coefficient of x gi *is the global optimal solution for the non - linear programming problem (XPQ1).

[0263] (Ⅱ) Substitute the relative entropy of the electronic node adaptation into the relative entropy aggregation non - linear programming (XPB1) to find the optimal adaptation solution of the electronic computing power node set Optimal adaptation solution:

[0264]

[0265] w s is the estimated value of the computing efficiency of the electronic node of the weight coefficient, then x gi * is the global optimal solution for the non - linear programming problem (XPB1).

[0266] When the demand service application scenario is of the semantic information meaning type and the request node is a quantum node or an electronic node, establish the relative entropy aggregation non - linear programming (XP) equation to find the global optimal solution of the adaptation relative entropy value.

[0267] Just change the superscript in the equation (XPQ1) or (XPB1) from "1" to "2". Similarly, for the computing power demand of the similar neural - based quantum computing power node the semantic information adaptation relative entropy or the computing power demand of the similar neural - based electronic computing power node and the computing power the semantic information adaptation relative entropy Substitute them into the above - mentioned equation for solution, and the judgment of the global optimal solution can be obtained (not elaborated here). the semantic information adaptation relative entropy

[0268] (3) Judge the global optimal solution of the equation. For the semantic information picture type, semantic meaning type service demands and quantum computing power nodes, electronic computing power nodes, obtain the non - linear programming (XP) equation. If there is an optimal solution, output it; otherwise, perform semantic information error correction and reconstruction.(3) Judge the global optimal solution of the equation. For the semantic information picture type, semantic meaning type service demands and quantum computing power nodes, electronic computing power nodes, obtain the non - linear programming (XP) equation. If there is an optimal solution, output it; otherwise, perform semantic information error correction and reconstruction.

[0269] Refer to Figure 7 In step S72, it includes:

[0270] S721: Identify the semantic information of the dynamic neuron, obtain a new feature information data set, calculate the semantic information entropy of the two identifications according to the data set obtained in step S5, and calculate the average value of the semantic information entropy as the conditional entropy;

[0271] S722: According to the service demand application scenario, calculate the minimum conditional entropy of the computing power node under different semantic information types as the data set of the new dynamic neuron, and return to step S5.

[0272] In this embodiment, for the AI to correct and reconstruct semantic information, an AI deep learning error correction algorithm is used to perform a secondary scan on the service demand task scenario as the semantic information source, and a dynamic neuron for the i-th demand node of the s-th subnet, where the l-th type of semantic information is picture type, meaning type, quantum node or electronic node, is obtained. Or A new feature information data set Or Perform error correction and reconstruction training on the new feature information data set to obtain a new set of dynamic neuron feature states after reconstruction, transfer to step S5, and repeat the further similarity training of the AI on the dynamic neurons.

[0273] The algorithm flow steps for the AI to correct and reconstruct semantic information training are as follows, including:

[0274] (1) The AI performs a secondary scan on the dynamic semantic information to obtain a new set of feature information measurement. When the service scenario is a semantic information picture type and the request node is a quantum or electronic node, the AI performs a secondary scan on the dynamic neuron Or Separate new feature information data sets are obtained Or

[0275] (2) Retrieve the feature information state set of the first scan and calculate the semantic information entropy of the two scans. Due to the uncertainty in the task scenario classification or the deep learning process, the feature information data sets obtained from the two scans of the dynamic neuron may be the same or different. To characterize this uncertainty of the dynamic neuron Or , it is represented by the semantic information entropy. The smaller the entropy value, the lower the uncertainty and the higher the accuracy.

[0276] When the service demand task scenario is a semantic information source picture type and the request node is a quantum or electronic node, the dynamic neuron Or The feature information data sets obtained from the two scans Or And the new feature information data set Or The feature information measure value Or May be the same or different. The uncertainty of the feature states of the two scans of the dynamic neuron Or Is represented by the semantic information entropy Or As:

[0277]

[0278] When the service demand task scenario is a semantic information meaning class and the requesting node is a quantum (or electronic) node, the dynamic neuron or The feature information data set obtained by two scans or and the new feature information data set or The semantic information entropy or is expressed as:

[0279]

[0280] Among them, or is the feature information state set obtained when the i-th demand node of the s-th subnet in the first scan is a quantum node or an electronic node, and the l-th semantic information source is a picture class or a meaning class or The feature information state set of the corresponding neural element.

[0281] (3) Calculate the conditional entropy after the second scan. The semantic information entropy or gives the uncertainty of each measure on each feature information component. To characterize the overall uncertainty of the feature state measure set after the second scan, the average value of the semantic information entropy is used to characterize it, which is defined as the conditional entropy.

[0282] When the given demand scenario is a semantic information picture class, a quantum node or an electronic node, the average value of the semantic information entropy (or ) (or ) is called the conditional entropy of the dynamic neuron (or ) of the given demand scenario picture class quantum (or electronic) node (or ):

[0283]

[0284] Judgment condition: When and are the same, Since Take:

[0285]

[0286] So there is:

[0287] Among them, It is the deep learning result of the first scan, that is, the probability measure value that can be adapted. Otherwise,

[0288] or

[0289] Judgment condition: When and are the same, Since Take:

[0290]

[0291] So there is:

[0292]

[0293] Among them, is the deep learning result of the first scan, that is, the probability measure value that can be adapted. Otherwise,

[0294] When the given demand scenario is of the semantic information meaning type, quantum node or electronic node, the semantic information entropy (or )'s average value (or ) is called the dynamic neuron of the semantic information meaning type quantum (or electronic) node of the service demand application scenario (or )'s conditional entropy (or ):

[0295]

[0296] Judgment condition: When and are the same, Since Take:

[0297]

[0298] So there is:

[0299]

[0300] Among them, is the deep learning result of the first scan, that is, the probability measure value that can be adapted. Otherwise,

[0301] or

[0302] Judgment condition: When is the same as , Since Take:

[0303]

[0304] So there is:

[0305]

[0306] Among them, is the deep learning result of the first scan, that is, the probability measure value that can be adapted. Otherwise,

[0307] (4) Calculate the minimum conditional entropy of the requested quantum (or electronic) node semantic picture class or semantic information class. Select the minimum conditional entropy among the l (l = 1, 2,...) class semantic information picture classes and meaning classes for l (l = 1, 2,...), and define the minimum conditional entropy of the semantic information picture class quantum (or electronic) node as (or ), and the minimum conditional entropy of the semantic information meaning class is or

[0308] Minimum conditional entropy of semantic information picture class quantum node

[0309]

[0310] Minimum conditional entropy of semantic information picture class or electronic node

[0311]

[0312] Minimum conditional entropy of semantic information meaning class quantum node

[0313]

[0314] Or the minimum conditional entropy of the semantic information meaning class electronic node

[0315]

[0316] (5) Reconstructed dynamic neuron feedback training. For the dataset of the new dynamic neuron feature information after reconstruction (or ), (or )The feedback is transferred to step S5 to restart the dynamic neuron AI similarity training.

[0317] In summary, the beneficial effects of the present invention are as follows:

[0318] 1. A hybrid computing power task parsing and adaptation method based on semantic information is designed. When the semantic information is picture type, meaning type, and the nodes are quantum nodes or electronic nodes, methods for constructing neural primitives of semantic information and dynamic neuron data are given; from the perspective of the whole process and engineering, a method for constructing a training and inference neural network is given; through steps such as calibrating dynamic neuron classification labels, scanning and extracting dynamic semantic information, performing similarity training on dynamic neurons to obtain similar neural primitives, performing inference training on similar neural primitives, judging and feedback of semantic adaptation relative entropy, and error correction and reconstruction of semantic information, the hybrid computing power task parsing and adaptation for semantic information is realized.

[0319] 2. By determining the segmentation granularity of semantic information, the purpose of segmenting semantic information and object detection is achieved; by giving prediction models for establishing true distribution probability and prediction distribution probability and weighted cross-entropy models for computing power task parsing and adaptation for semantic information, and selecting the fuzzy minimum weighted cross-entropy for target evaluation as the optimal similar neural primitives for the data of secondary training.

[0320] 3. By extracting the computing power capability set corresponding to the computing power node set, a semantic information adaptation relative entropy calculation method is given, realizing the mapping between the data set of dynamic neurons and the data set of neural primitives.

[0321] 4. Using the state set of similar neural primitives after training of the neural network as input for secondary training, a one-to-one mapping relationship is established between the request node and the computing power capability set of the computing power node, and the optimal solution for computing power adaptation of unknown semantic information can be obtained through known semantic information.

[0322] 5. An algorithm for semantic information adaptation relative entropy is given. A semantic information adaptation relative entropy model for computing power task parsing is established, and a method for finding the optimal solution of the semantic adaptation entropy value global optimal solution processing value of the relative entropy aggregation nonlinear programming equation for computing power task parsing is given.

[0323] 6. A method for error correction and reconstruction of semantic information. The service demand application scenario for semantic information is scanned twice to obtain a new feature information data set of dynamic neurons for error correction and reconstruction training, and the state set of the new dynamic neurons after reconstruction is obtained, and further computing power task parsing and adaptation are carried out, improving the timeliness, accuracy, reliability and overall efficiency of the computing power service.

[0324] Refer to Figure 8, and further includes an embodiment of a computing power task parsing and adaptation control system, where the control system is applicable to any of the parsing and adaptation methods in the above technical solutions, including:

[0325] A data storage module 1 for storing the characteristic information data sets, computing power demand ability measure sets, and state sets corresponding to neural elements, dynamic neurons, and similar neural elements;

[0326] A neural network training module 2 for performing neural network training iterations based on the data in the data storage module to determine the state set of similar neural elements;

[0327] A calibration module 3 for classifying and labeling the types of semantic information of dynamic neurons according to the service demand application scenario;

[0328] A semantic information extraction module 4 for classifying according to the computing power request of the service demand application scenario based on the type of semantic information, identifying the characteristic information data set and the computing power demand ability measure set, and storing them in the data storage module;

[0329] A judgment and feedback module 5 for solving and training the global optimal solution processing value of the semantic information adaptation relative entropy with respect to the provided computing power being a quantum node or an electronic node;

[0330] An error correction and reconstruction module 6 for updating and identifying the characteristic information, performing error correction and reconstruction training according to the service demand application scenario, and obtaining a new state set.

[0331] In this embodiment, the data storage module 1 is used to store neural elements and the corresponding characteristic information data sets and computing power demand ability measure sets. The neural elements defined as the semantic information source according to the semantic information "basic elements" of categories such as semantic information pictures and meanings form a standard library of semantic information source neural elements with the data set composed of neural elements.

[0332] The neural network training module 2 is used to provide support for subsequent AI training and inference of dynamic neurons. The deep neural network is jointly composed of neural elements, dynamic neurons formed by any request node with the service application scenario state being a picture type or a meaning type at any time being a quantum node or an electronic node, and a quantum-electronic hybrid computing power network with the provided computing power node being a quantum node or an electronic node.

[0333] A feature information data set for extracting semantic information and a computing power requirement ability measurement set are used to determine the segmentation granularity of semantic information, perform object detection, obtain a feature information state set through computing power learning and training according to the request node, and evaluate and calculate the target dynamic neurons to obtain similar neural primitives; determine the computing power requirement value of the similar neural primitives, and use the feature information state set and computing power ability data set of the similar neural primitives as the feature information state set and computing power ability requirement data set of the dynamic neurons, and store them in the semantic information dynamic neuron database for subsequent secondary training of the dynamic neurons.

[0334] Perform secondary training on the feature state set of similar neural primitives of the trained neural network with respect to computing power nodes being quantum or electronic to obtain the optimal computing power task adaptation node.

[0335] The calibration module 3 is used to calibrate the classification labels of dynamic neurons according to service demand application scenarios such as semantic information picture classes, meaning classes, etc., and store the dynamic neurons calibrated with classification labels according to service demand application scenarios in the semantic information dynamic neuron database.

[0336] The semantic information extraction module 4 is used to, according to the computing power request of the service demand application scenario, based on the semantic information source being a picture class or a meaning class, perform deep learning scanning, recognition and analysis on the dynamic neurons of the service demand application scenario by AI, determine the dynamic neuron feature information data set and computing power requirement ability value, and store them in the semantic information dynamic neuron database to prepare training data for subsequent determination of similar neural primitives.

[0337] The judgment and feedback module 5 is used to solve and train the semantic adaptation relative entropy value with respect to the global optimal solution processing value of the computing power providing node being a quantum node or an electronic node. If there is a global optimal solution, it indicates that the computing power providing node corresponding to the global optimal solution processing value is the optimal output result, which can be stored as the computing power task parsing adaptation information computing power node information for subsequent output; if there is no optimal solution, correct and reconstruct the semantic information.

[0338] The error correction and reconstruction module 6 is used to perform error correction and reconstruction training on the feature states of the service demand task scenario by using the AI deep learning error correction algorithm, obtain a new feature state data set of dynamic neurons for the quantum node or electronic node of the picture class or meaning class with respect to the request node, transfer to step S5 for retraining to obtain a new state set, and enhance the accuracy of computing power task adaptation.

[0339] There is also an embodiment of a computer-readable storage medium that stores a computer program for running the parsing and adaptation method, wherein the computer program causes the computer to perform the following steps:

[0340] S1: Establish a data set of neural primitives of semantic information according to the type of semantic information;

[0341] S2: Identify the dynamic neurons for semantic information, obtain the set of request nodes for the computing power task, and establish a neural network by combining the set of computing power nodes corresponding to the set of request nodes.

[0342] S3: Classify the dynamic neurons according to the type of semantic information, and calibrate the set of request nodes according to the application scenario of the service demand.

[0343] S4: Identify and analyze the dynamic neurons according to the type of semantic information to determine the data set of the dynamic neurons.

[0344] S5: According to the data set of the neural primitives, identify and extract the data set of the dynamic neurons, determine the segmentation granularity of the semantic information of the dynamic neurons, and train through the neural network according to the set of request nodes to update the state set of the dynamic neurons. Obtain the similar neural primitives of the dynamic neurons through the computing power task parsing algorithm, and train according to the similar neural primitives to obtain the state set of the similar neural primitives.

[0345] S6: Use the trained neural network, input the state set of the similar neural primitives for training, and calculate the semantic information adaptation relative entropy data set between the computing power nodes and the type of semantic information.

[0346] S7: Calculate the global optimal solution according to the semantic information adaptation relative entropy data set, and determine whether the global optimal solution exists. If it exists, output the global optimal solution.

[0347] Among them, the computer-readable storage medium can be a computer storage medium or a communication medium. The communication medium includes any medium that facilitates the transmission of a computer program from one place to another. The computer storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer. For example, the computer-readable storage medium is coupled to the processor, so that the processor can read information from the computer-readable storage medium and write information to the computer-readable storage medium. Of course, the computer-readable storage medium can also be a component of the processor. The processor and the computer-readable storage medium can be located in an application specific integrated circuit (ASIC). In addition, the ASIC can be located in the user equipment. Of course, the processor and the computer-readable storage medium can also exist as discrete components in the communication device.

[0348] Specifically, the computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically-Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable read-only memory (PROM), Read-Only Memory (ROM), magnetic memory, flash memory, a magnetic disk, or an optical disk. The storage medium can be any available medium accessible by a general-purpose or special-purpose computer.

[0349] An embodiment of an electronic device is further provided, including:

[0350] One or more processors; a memory; and

[0351] One or more programs, where one or more programs are stored in the memory and are configured to be executed by one or more processors. The programs include steps for performing the following:

[0352] S1: Establish a data set of neural primitives of semantic information according to the type of semantic information;

[0353] S2: Identify the dynamic neurons of semantic information to obtain a set of request nodes for computing power tasks, and combine the set of computing power nodes corresponding to the set of request nodes to establish a neural network;

[0354] S3: Classify the dynamic neurons according to the type of semantic information, and calibrate the set of request nodes according to the service demand application scenario;

[0355] S4: Identify and analyze the dynamic neurons according to the type of semantic information to determine the data set of dynamic neurons;

[0356] S5: According to the data set of neural primitives, identify and extract the data set of dynamic neurons, determine the segmentation granularity of the semantic information of the dynamic neurons, train through the neural network according to the set of request nodes, update the state set of the dynamic neurons, obtain the similar neural primitives of the dynamic neurons through the computing power task parsing algorithm, and train according to the similar neural primitives to obtain the state set of the similar neural primitives;

[0357] S6: Using the trained neural network, input the state sets of similar neural primitives for training, and calculate the semantic information adaptation relative entropy dataset between the computing power nodes and the types of semantic information;

[0358] S7: Calculate the global optimal solution based on the semantic information adaptation relative entropy dataset, and determine whether the global optimal solution exists. If it exists, output the global optimal solution.

[0359] A memory for storing computer programs. This memory may include high-speed random access memory (RAM), and may also include non-volatile memory (NVM), such as at least one disk memory, and can also be a USB flash drive, a portable hard drive, a read-only memory, a disk, or an optical disc, etc.

[0360] A processor for executing the computer programs stored in the memory to implement the parsing and adaptation method in the above embodiments. This processor can be a central processing unit (CPU), or can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or this processor can also be any conventional processor, etc. The steps of the method disclosed in combination with the invention can be directly implemented by the execution of the hardware processor, or can be implemented by the combination of the hardware and software modules in the processor.

[0361] Optionally, the memory can be either independent or integrated with the processor.

[0362] When the memory is a device independent of the processor, the electronic device can also include a bus. This bus is used to connect the memory and the processor. This bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc.

[0363] It should be noted that through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the essence of the above technical solution or the part that contributes to the prior art can be embodied in the form of a software product, and this computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments. In this article, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, the element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article or device including the said element.

[0364] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A computing task parsing and adaptation method, characterized in that: include: S1: according to the type of semantic information, establish a data set of neural primitives of the semantic information; S2: Identify the dynamic neurons of the semantic information, obtain the request node set of the computing task, and establish a neural network by combining the computing node set corresponding to the request node set; S3: classifying the dynamic neurons according to the type of the semantic information, and calibrating the request node set according to the service demand application scenario; S4: performing identification analysis on the dynamic neuron according to the type of the semantic information to determine a data set of the dynamic neuron; S5: According to the data set of the neural primitives, identify and extract the data set of the dynamic neurons, determine the segmentation granularity of the semantic information of the dynamic neurons, train through the neural network according to the request node set, update the state set of the dynamic neurons, obtain similar neural primitives of the dynamic neurons through the computing task parsing algorithm, and train according to the similar neural primitives to obtain the state set of similar neural primitives; S6: using the trained neural network, inputting the state set of the similar neural primitives for training, and calculating the semantic information adaptation relative entropy data set between the computing power nodes and the types of semantic information; S7: Calculate the global optimal solution according to the semantic information adaptation relative entropy data set, determine whether the global optimal solution exists, and if so, output the global optimal solution.

2. The analysis and adaptation method according to claim 1, characterized in that: In step S1, it includes: S11: Semantic information is divided into image information and meaning information, and the neural primitives of the semantic information are identified to obtain feature information and computing power requirement information of the image neural primitives and the meaning neural primitives respectively; S12: Establish feature information datasets and computing power requirement capability measurement sets of image-based neural primitives and meaning-based neural primitives, calculate the union, and obtain the dataset of neural primitives.

3. The analysis and adaptation method according to claim 1, characterized in that: In step S2, it includes: Identify the dynamic neurons of the semantic information, obtain quantum request nodes or electronic request nodes, determine the corresponding quantum computing power nodes or electronic computing power nodes according to the quantum request nodes or electronic request nodes, and establish a neural network according to the quantum request nodes, electronic request nodes, quantum computing power nodes, electronic computing power nodes and information transmission network.

4. The analysis and adaptation method according to claim 1, characterized in that: In step S3, it includes: According to the type of semantic information, the request nodes of the sub-network in the neural network at any time are identified, and the request nodes that meet the service demand application scenario are calibrated to obtain a calibrated request node set.

5. The analysis and adaptation method according to claim 1, characterized in that: In step S4, it includes: S41: Scan the dynamic neuron, select a calibrated request node, obtain the data set of the dynamic neuron under the request node, and obtain the characteristic information data set of the dynamic neuron; S42: Determine a feature information data set of the dynamic neuron according to the service demand application scenario, analyze it with the feature information data set of the neural primitive, and obtain a state set of the dynamic neuron.

6. The analysis and adaptation method according to claim 1, characterized in that: In step S5, it includes: S51: Segmenting the semantic information and detecting the target according to the feature information data set, and determining the segmentation granularity of the semantic information; S52: Taking the neural primitive as the real distribution and the dynamic neuron as the predicted distribution, a prediction model of the real distribution probability and the predicted distribution probability is established, and the state set of the neural primitive adapted to the dynamic neuron is obtained according to the state set of the dynamic neuron; S53: taking the characteristic information data set of the dynamic neuron as the prediction distribution, performing probability distribution training, obtaining the prediction distribution probability, and performing normalization processing; The real distribution of the neural primitives adapted to the dynamic neurons is indexed, the real distribution probability is calculated, and normalized; S54: Use weighted cross entropy to measure the difference between the actual distribution probability of the neural motif and the predicted distribution probability of the dynamic neuron, use the weighted cross entropy as neural network training data, train to obtain a state set of similar neural motifs, and select the minimum weighted cross entropy as the optimal similar neural motif.

7. The analysis and adaptation method according to claim 1, characterized in that: In step S6, it includes: S61: extracting the state set, feature information data set and computing power requirement capability measurement set of the neural primitive corresponding to the optimal similar neural primitive as training data, and performing secondary training on the neural network; S62: Determine a computing power capability set corresponding to the computing power node set, and calculate the semantic information adaptation relative entropy according to the service demand application scenario and the type of semantic information to obtain a semantic information adaptation relative entropy data set.

8. The analysis and adaptation method according to claim 1, characterized in that: In step S7, it includes: S71: According to the computing power capability set and the set weight coefficient, a relative entropy aggregation nonlinear programming equation is established, and the semantic information of the computing power node set is adapted to the relative entropy data set and substituted into the relative entropy aggregation nonlinear programming equation to calculate the global optimal solution, and determine whether the global optimal solution exists. If so, the global optimal solution adapted to the computing power task analysis is output; S72: If it does not exist, the state set of the similar neural primitives is updated, error correction and reconstruction training is performed, and the process returns to step S5 to obtain a new state set.

9. The analysis and adaptation method according to claim 8, characterized in that: In step S72, it includes: S721: Identify the semantic information of the dynamic neuron, obtain a new feature information data set, calculate the semantic information entropy of the two identifications according to the data set obtained in step S5, and calculate the average value of the semantic information entropy as the conditional entropy; S722: According to the service demand application scenario, calculate the minimum conditional entropy of the computing power node under different types of semantic information as the data set of the new dynamic neuron, and return to step S5.

10. A computing task analysis and adaptation control system, characterized in that: The control system is applicable to the analytical adaptation method according to any one of claims 1 to 9, including: A data storage module, used to store feature information data sets, computing power requirement capability measurement sets and state sets corresponding to neural primitives, dynamic neurons and similar neural primitives; A neural network training module, which performs neural network training iterations according to the data in the data storage module to determine a state set of similar neural primitives; The calibration module classifies and labels the types of semantic information of dynamic neurons according to the service demand application scenarios; The semantic information extraction module is used to classify the computing power request according to the service demand application scenario based on the type of semantic information, identify the feature information data set and the computing power demand capability measurement set, and store them in the data storage module; The judgment feedback module is used to find the solution training of semantic information adaptation relative entropy with respect to the global optimal solution processing value provided by the computing power of quantum nodes or electronic nodes; The error correction and reconstruction module updates and identifies the feature information, performs error correction and reconstruction training according to the service demand task scenario, and obtains a new state set.

11. A computer-readable storage medium, characterized in that: It stores a computer program for the parsing and adaptation method, wherein the computer program enables a computer to execute the parsing and adaptation method as described in any one of claims 1-9.

12. An electronic device, characterized in that: include: one or more processors; Memory; as well as One or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include instructions for executing the parsing adaptation method according to any one of claims 1 to 9.

Images