Task processing method and system for neural network model, storage medium and electronic device

Abstract

The application discloses a task processing method for a neural network model, a storage medium and an electronic device, relates to the technical field of smart homes, and the task processing method for the neural network model comprises the following steps: obtaining a task demand, and selecting a target expert model from a mixed expert model according to a routing decision mechanism; wherein the task demand comprises input data, and the routing decision mechanism is used to route an expert model that is suitable for a task corresponding to the task demand from the mixed expert model; input data is written into the target expert model for model training, and a model output result is obtained; and a task execution strategy corresponding to the model output result is driven to be executed according to the model output result. The method can improve the efficiency and response speed of task processing. The application further discloses a task processing system for a neural network model.

Description

Technical Field

[0001] This application relates to the field of smart homes, and more specifically, to a task processing method and system for neural network models, a storage medium, and an electronic device. Background Technology

[0002] Currently, with the rapid development of IoT and AI technologies, smart home systems are rapidly becoming widespread in users' homes. A smart home system mainly consists of a central processing unit for controlling home appliances, sensors for detecting different types of data, and actuators for executing device commands. However, current smart home systems employ a single model, which means they excel at handling specific tasks, but their performance may degrade when handling different types of or concurrent tasks.

[0003] To address the issue of system performance degradation when handling different types or concurrent tasks, a task processing method for smart homes has been disclosed, which includes: implementing automation strategies based on cloud computing services or local centralized intelligent control units, and using standard machine learning algorithms for task processing.

[0004] In the process of implementing the embodiments of this disclosure, at least the following problems were found in the related art:

[0005] While the relevant technologies can maintain system performance to a certain extent when handling different types or concurrent tasks, they also require sending task-related data to a cloud server for processing. This prevents the intelligent control unit from processing tasks quickly, resulting in longer response times and impacting task processing efficiency.

[0006] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] To provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. This summary is not intended as a general commentary, nor is it intended to identify key / important components or describe the scope of protection of these embodiments, but rather as a prelude to the detailed description that follows.

[0008] This disclosure provides a task processing method, system, storage medium, and electronic device for neural network models to improve task processing efficiency and response speed.

[0009] In some embodiments, the method includes: obtaining task requirements; selecting a target expert model from a hybrid expert model according to a routing decision mechanism; wherein the task requirements include input data, and the routing decision mechanism is used to route an expert model from the hybrid expert model that is compatible with the task requirements; writing the input data to the target expert model for model training to obtain model output results; and driving the execution of a task execution strategy corresponding to the model output results according to the model output results.

[0010] In some embodiments, selecting a target expert model from hybrid expert models according to a routing decision mechanism includes: extracting semantic data from input data and determining semantic complexity; the semantic complexity is determined based on the semantic data and a semantic complexity evaluation function; determining the matching degree between the task requirements and the application domain of each model based on the semantic data and the model application domain of each expert model; weighting the matching degree between the task requirements and the application domain of each model with the semantic complexity to determine the evaluation score of each expert model; and selecting the expert model with the highest evaluation score as the target expert model.

[0011] In some embodiments, determining the matching degree between task requirements and each model application domain based on semantic data and the model application domain of each expert model includes: determining the professional domain corresponding to the task requirements based on semantic data; obtaining the model application domain of each expert model; and determining the matching degree between task requirements and each model application domain.

[0012] In some embodiments, selecting a target expert model from the hybrid expert models according to the routing decision mechanism includes: obtaining the accuracy and average response time of each expert model; determining the performance evaluation score PerScore of each expert model based on the accuracy and average response time of each expert model; obtaining the average load AverLoad of each expert model; and selecting the expert model with the highest PerScore / (AverLoad+1) as the target expert model.

[0013] In some embodiments, determining the performance evaluation score of each expert model based on its accuracy and average response time includes: weighting the accuracy and the reciprocal of the average response time of each expert model to obtain the performance evaluation score of each expert model.

[0014] In some embodiments, obtaining the average load of each expert model includes: obtaining the current load and historical performance evaluation score of each expert model; and determining the average load of each expert model based on historical load data and the number of loads.

[0015] In some embodiments, the task execution strategy corresponding to the model output result is driven to be executed based on the model output result, including: determining the target executor and the task execution strategy based on the model output result; and sending control instructions to the target executor to make the target executor execute the task execution strategy.

[0016] In some embodiments, the system includes: a smart home device configured to collect device input data; a router configured to obtain task requirements, select a target expert model from a hybrid expert model according to a routing decision mechanism, and write input data to the target expert model for model training to obtain model output results; wherein, the task requirements include input data, which includes device input data and / or user input data, and the routing decision mechanism is used to route an expert model from the hybrid expert model that is compatible with the task corresponding to the task requirements; a control unit configured to drive the execution of a task execution strategy corresponding to the model output results according to the model output results; and an executor configured to execute the task execution strategy corresponding to the model output results.

[0017] In some embodiments, the storage medium, the computer-readable storage medium, includes a stored program, wherein the program, when executed, performs the task processing method for a neural network model as described in any of the preceding embodiments.

[0018] In some embodiments, the electronic device includes a memory and a processor, the memory storing a computer program, and the processor being configured to execute the task processing method for a neural network model as described above through the computer program.

[0019] The task processing method, system, storage medium, and electronic device for neural network models provided in this disclosure can achieve the following technical effects:

[0020] This embodiment of the disclosure selects a suitable expert model for the task from a hybrid expert model based on the local routing decision of the router. The target expert model is then used to train the input data to obtain the model output. The corresponding task execution strategy is then driven by the control unit to ensure a high degree of adaptation between the task execution strategy and the task requirements. Furthermore, this embodiment eliminates the need to upload the input data containing the task requirements to a cloud server for processing; task execution can be achieved through the collaboration of the router and control unit. This improves task processing efficiency and response speed, thereby giving the smart home system better adaptability.

[0021] The above general description and the description below are exemplary and illustrative only and are not intended to limit this application. Attached Figure Description

[0022] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0023] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the hardware environment for a task processing method for a neural network model according to an embodiment of this application;

[0025] Figure 2 This is a schematic diagram of a task processing method for a neural network model provided in an embodiment of this disclosure;

[0026] Figure 3 This is a schematic diagram of another task processing method for a neural network model provided in an embodiment of this disclosure;

[0027] Figure 4 This is a schematic diagram of another task processing method for a neural network model provided in an embodiment of this disclosure;

[0028] Figure 5 This is a schematic diagram of another task processing method for a neural network model provided in an embodiment of this disclosure;

[0029] Figure 6 This is a schematic diagram of another task processing method for a neural network model provided in an embodiment of this disclosure;

[0030] Figure 7 This is a schematic diagram of a task processing system for a neural network model provided in an embodiment of this disclosure;

[0031] Figure 8 This is a schematic diagram of another task processing system for neural network models provided in an embodiment of this disclosure.

[0032] Figure label:

[0033] 102: Electronic devices; 104(201): Smart home devices; 202: Routers;

[0034] 203: Control unit; 204: Actuator;

[0035] 70: Task processing systems for neural network models;

[0036] 700: Processor; 701: Memory; 702: Communication interface; 703: Bus. Detailed Implementation

[0037] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0038] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0039] According to one aspect of the embodiments of this application, a task processing method for neural network models is provided. This task processing method for neural network models is widely applicable to whole-house intelligent digital control application scenarios such as smart homes, smart home ecosystems, and intelligence house ecosystems. Optionally, in this embodiment, the above-mentioned task processing method for neural network models can be applied to, for example... Figure 1 The hardware environment shown consists of electronic device 102 and smart home device 104. For example... Figure 1 As shown, smart home device 104 is connected to electronic device 102 via a network.

[0040] The aforementioned networks may include, but are not limited to, at least one of the following: wired network, wireless network. The aforementioned wired network may include, but is not limited to, at least one of the following: wide area network, metropolitan area network, local area network. The aforementioned wireless network may include, but is not limited to, at least one of the following: Wi-Fi (Wireless Fidelity), Bluetooth. Electronic device 102 may not be limited to PCs, mobile phones, or tablets. Smart home devices 104 may not be limited to smart air conditioners, smart range hoods, smart refrigerators, smart ovens, smart stoves, smart washing machines, smart water heaters, smart washing equipment, smart dishwashers, smart projectors, smart TVs, smart clothes racks, smart curtains, smart audio-visual equipment, smart sockets, smart speakers, smart speakers, smart fresh air systems, smart kitchen and bathroom equipment, smart bathroom equipment, smart robot vacuum cleaners, smart window cleaning robots, smart mopping robots, smart air purifiers, smart steam ovens, smart microwave ovens, smart water heaters, smart air purifiers, smart water dispensers, smart door locks, etc.

[0041] Combination Figure 2 As shown, this disclosure provides a task processing method for a neural network model, including:

[0042] S01, the router obtains the task requirements and selects a target expert model from the hybrid expert models according to the routing decision mechanism. The task requirements include input data, and the routing decision mechanism is used to route the expert model from the hybrid expert models that best matches the task requirements.

[0043] In this step, the input data includes user input data sent by the user and / or device input data sent by smart home devices. As an example, user input data includes voice data or touch data. Device input data includes environmental data or audio data. Environmental data includes some or all of temperature, humidity, wind speed, and environmental quality data. Environmental data is obtained from different types of sensors. After receiving the task request, the router can indirectly determine the user's subjective requirements and / or the objective device data transmitted by the devices based on the input data included in the task request, thereby accurately determining the task requirements. The hybrid expert model is composed of multiple specialized sub-models (i.e., expert models), each expert model exhibiting good performance in its area of ​​expertise.

[0044] S02, the router writes input data to the target expert model for model training and obtains the model output result. It should be noted that after writing input data to the target expert model for training and obtaining the model output result, the router also sends the model output result to the control unit.

[0045] S03, the control unit drives the execution of the task execution strategy corresponding to the model output result based on the model output result.

[0046] The task processing method for neural network models provided in this disclosure involves obtaining task requirements and then selecting a target expert model from a hybrid expert model based on a routing decision mechanism to determine the appropriate expert model for the task requirements. Input data is then written to the target expert model for training, yielding the model output. Finally, the control unit receives the model output and executes the task execution strategy corresponding to that output. This disclosure utilizes the router's local routing decision to select the appropriate expert model from the hybrid expert model, trains the input data using the target expert model to obtain the model output, and drives the corresponding task execution strategy through the control unit to ensure a high degree of adaptation between the task execution strategy and the task requirements. Furthermore, this disclosure eliminates the need to upload the input data contained in the task requirements to a cloud server for processing; task execution can be achieved through the collaboration of the router and control unit. This improves task processing efficiency and response speed, thereby enhancing the adaptive capabilities of the smart home system.

[0047] Optionally, combined Figure 3 As shown, the router selects the target expert model from the hybrid expert models according to the routing decision mechanism, including:

[0048] S11, the router extracts semantic data from the input data and determines the semantic complexity. The semantic complexity is determined based on the semantic data and the semantic complexity evaluation function.

[0049] In this step, the router determines semantic complexity as follows: The router parses the semantic data to obtain multiple words that construct the semantic data, as well as the frequency and part-of-speech tag of each word in the semantic data; the router inputs each word, its frequency, and its part-of-speech tag into the semantic complexity evaluation function to determine the semantic complexity evaluation value of each word; the router sums the speech complexity evaluation values ​​of all words to obtain the semantic complexity. The semantic complexity evaluation function is pre-defined, with the word frequency and part-of-speech tag as independent variables and the speech complexity evaluation value as the dependent variable.

[0050] S12, the router determines the matching degree between the task requirements and the application domain of each model based on the semantic data and the model application domain of each expert model.

[0051] In this step, the application domain of the expert model can be obtained by querying a model application domain database. This database stores different expert models and their corresponding application domains. As an example, application domains may include speech recognition, natural language processing, image recognition, environmental perception, or security monitoring.

[0052] S13, the router performs weighted processing on the matching degree between the task requirements and the application domain of each model, as well as the semantic complexity, to determine the evaluation score of each expert model.

[0053] In this step, the router weights the matching degree between the task requirements and the application domain of each model, as well as the semantic complexity, to determine the evaluation score for each expert model, including:

[0054] Score(i)=α×Complexity+β×DMatch(i);

[0055] Where i = 1, 2, ..., N, Score(i), Complexity, and DMatch represent the evaluation score, semantic complexity, and matching degree between the task requirements and the application domain of the i-th expert model, respectively; α and β represent the first and second weight coefficients, respectively; and N represents the total number of expert models in the hybrid expert model. α + β = 1, 0 < α < 1 and 0 < β < 1.

[0056] S14, the router selects the expert model with the highest evaluation score as the target expert model.

[0057] In this step, the router selects the expert model with the highest evaluation score as the target expert model. This includes: the router sorts the evaluation scores of all expert models, selects the expert model with the highest evaluation score based on the sorting results, and determines it as the target expert model.

[0058] In this embodiment, after receiving the task requirements, semantic data is first extracted from the input data and the semantic complexity is determined. Then, based on the semantic data and the application domain of each expert model, the matching degree between the task requirements and the application domain of each model is determined. Finally, the semantic complexity and the matching degree between the task requirements and the application domain of each model are weighted to determine the evaluation score of each expert model, and the expert model with the highest evaluation score is selected as the target expert model. Thus, this embodiment evaluates the expert model scores by comprehensively considering both semantic complexity and the matching degree between the task requirements and the application domain of each model, ensuring the reliability of the evaluation scores and improving the adaptability of the target expert model to the task requirements.

[0059] Optionally, the router determines the matching degree between the task requirements and the application domain of each expert model based on semantic data and the model application domain of each model, including:

[0060] The router determines the professional field corresponding to the task requirements based on semantic data.

[0061] The router obtains the application domain of each expert model and determines the degree of matching between the task requirements and the application domain of each model.

[0062] Thus, this embodiment first determines the professional field corresponding to the task requirements based on semantic data. Then, it obtains the application field of each expert model and determines the matching degree between the patent field of the task requirements and each application field. Determining the matching degree helps obtain accurate evaluation scores, thereby facilitating the selection of a suitable target expert model for the task requirements.

[0063] Optionally, combined Figure 4 As shown, the router selects the target expert model from the hybrid expert models according to the routing decision mechanism, including:

[0064] S21, the router obtains the accuracy and average response time of each expert model.

[0065] S22, the router determines the performance evaluation score PerScore for each expert model based on the accuracy and average response time of each expert model.

[0066] In this step, the accuracy of the expert model is positively correlated with the performance evaluation score, while the average response time of the expert model is negatively correlated with the performance evaluation score.

[0067] S23, the router obtains the average load (AverLoad) for each expert model.

[0068] S24, the router selects the expert model with the highest PerScore / (AverLoad+1) as the target expert model.

[0069] Thus, the present invention can select a target expert model by combining the performance evaluation score and average load of the expert model. The selected target expert model has a relatively low average load and a high performance evaluation score compared to other expert models, which makes the selected target expert model have a shorter response time and is conducive to improving the efficiency of task processing.

[0070] Optionally, the router determines a performance evaluation score for each expert model based on its accuracy and average response time, including:

[0071] The router weights the accuracy of each expert model with the reciprocal of its average response time to obtain a performance evaluation score for each expert model.

[0072] Thus, this embodiment of the disclosure can comprehensively evaluate the performance of the model by combining the accuracy and average response time of the expert model in the historical operation phase, thereby improving the accuracy of the calculation of the expert model performance evaluation score and ensuring that the selection of the target expert model is sufficiently reliable.

[0073] As an example, PerScore = γ × Accuracy + δ × (1 / ResTime);

[0074] Where PerScore, Accuracy, and ResTime represent the performance evaluation score, accuracy, and average response time of the expert model, respectively, and γ and δ represent the third and fourth weighting coefficients, respectively.

[0075] γ+δ=1, 0<γ<1 and 0<δ<1.

[0076] Optionally, combined Figure 5 As shown, the router obtains the average load (AverLoad) for each expert model, including:

[0077] S31, the router obtains the current load and historical performance evaluation score of each expert model.

[0078] S32, the router determines the average load (AverLoad) for each expert model based on historical load data and the number of loads.

[0079] This helps improve the accuracy of the expert model performance evaluation score calculation, thereby ensuring that the selection of the target expert model is sufficiently reliable.

[0080] In this step, Among them, Load j Let represent the j-th load data point of the expert model, and m represent the total number of load data points for the expert model. Each load data point of the expert model can be obtained by periodically collecting historical load data from the expert model. Understandably, the larger the total number of load data points for the expert model, the more reliable the average load value will be.

[0081] Combination Figure 6 As shown, this disclosure also provides a task processing method for a neural network model, including:

[0082] S41, the router obtains the task requirements and selects a target expert model from the hybrid expert models according to the routing decision mechanism. The task requirements include input data, and the routing decision mechanism is used to route the expert model from the hybrid expert models that best matches the task requirements.

[0083] S42, the router writes input data to the target expert model for model training and obtains the model output results.

[0084] S43, the control unit determines the target executor and task execution strategy based on the model output results.

[0085] S44, the control unit sends control commands to the target actuator to cause the target actuator to execute the task execution strategy.

[0086] The task processing method for neural network models provided in this disclosure involves obtaining task requirements and then selecting a target expert model from a hybrid expert model based on a routing decision mechanism to determine the appropriate expert model for the task requirements. Input data is then written to the target expert model for training, yielding the model output. Finally, the control unit receives the model output, determines the target executor and task execution strategy, and sends control commands to the target executor to execute the task execution strategy. This disclosure utilizes the router's local routing decision to select the appropriate expert model from the hybrid expert model, trains the input data using the target expert model to obtain the model output, and controls the target executor to execute the corresponding task execution strategy, ensuring a high degree of adaptation between the task execution strategy and the task requirements. Furthermore, this disclosure eliminates the need to upload the input data contained in the task requirements to a cloud server for processing; task execution can be achieved through the collaboration of the router and control unit. This improves task processing efficiency and response speed, thereby enhancing the adaptive capabilities of the smart home system.

[0087] Combination Figure 7 As shown, this disclosure provides a task processing system 200 for neural network models, including a smart home device 201, a router 202, a control unit 203, and an actuator 204. The smart home device 201 is configured to collect device input data. The router 202 is configured to obtain task requirements, select a target expert model from the hybrid expert models according to a routing decision mechanism, write input data to the target expert model for model training, and obtain model output results. The routing decision mechanism is used to route an expert model from the hybrid expert models that matches the task requirements; wherein, the task requirements include input data, which includes device input data and / or user input data. The control unit 203 is configured to drive the execution of the task execution strategy corresponding to the model output results based on the model output results. The actuator 204 is configured to execute the task execution strategy corresponding to the model output results.

[0088] The task processing system 200 for neural network models provided in this disclosure can select a suitable expert model from a hybrid expert model based on the task requirements through local routing decisions of the router. The target expert model is then used to train the input data to obtain the model output. The control unit drives the executor to execute the corresponding task execution strategy, ensuring a high degree of adaptation between the strategy and the task requirements. Furthermore, this disclosure eliminates the need to upload the input data for the task requirements to a cloud server for processing; task execution can be achieved through the collaboration of the router, control unit, and executor. This improves task processing efficiency and response speed, thereby enhancing the adaptability of the smart home system.

[0089] Combination Figure 8 As shown, this disclosure provides a task processing system 70 for a neural network model, including a processor 700 and a memory 701. Optionally, the device 70 may further include a communication interface 702 and a bus 703. The processor 700, communication interface 702, and memory 701 can communicate with each other via the bus 703. The communication interface 702 can be used for information transmission. The processor 700 can call logical instructions in the memory 701 to execute the task processing method for the neural network model described in the above embodiment.

[0090] Furthermore, the logic instructions in the aforementioned memory 701 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0091] The memory 701, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as program instructions / modules corresponding to the methods in the embodiments of this disclosure. The processor 700 executes functional applications and data processing by running the program instructions / modules stored in the memory 701, thereby implementing the task processing method for the neural network model described above.

[0092] The memory 701 may include a program storage area and a data storage area. The program storage area may store the operating system and application programs required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 701 may include high-speed random access memory and may also include non-volatile memory.

[0093] This disclosure provides an electronic device, including a product body and the aforementioned task processing system 70 for a neural network model. The task processing system 70 for the neural network model is mounted on the product body. The mounting relationship described herein is not limited to placement within the product body, but also includes mounting connections with other components of the electronic device, including but not limited to physical connections, electrical connections, or signal transmission connections. Those skilled in the art will understand that the task processing system 70 for the neural network model can be adapted to feasible electronic device bodies to achieve other feasible embodiments.

[0094] This disclosure provides a computer-readable storage medium storing computer-executable instructions configured to perform the above-described task processing method for a neural network model.

[0095] The technical solutions of this disclosure can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes one or more instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in this disclosure. The aforementioned storage medium can be a non-transitory storage medium, such as a USB flash drive, external hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, etc., and other media capable of storing program code.

[0096] The foregoing description and accompanying drawings fully illustrate embodiments of this disclosure to enable those skilled in the art to practice them. Other embodiments may include structural, logical, electrical, procedural, and other changes. The embodiments represent only possible variations. Individual components and functions are optional unless explicitly required, and the order of operation may vary. Parts and features of some embodiments may be included in or replace parts and features of other embodiments. Moreover, the terminology used in this application is for describing embodiments only and is not intended to limit the claims. As used in the description of embodiments and claims, the singular forms “a,” “an,” and “the” are intended to equally include the plural forms unless the context clearly indicates otherwise. Similarly, the term “and / or” as used in this application means including one or more of the associated listed items and all possible combinations thereof. Additionally, when used in this application, the term "comprise" and its variations "comprises" and / or "comprising" refer to the presence of stated features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof. Without further limitations, an element defined by the phrase "comprises a..." does not exclude the presence of other identical elements in the process, method, or apparatus that includes said element. In this document, each embodiment may focus on the differences from other embodiments, and similar or identical parts between embodiments can be referred to mutually. For methods, products, etc., disclosed in the embodiments, if they correspond to the method section disclosed in the embodiments, the relevant parts can be referred to the description of the method section.

[0097] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments of this disclosure. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0098] The methods and products (including but not limited to devices and equipment) disclosed in the embodiments herein can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units may be merely a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the shown or discussed units may be through some interfaces, and the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units may be selected to implement this embodiment according to actual needs. Furthermore, the functional units in the embodiments of this disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0099] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. In the descriptions corresponding to the flowcharts and block diagrams in the accompanying drawings, the operations or steps corresponding to different blocks may also occur in a different order than disclosed in the description, and sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. Each block in a block diagram and / or flowchart, and combinations of blocks in a block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

Claims

1. A task processing method for neural network models, characterized in that, include: Obtain the task requirements and select the target expert model from the hybrid expert models according to the routing decision mechanism; wherein, the task requirements include input data, and the routing decision mechanism is used to route the expert model from the hybrid expert models that is suitable for the task corresponding to the task requirements. Input data is written into the target expert model for model training, and the model output results are obtained. Based on the model output, drive the execution of the task execution strategy corresponding to the model output; Based on the routing decision mechanism, a target expert model is selected from the hybrid expert models, including: Semantic data is extracted from the input data, and semantic complexity is determined; semantic complexity is determined based on the semantic data and the semantic complexity evaluation function. Based on semantic data and the application domain of each expert model, determine the degree of matching between task requirements and the application domain of each model; The matching degree between the task requirements and the application domain of each model and the semantic complexity are weighted to determine the evaluation score of each expert model; Select the expert model with the highest evaluation score as the target expert model; or, Obtain the accuracy and average response time for each expert model; Based on the accuracy and average response time of each expert model, a performance evaluation score is determined for each expert model. ; Obtain the average load for each expert model ; choose The highest level of expert model is the objective expert model.

2. The method according to claim 1, characterized in that, Based on semantic data and the application domain of each expert model, determine the matching degree between task requirements and the application domain of each model, including: Based on semantic data, determine the professional field corresponding to the task requirements; Obtain the application domain of each expert model and determine the degree of matching between task requirements and each model's application domain.

3. The method according to claim 1, characterized in that, Based on the accuracy and average response time of each expert model, a performance evaluation score is determined for each expert model, including: The accuracy and the reciprocal of the average response time of each expert model are weighted to obtain a performance evaluation score for each expert model.

4. The method according to claim 3, characterized in that, Obtain the average load for each expert model, including: Obtain the current load and historical performance evaluation score for each expert model; The average load for each expert model is determined based on historical load data and the number of loads.

5. The method according to any one of claims 1 to 4, characterized in that, Based on the model output, the corresponding task execution strategy is driven to execute the model output, including: Based on the model output, determine the target executor and task execution strategy; Send control commands to the target executor to make the target executor execute the task execution strategy.

6. A task processing system for neural network models, characterized in that, include: Smart home devices are configured to collect input data from the devices; A router is configured to obtain task requirements, select a target expert model from a hybrid expert model based on a routing decision mechanism, and write input data to the target expert model for training to obtain model output results. The task requirements include input data, which includes device input data and / or user input data. The routing decision mechanism is used to route expert models from the hybrid expert model that are suitable for the task requirements. Selecting a target expert model from the hybrid expert model based on the routing decision mechanism includes: extracting semantic data from the input data and determining semantic complexity; the semantic complexity is determined based on the semantic data and a semantic complexity evaluation function; determining the matching degree between the task requirements and the application domain of each model based on the semantic data and the model application domain of each expert model; weighting the matching degree between the task requirements and the application domain of each model with the semantic complexity to determine the evaluation score of each expert model; selecting the expert model with the highest evaluation score as the target expert model; or, obtaining the accuracy and average response time of each expert model; and determining the performance evaluation score of each expert model based on its accuracy and average response time. ; Obtain the average load for each expert model ;choose The highest level of expert model is the target expert model; The control unit is configured to drive the execution of the task execution strategy corresponding to the model output results based on the model output results; The executor is configured to execute the task execution strategy corresponding to the output of the execution model.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program, when executed, performs the task processing method for a neural network model as described in any one of claims 1 to 5.

8. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the task processing method for a neural network model according to any one of claims 1 to 5 through the computer program.

Images