Task execution method, system, electronic device, storage medium and program product

Abstract

The present disclosure provides a task execution method, system, electronic device, storage medium and program product, which can be applied to the fields of artificial intelligence technology and distributed technology. The task execution method comprises: in response to a training request of a first model by a user, training the first model based on a labeled sample of the user to obtain a second model; and in response to an application result of the second model meeting a preset condition, replacing the first model with the second model, wherein the first model and the second model are both used to assist a demand side in processing a business having a target type.

Description

Technical Field

[0001] This disclosure relates to the fields of artificial intelligence and distributed technologies, and more specifically, to a task execution method, system, electronic device, storage medium, and program product. Background Technology

[0002] Large models refer to a class of artificial intelligence models with a large number of parameters built by artificial neural networks. These models have high versatility and generalization ability and can be applied to fields such as natural language processing, image recognition, and speech recognition.

[0003] Faced with a cross-regional group-based system architecture, how to provide large models for specific types of business, as well as stable and closed-loop large model optimization strategies, in order to achieve model optimization and resource sharing within the group-based system architecture, is an urgent problem to be solved. Summary of the Invention

[0004] In view of this, the present disclosure provides a task execution method, system, electronic device, storage medium, and program product.

[0005] According to one aspect of this disclosure, a task execution method is provided, comprising: in response to a user's training request for a first model, training the first model based on the user's labeled samples to obtain a second model; and in response to the application result of the second model meeting the preset conditions, replacing the first model with the second model, wherein both the first model and the second model are used to assist the requester in processing business with a target type.

[0006] According to another aspect of this disclosure, a task execution system is provided, comprising: a central cloud; and at least one edge cloud connected to the central cloud, configured to: in response to a user's training request for a first model, train the first model based on the user's labeled samples to obtain a second model; and in response to the application result of the second model meeting the preset conditions, replace the first model with the second model, wherein both the first model and the second model are used to assist the requester in processing business with a target type.

[0007] According to another aspect of this disclosure, an electronic device is provided, comprising: one or more processors; and a memory for storing one or more instructions, wherein, when executed by the one or more processors, the one or more processors cause the one or more processors to perform the method as described in this disclosure.

[0008] According to another aspect of this disclosure, a computer-readable storage medium is provided having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the methods described in this disclosure.

[0009] According to another aspect of this disclosure, a computer program product is provided, which includes computer-executable instructions that, when executed, are used to perform the methods described in this disclosure. Attached Figure Description

[0010] The above and other objects, features, and advantages of this disclosure will become clearer from the following description of embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0011] Figure 1A The system architecture of a task execution system according to an embodiment of the present disclosure is illustrated schematically;

[0012] Figure 1B The system architecture of a task execution system according to an embodiment of the present disclosure is illustrated schematically;

[0013] Figure 2 A flowchart illustrating a task execution method according to an embodiment of the present disclosure is shown schematically.

[0014] Figure 3 The illustration shows an example diagram of the process for obtaining performance evaluation results for a first model of a target type according to an embodiment of the present disclosure;

[0015] Figure 4 The illustration shows an example diagram of a process for obtaining labeled samples according to an embodiment of the present disclosure;

[0016] Figure 5 The illustration shows an example schematic diagram of a provider's addition process according to an embodiment of the present disclosure;

[0017] Figure 6A The illustration shows an example of a process of training a first model using labeled samples to obtain a second model according to an embodiment of the present disclosure.

[0018] Figure 6B This illustration schematically shows an example of a process of training a first model using labeled samples to obtain a second model according to another embodiment of the present disclosure;

[0019] Figure 7 The illustration shows an example schematic diagram of the process of adding an edge cloud according to an embodiment of the present disclosure;

[0020] Figure 8 This illustration schematically shows an example diagram of a task execution interaction process according to an embodiment of the present disclosure;

[0021] Figure 9 A block diagram of a task execution apparatus according to embodiments of the present disclosure is schematically shown; and

[0022] Figure 10 A block diagram of an electronic device suitable for implementing a task execution method according to an embodiment of the present disclosure is shown schematically. Detailed Implementation

[0023] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0024] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0025] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0026] When using expressions such as "at least one of A, B, and C", they should generally be interpreted in accordance with the meaning that is commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B, and C, etc.).

[0027] In the technical solution of this invention, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0028] In one example, taking the group-based architecture in the medical field as an example, the group-based architecture can include multiple hospitals, and each hospital can have multiple medical systems and multiple diagnostic and treatment devices.

[0029] However, because the medical systems of various hospitals are isolated from each other and each hospital has its own multiple diagnostic and treatment devices, data and clinical experience cannot be shared among the hospitals, and the equipment maintenance costs are high.

[0030] Furthermore, while large models can provide explanations for certain types of business (e.g., the medical concept of hypertension) for the group-based system architecture in the medical field, their explanations for certain specific types of business (e.g., precision medication for hypertension) are not ideal. Therefore, how to provide a stable and closed-loop large model optimization strategy to achieve model optimization and resource sharing within the group-based system architecture is an urgent problem to be solved.

[0031] To this end, this disclosure provides a task execution method, system, electronic device, storage medium, and program product, which can be applied to the fields of artificial intelligence and distributed technologies. The task execution method includes: in response to a user's training request for a first model, training the first model based on the user's labeled samples to obtain a second model; and in response to the application result of the second model meeting preset conditions, replacing the first model with the second model, wherein both the first and second models are used to assist the requester in processing business with a target type.

[0032] Figure 1A The system architecture of a task execution system according to an embodiment of this disclosure is illustrated schematically. It should be noted that... Figure 1A The examples shown are merely examples of system architectures that can be applied to the embodiments of this disclosure, in order to help those skilled in the art understand the technical content of this disclosure, but do not mean that the embodiments of this disclosure cannot be used in other devices, systems, environments or scenarios.

[0033] like Figure 1A As shown, the task execution system 100A according to this embodiment may include a central cloud 110 and at least one edge cloud connected to the central cloud 110. For example, the at least one edge cloud may include edge cloud 121, edge cloud 122, and edge cloud 123. The central cloud 110 and the at least one edge cloud can interact via a network, which serves as a medium for providing communication links between different devices.

[0034] Central cloud (110) can refer to a cloud computing platform located in a remote data center, possessing powerful computing and storage capabilities, responsible for handling large-scale data and complex computing tasks. Edge cloud can refer to cloud computing nodes located at the network edge, close to data sources and users, capable of providing low-latency, high-bandwidth data processing and storage services.

[0035] In one example, edge cloud 121, edge cloud 122, and edge cloud 123 can be configured to: in response to a user's training request for the first model, train the first model based on the user's labeled samples to obtain the second model; in response to the application result of the second model meeting preset conditions, replace the first model with the second model, wherein both the first model and the second model are used to assist the demand side in processing business with target types.

[0036] It should be noted that the aforementioned central cloud 110, edge cloud 121, edge cloud 122 and edge cloud 123 can each be network-attached storage (NAS) and share the network, thereby realizing a data-centric approach that separates storage devices from servers and centrally manages data.

[0037] According to embodiments of this disclosure, dynamic evaluation and optimization of model performance are achieved through the collaborative work of the central cloud and edge cloud, effectively meeting the high performance requirements of different business scenarios. Furthermore, by promptly replacing the first model with an optimized second model, the model's performance and adaptability are improved, contributing to enhanced quality and efficiency in the processing of business applications.

[0038] In embodiments of this disclosure, the task execution system may further include at least one demander connected to the edge cloud and at least one provider connected to the demander, as will be described below. Figure 1B A task execution system provided in another embodiment of this disclosure will be described.

[0039] Figure 1B The system architecture of a task execution system according to an embodiment of the present disclosure is illustrated schematically.

[0040] like Figure 1B As shown, the task execution system 100B according to this embodiment may include a central cloud 110, at least one edge cloud connected to the central cloud 110, at least one demander connected to the edge cloud, and at least one provider connected to the demander.

[0041] For example, at least one edge cloud may include edge cloud 121, edge cloud 122, and edge cloud 123. At least one demander may include demander 131 and demander 132 connected to edge cloud 122. At least one provider may include provider 141 connected to demander 131, and provider 142 and provider 143 connected to demander 132. The central cloud 110 can interact with at least one edge cloud, the edge cloud with at least one demander, and the demander with at least one provider via a network, which serves as a medium for providing communication links between different devices.

[0042] For example, in the medical field, the central cloud 110 can be a medical data center, the edge cloud can be the edge cloud servers inside various hospitals, the demand side can be the diagnosis and treatment platform, and the provider can be the diagnosis and treatment equipment.

[0043] In one example, providers 141, 142, and 143 can all be configured to provide collected data for candidate types. For example, in the medical field, candidate types could be various disease types, such as pneumonia, diabetes, etc. Collected data refers to the raw business data collected by the provider through its equipment. Examples include patient X-ray images and electrocardiogram data collected by medical equipment.

[0044] In another example, both demanders 131 and 132 can be configured to, in response to receiving model information for the second model, deploy the second model to replace the first model. For example, again in the medical field, a second model for disease diagnosis can be deployed to demanders 131 and 132, replacing the original first model to improve the accuracy and efficiency of diagnosis.

[0045] It should be noted that in the task execution system 100B, the central cloud 110 and edge clouds 121, 122 and 123 can be used as the cloud side, the demand side 131 and 132 as the edge side, and the provider 141, 142 and 143 as the end side, thus forming a cloud-edge-end architecture, and each part can work together to realize tasks such as data processing, model training and application, and business processing.

[0046] For example, in the medical field, central cloud 110 can be used for central management; edge clouds 121, 122, and 123 can be used for business-side model training; demand side 131 and demand side 132 can be used to assist doctors in actual diagnosis and treatment of patients; and provider 141, provider 142, and provider 143 can be used to achieve access to intelligent agents and devices to obtain basic patient user data information.

[0047] Thus, the 100B task execution system can achieve closed-loop management of business processes by acquiring data from the underlying layer, using the acquired data for cloud training, and then feeding the trained model back to the diagnosis and treatment platform to empower doctors.

[0048] According to embodiments of this disclosure, by constructing a task execution system comprising a central cloud, edge cloud, demand side, and provider, and adopting a cloud-edge-device architecture, a cross-regional, group-level management architecture deployment system is achieved. Simultaneously, a stable and closed-loop complete process from data acquisition and model training to model deployment is realized. This not only efficiently processes and analyzes large amounts of data, continuously optimizes models, and improves model performance and accuracy, but also quickly responds to changes in business needs, promptly deploying optimized models to demand sides, thereby improving the efficiency and quality of business processing. Furthermore, various components of the system can be expanded by adding edge cloud, demand side, and provider, enhancing the system's flexibility, scalability, and reliability.

[0049] It should be noted that the task execution method provided in this embodiment can generally be executed by edge cloud 121, edge cloud 122, and edge cloud 123. Accordingly, the task execution device provided in this embodiment can generally be disposed in edge cloud 121, edge cloud 122, and edge cloud 123.

[0050] Alternatively, the task execution method provided in this embodiment of the disclosure can also be executed by the central cloud 110. Correspondingly, the task execution device provided in this embodiment of the disclosure can also be disposed in the central cloud 110.

[0051] It should be understood that Figure 1A and 1B The number of central clouds, edge clouds, demanders, and providers shown is merely illustrative. Depending on implementation needs, there can be any number of central clouds, edge clouds, demanders, and providers.

[0052] It should be noted that the sequence numbers of the operations in the following methods are for descriptive purposes only and should not be considered as indicating the execution order of the operations. Unless explicitly stated otherwise, the method does not need to be executed in the exact order shown.

[0053] The foregoing has described the system architecture for applying task execution methods provided in this disclosure. The following will use... Figure 2 As an example, the task execution process of this disclosure will be further explained.

[0054] Figure 2 A flowchart illustrating a task execution method according to an embodiment of the present disclosure is shown schematically.

[0055] like Figure 2 As shown, the task execution method 200 includes operations S210 to S220.

[0056] In operation S210, in response to the user's training request for the first model, the first model is trained based on the user's labeled samples to obtain the second model.

[0057] In operation S220, in response to the application result of the second model meeting the preset conditions, the first model is replaced with the second model. Both the first and second models are used to assist the demand side in processing business with target types.

[0058] The first model can be a model used to assist the requester in processing business with a target type. The target type refers to a specific classification or category of business, used to distinguish different types of business requirements. In one example, a first preset duration for performance evaluation of the first model can be set. In this case, the performance evaluation operation of the first model can be triggered every first preset duration. In another example, a first preset evaluation condition can be set to determine whether performance evaluation of the first model is necessary. In this case, the performance evaluation operation of the first model can be triggered in response to the satisfaction of the first preset evaluation condition. In yet another example, the performance evaluation operation of the first model can also be manually triggered by the user. For example, the user can trigger the performance evaluation operation based on the performance of the first model in assisting the processing of business with a target type.

[0059] It should be noted that the aforementioned first preset duration and first preset evaluation conditions can be set according to the target type, or according to the demand side applying the first model for the target type, or according to the data update frequency of the target type's business, or according to the urgency of the target type's business, or can be set by combining at least two of the target type, demand side, data update frequency, and urgency. No limitation is made here.

[0060] When a performance evaluation operation on the first model is triggered, the results of the first model processing the target type of business can be obtained, and the results can be used to evaluate the performance, resulting in a performance evaluation result. The performance evaluation result refers to the result obtained by quantitatively evaluating the performance of the first model on a specific task. For example, the performance of the first model in a business with a target type can be measured using at least one of the following metrics: accuracy, recall, and F1 score.

[0061] Preset conditions can be pre-defined model performance standards used to determine whether the model meets deployment requirements. Preset conditions can be configured according to actual business needs and are not limited here. For example, a preset condition could be that the model must achieve an accuracy rate of over 90% when processing target type business.

[0062] For example, if the performance evaluation result of the first model used for the target type meets the preset conditions, it is not necessary to retrain the first model. Instead, at least one demand party that has deployed the first model can continue to use the first model to process the business of the target type until the next performance evaluation operation for the first model is triggered. Alternatively, if the performance evaluation result of the first model used for the target type does not meet the preset conditions, labeled samples for the target type can be obtained.

[0063] Labeled samples refer to the dataset used for optimizing and training the first model. The method of obtaining labeled samples can be configured according to actual business needs and is not limited here. For example, labeled samples can be obtained from at least one client that has deployed the first model. Alternatively, labeled samples can also be pre-labeled datasets with target types and corresponding labels. Labeling methods can include at least one of intelligent labeling, manual labeling, and collaborative labeling, and are not limited here.

[0064] After obtaining labeled samples for the target type, the currently deployed first model can be further trained and optimized using these labeled samples to obtain a second model with better performance. For example, in the medical field, the target type can be a disease type, and the labeled samples can be medical image data and diagnostic results for a specific disease type. By using these labeled samples to further optimize the first model, a second model with better performance for medical image diagnosis of a specific disease type can be obtained.

[0065] The model information for the second model can include its structure, parameters, weights, etc., describing the model's state and function. For example, the model information for the second model might include the number of layers in the neural network model, the number of neurons in each layer, and the connection weights. After obtaining the second model, the relevant information of the trained second model can be transmitted to the requesting party that has already deployed and is using the first model, so that they can update and replace the model. For example, the model information of the newly trained second model can be sent to the requesting party using the old version of the first model via network transmission. The requesting party refers to an entity that uses the first and second models to process business with a target type. For example, in the medical field, the requesting party could be a diagnosis and treatment platform.

[0066] According to embodiments of this disclosure, by dynamically responding to training requests for the first model, using labeled samples to train the first model in a targeted manner, and replacing the first model with the second model when the application results meet preset conditions, the performance and adaptability of the model can be effectively improved, ensuring that the model runs stably and reliably in actual business, thereby helping to improve the quality and efficiency of business processing.

[0067] In one example, taking the medical field, the demand side can include a diagnostic platform, which is a platform that assists doctors in diagnosing target-type business using a first or second model, in order to provide remote diagnostic services to patients. The provider connected to the demand side can include diagnostic equipment, such as equipment capable of collecting patient data like X-rays and CT scans, which will be used for disease diagnosis.

[0068] The target type can include disease types, such as pneumonia, diabetes, heart disease, etc. The first model or the second model will provide auxiliary diagnosis for these specific disease types. Both the first model and the second model can be used to assist the diagnosis and treatment platform in diagnosing diseases based on their specific types.

[0069] According to embodiments of this disclosure, by using the diagnostic platform as the demand side, the diagnostic equipment as the provider, and the disease type as the target type, and by using the first model and the second model to assist the diagnostic platform in disease diagnosis, it is possible to achieve accurate diagnostic assistance for specific disease types, thereby reducing the pressure on doctors, improving their work efficiency, and ultimately enabling patients to benefit from more accurate and efficient medical diagnostic services.

[0070] In the embodiments of this disclosure, the training request for the first model may be triggered in response to a situation where the performance evaluation result of the first model does not meet a preset condition. The following will utilize... Figure 3 The process of obtaining the performance evaluation results of the first model used for the target type is further explained.

[0071] Figure 3 The illustration shows an example schematic diagram of the process for obtaining performance evaluation results for a first model of a target type according to an embodiment of the present disclosure.

[0072] like Figure 3 As shown in 300, the process of obtaining the performance evaluation results of the first model for the target type may include operations S301 to S306.

[0073] When operating S301, the provider can obtain the business data to be processed.

[0074] In operation S302, the provider sends the business data to be processed to the intelligent agent.

[0075] When operating S303, the agent can call the first model deployed by the requester after receiving the business data to be processed.

[0076] When operating S304, the agent can utilize the first model to process the business data to be processed and obtain the business processing result. The business processing result refers to the result obtained by the agent after processing the business data to be processed from the provider by calling the first model deployed by the demand side. For example, in a medical scenario, the diagnosis and treatment platform calls the first model to analyze patient image data from medical equipment and obtain disease diagnosis results.

[0077] When operating S305, the agent can send the business processing results to the edge cloud.

[0078] In operation S306, after receiving the business processing result, the edge cloud can determine the performance evaluation result for the first model based on the business processing result.

[0079] In one example, in addition to the aforementioned metrics such as accuracy, recall, and F1 score, the performance of the first model can be evaluated from multiple dimensions, including processing speed, resource consumption, and stability. For instance, the response time of the first model under high concurrency requests and its operating efficiency under different hardware environments can be evaluated.

[0080] According to embodiments of this disclosure, the performance evaluation mechanism based on actual business processing results can accurately and realistically reflect the performance of the first model in the target type of business. This not only improves the accuracy and reliability of performance evaluation, but also helps to promptly identify problems and shortcomings of the model in actual use, thereby triggering subsequent model optimization operations and improving the intelligence level and operational efficiency of the task execution process.

[0081] The following will utilize Figure 4 The process of obtaining the labeled samples provided in this disclosure will be further explained.

[0082] Figure 4 The illustration shows an example schematic diagram of the process of obtaining labeled samples according to an embodiment of the present disclosure.

[0083] like Figure 4 As shown in Figure 400, the process of obtaining labeled samples may include operations S401 to S406.

[0084] It should be noted beforehand that a demand side can be associated with at least one provider. A provider is an entity used to process at least one business with candidate types. For example, in a medical scenario, a provider could be various types of diagnostic and treatment equipment capable of collecting patient image data, blood test data, etc.

[0085] Labeled samples can be determined based on data from a provider. For example, in operation S401, the provider can collect data for candidate types.

[0086] In S402 operation, the provider can send the collected data to the agent deployed by the provider. An agent is a tool or software deployed at the provider's location, compatible with multiple providers, used to perform specific data processing and transmission tasks. For example, data acquisition and conversion software installed on diagnostic equipment can preprocess the raw data collected by the equipment.

[0087] It should be noted that intelligent agents can operate independently or form a dual-machine cluster system (Highly Available, HA) to improve availability and ensure business continuity.

[0088] When operating S403, the agent can convert collected data into data in a preset format. Collected data refers to the raw business data gathered by the provider. Examples include X-ray images and electrocardiogram data collected by diagnostic equipment. The preset format can be determined based on the provider's device identifier and candidate types to ensure data consistency and compatibility during transmission and processing.

[0089] When operating S404, the agent can send data in a preset format to the edge cloud.

[0090] According to embodiments of this disclosure, through the association between the demand side and the provider, and the format conversion and transmission of the collected data by the intelligent agent, efficient integration and transmission of multi-source heterogeneous data can be achieved, improving the efficiency and accuracy of data processing and ensuring the consistency and availability of data in the edge cloud.

[0091] When operating S405, the edge cloud can store candidate types and preset formats of data into a dataset. This dataset can then be used for subsequent processing and analysis. Furthermore, the edge cloud can synchronize its respective datasets to the central cloud, enabling the central cloud to manage and monitor data from various providers, as well as to share resources and models.

[0092] When operating S406, if it is necessary to optimize the training of the first model, the edge cloud can identify the data corresponding to the candidate type that matches the target type in the dataset as labeled samples.

[0093] For example, the dataset can store data corresponding to at least one candidate type. When it is necessary to optimize the training of the first model, at least one candidate type can be matched according to the target type corresponding to the first model, so as to determine the data corresponding to the candidate type that matches the target type and obtain labeled samples.

[0094] During this process, Edge Cloud can also filter and clean the data corresponding to candidate types that match the target type to remove noisy, erroneous, and duplicate data, thereby improving the quality of labeled samples. Furthermore, Edge Cloud can perform data augmentation on the data corresponding to candidate types that match the target type to increase the diversity and quantity of labeled samples.

[0095] According to embodiments of this disclosure, by automatically receiving, storing, and filtering data in a preset format, labeled samples that match the target type can be constructed. This not only achieves automated data management but also ensures the accuracy and relevance of the labeled samples, providing a high-quality data foundation for subsequent model training.

[0096] In the embodiments of this disclosure, the above-described task execution method can be applied to a target edge cloud. The target edge cloud can access a new provider by responding to a first request from a provider to be added, thereby enhancing the scalability of the task execution process for the provider.

[0097] Figure 5 The illustration shows an example schematic diagram of a provider's addition process according to an embodiment of the present disclosure.

[0098] like Figure 5 As shown in Figure 500, the provider's new process may include operations S501 to S505.

[0099] When operating S501, new providers can install smart agents and interact with the edge cloud through these agents.

[0100] In operation S502, a new provider can send its first request to the edge cloud. When sending the first request, the new provider can also specify the training samples to be used to train the model.

[0101] When operating S503, the edge cloud can use the training samples indicated by the first request to train the initial model and obtain the third model.

[0102] When operating S504, the edge cloud can send the model information of the third model to the demand side associated with the provider to be added, so that the demand side can deploy the third model based on the model information.

[0103] When operating S505, a new provider can use an agent to call a third model from the demand side.

[0104] According to embodiments of this disclosure, by responding to a first request from a new provider, training and deploying an initial model using training samples indicated by the new provider, dynamic expansion and personalization of the model are achieved. This not only enables rapid adaptation to the data and business needs of new providers and improves the applicability and accuracy of the resulting third model, but also enhances the flexibility and scalability of the task execution system.

[0105] For new providers that have already connected to the edge cloud and deployed the third model, the third model can be optimized and trained by performing the following operations: S506 to S508.

[0106] When operating S506, the edge cloud can perform performance evaluation on the third model. If the performance evaluation result does not meet the preset conditions, the edge cloud can retrain the third model to obtain the fourth model.

[0107] In one example, a second preset duration can be set for performance evaluation of the third model. In this case, the performance evaluation operation of the third model can be triggered every second preset duration. In another example, a second preset evaluation condition can be set to determine whether performance evaluation of the third model is necessary. In this case, the performance evaluation operation of the third model can be triggered in response to the satisfaction of the second preset evaluation condition. In yet another example, the performance evaluation operation of the third model can also be manually triggered by the user. For example, the user can trigger the performance evaluation operation based on the performance of the third model in assisting in processing business with candidate types.

[0108] It should be noted that the aforementioned second preset duration and second preset evaluation conditions can be set according to the business type used by the third model, or according to the demand side applying the third model to the business type, or according to the data update frequency of the business type, or according to the urgency of the business type, or a combination of at least two of the business type, demand side, data update frequency and urgency. No limitation is made here.

[0109] In operation S507, the edge cloud can send the model information of the fourth model to the demand side associated with the provider to be added, so that the demand side can replace the third major model with the fourth model based on the model information. In operation S508, the provider to be added can use an agent to call the fourth model from the demand side.

[0110] The above task execution method can be applied to a target edge cloud, which can interact with the central cloud via a virtual local area network (VLAN). The central cloud can also interact with at least one other edge cloud via a VLAN.

[0111] It should be noted that the process of training the first model using labeled samples to obtain the second model can be performed by the target edge cloud, the central cloud, or other edge clouds connected to the central cloud; no limitation is made here.

[0112] In one example, if no server is deployed locally on the target edge cloud, or if a server is deployed locally but is in use, labeled samples can be sent to the central cloud so that the central cloud can use the labeled samples to train the first model and obtain the second model.

[0113] A target edge cloud without locally deployed servers refers to a local environment in which no physical or virtual servers are deployed. For example, a small telemedicine site, due to its small size, does not have its own servers, and all data processing and storage rely on the edge cloud and the central cloud.

[0114] "Locally deployed servers are in a occupied state" means that there are servers in the local environment where the target edge cloud is located, but these servers are currently occupied by other tasks or services and cannot handle new training tasks. For example, a hospital's local servers are processing a large number of patient registration and preliminary diagnosis tasks and cannot respond to new model training requests in a timely manner.

[0115] According to embodiments of this disclosure, by rationally allocating training tasks to the central cloud when the target edge cloud resources are insufficient, it is possible not only to fully utilize the computing power of the central cloud and improve the speed and quality of model training, but also to ensure that the target edge cloud can still obtain a high-performance model even when resources are limited, thereby achieving optimized utilization of computing resources and efficient model training.

[0116] The following will utilize Figure 6A The process of training the first model using labeled samples, performed by the central cloud, to obtain the second model, and the utilization of... Figure 6B The process of training the first model using labeled samples to obtain the second model for other edge clouds connected to the central cloud is explained separately.

[0117] Figure 6A The illustration shows an example of a process of training a first model using labeled samples to obtain a second model according to an embodiment of the present disclosure.

[0118] like Figure 6A As shown in 600A, if no server is deployed locally in the target edge cloud 610, or if a server is deployed locally but the server is occupied, the target edge cloud 610 can send the labeled samples to the central cloud 620.

[0119] After receiving labeled samples, the central cloud 620 can obtain resource usage information. This information characterizes the usage of servers deployed in the central cloud 620 and the usage of servers deployed in at least one candidate edge cloud. For example, the resource usage information includes the server usage of the central cloud 620, the server usage of candidate edge cloud 631, candidate edge cloud 632, and candidate edge cloud 633.

[0120] Server usage metrics can include CPU utilization, memory usage, and network bandwidth, used to determine whether a server is idle. For example, if a server's CPU utilization is below 20% and its memory usage is below 30%, the server can be considered idle.

[0121] After obtaining resource occupancy information, the central cloud 620 can determine, based on the resource occupancy information, the target cloud with an idle server among the central cloud 620 and at least one candidate edge cloud (candidate edge cloud 631, candidate edge cloud 632 and candidate edge cloud 633).

[0122] If there are idle servers in the central cloud 620, the central cloud 620 can be prioritized as the target cloud. The first model can be trained on the central cloud 620 using labeled samples for the target type to obtain the second model.

[0123] Figure 6B The illustration shows an example of a process of training a first model using labeled samples to obtain a second model according to another embodiment of the present disclosure.

[0124] like Figure 6B As shown in 600B, if there are no idle servers in the central cloud 620, the central cloud 620 can determine the target cloud with idle servers in at least one candidate edge cloud (candidate edge cloud 631, candidate edge cloud 632 and candidate edge cloud 633) based on resource occupancy information.

[0125] For example, if the central cloud 620 determines that there are idle servers in the candidate edge cloud 632 based on resource occupancy information, the candidate edge cloud 632 can be identified as the target cloud, and labeled samples for the target type can be sent to the candidate edge cloud 632 so that the candidate edge cloud 632 can use the labeled samples for the target type to train the first model and obtain the second model.

[0126] According to embodiments of this disclosure, by dynamically selecting idle servers for model training based on resource occupancy information, not only can the model training time be effectively shortened, but also the timely updating and optimization of the model can be ensured, thereby making full use of the computing resources of the central cloud and edge cloud, and improving resource utilization and training efficiency.

[0127] In embodiments of this disclosure, the central cloud can access a new edge cloud by responding to a second request from the edge cloud to be added, thereby enhancing the scalability of the task execution process for the edge cloud.

[0128] Figure 7 The illustration shows an example schematic diagram of the process of adding an edge cloud according to an embodiment of the present disclosure.

[0129] like Figure 7 As shown in Figure 700, the edge cloud 740 to be added can be pre-associated with at least one demander, each demander can be pre-connected with at least one provider, and each provider can deploy an agent.

[0130] For example, the newly added edge cloud 740 can connect to the demand side 731 via a virtual LAN, the demand side 731 can connect to the provider 721 via a virtual LAN, and the provider 721 can deploy an intelligent agent 711.

[0131] Alternatively, the newly added edge cloud 740 can also connect to the demand side 732 via a virtual LAN. The demand side 732 can connect to the providers 722 and 723 via a virtual LAN. The provider 722 can deploy an agent 712, and the provider 723 can deploy an agent 713.

[0132] When it is necessary to connect the newly added edge cloud 740 to the central cloud 750, the newly added edge cloud 740 can send a second request to the central cloud 750. The second request can refer to a connection request received by the central cloud 750 from the newly added edge cloud 740. For example, if a new hospital deploys its own edge cloud server and wishes to establish a connection with the central cloud 750 to share resources and data, it can send a second request to the central cloud 750.

[0133] When the central cloud 750 receives a second request from the edge cloud to be added, it can connect to the edge cloud 740 via the local area network. For example, by configuring network parameters and setting routing rules, the edge cloud 740 to be added can communicate with the central cloud 750.

[0134] According to embodiments of this disclosure, the dynamic expansion mechanism between the central cloud and the edge cloud to be added not only enables rapid response to changes in business needs and improves the adaptability and scalability of the task execution system, but also fully leverages the advantages of the central cloud and the edge cloud to optimize the utilization of computing and storage resources, thereby achieving efficient integration and sharing of resources and data.

[0135] The following will combine Figure 8 The following is an exemplary description of the task execution interaction process provided in the embodiments of this disclosure.

[0136] Figure 8 The illustration shows an example diagram of a task execution interaction process according to an embodiment of the present disclosure.

[0137] like Figure 8 As shown, in 800, the task execution process may include operations S801 to S812.

[0138] When operating S801, the provider can send data in a preset format belonging to the candidate type to the edge cloud.

[0139] In operation S802, after receiving data of a preset format belonging to a candidate type, the edge cloud can store the candidate type and the preset format data into a data set.

[0140] When operating S803, the requester can use the first model to process the business data to be processed and obtain the business processing results.

[0141] When operating S804, the requester can send the business processing results to the edge cloud.

[0142] When operating S805, the edge cloud can determine the performance evaluation results based on the business processing results.

[0143] When operating S806, in response to the performance evaluation results not meeting the preset conditions, the edge cloud can identify the data corresponding to the candidate type that matches the target type in the dataset as labeled samples.

[0144] In operating S807, one of the edge cloud, the central cloud, and other edge clouds connected to the central cloud can be used to train the first model using labeled samples for the target type to obtain the second model.

[0145] For example, if a server is deployed locally in the edge cloud and is in an idle state, operation S8071 can be executed. In operation S8071, the edge cloud can use labeled samples for the target type to train the first model and obtain the second model.

[0146] Alternatively, if no server is deployed locally on the edge cloud, or if a server is deployed locally but is in use, operations S8072 to S8073 can be executed. In operation S8072, the edge cloud can send labeled samples to the central cloud. In operation S8073, the central cloud or other edge clouds connected to the central cloud can use the labeled samples to train the first model and obtain the second model.

[0147] When operating S808, the edge cloud can determine the application results of the second model.

[0148] When operating S809, in response to the application results meeting preset conditions, the edge cloud can send the model information of the second model to the requester.

[0149] When operating S810, after receiving the model information of the second model, the requester can use the second model to replace the first model.

[0150] When operating S811, in response to the application results meeting preset conditions, the edge cloud can also send the model information of the second model to the central cloud.

[0151] When operating S812, after receiving the model information of the second model, the central cloud can also associate and store the model information of the second model with the target type of the second model in a model collection. A model collection refers to a repository or data structure that includes information on multiple models, used for centralized management and retrieval of various models.

[0152] In one example, each edge cloud connected to the central cloud can match the various types in the model set according to the type of the model to be deployed, so as to determine the model to be deployed in the model set, obtain the corresponding model information, and deploy the required model based on the model information.

[0153] According to embodiments of this disclosure, by sending the model information of the second model to the central cloud and storing it in the model collection in association with the target type, it is not only convenient to uniformly manage and maintain models of different target types, improve the reusability and sharing of models, but also provide the central cloud with more comprehensive model resources, enhance its support capabilities for various businesses, and realize centralized management and efficient retrieval of models.

[0154] The above are merely exemplary embodiments, but are not limited thereto. Other task execution methods known in the art may also be included, as long as they can improve the quality and efficiency of business processing.

[0155] Based on the above task execution method, the present invention also provides a task execution device. The following will be combined with... Figure 9 The device is described in detail.

[0156] Figure 9 A block diagram of a task execution apparatus according to an embodiment of the present disclosure is shown schematically.

[0157] like Figure 9 As shown, the task execution device 900 may include a training module 910 and a sending module 920.

[0158] The training module 910 is used to respond to the user's training request for the first model, and train the first model based on the user's labeled samples to obtain the second model.

[0159] The sending module 920 is used to replace the first model with the second model in response to the application result of the second model meeting the preset conditions. Both the first model and the second model are used to assist the requester in processing business with target type.

[0160] Any one or more of the modules, submodules, units, and subunits according to embodiments of this disclosure, or at least part of the functions of any one or more of them, can be implemented in one module. Any one or more of the modules, submodules, units, and subunits according to embodiments of this disclosure can be implemented by dividing them into multiple modules. Any one or more of the modules, submodules, units, and subunits according to embodiments of this disclosure can be at least partially implemented as hardware circuitry, such as Field Programmable Gate Arrays (FPGAs), Programmable Logic Arrays (PLAs), Systems-on-Chip, Systems-on-Substrate, Systems-on-Package, Application-Specific Integrated Circuits (ASICs), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, one or more of the modules, submodules, units, and subunits according to embodiments of this disclosure can be at least partially implemented as computer program modules, which, when run, can perform corresponding functions.

[0161] It should be noted that the task execution device part in the embodiments of this disclosure corresponds to the task execution method part in the embodiments of this disclosure. The specific description of the task execution device part is referred to in the task execution method part, and will not be repeated here.

[0162] Figure 10 A block diagram of an electronic device suitable for implementing a task execution method according to an embodiment of the present disclosure is shown schematically. Figure 10 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0163] like Figure 10As shown, a computer electronic device 1000 according to an embodiment of the present disclosure includes a processor 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage portion 1009 into a random access memory (RAM) 1003. The processor 1001 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1001 may also include onboard memory for caching purposes. The processor 1001 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0164] RAM 1003 stores various programs and data required for the operation of electronic device 1000. Processor 1001, ROM 1002 and RAM 1003 are interconnected via bus 1004.

[0165] According to embodiments of this disclosure, the electronic device 1000 may further include an input / output (I / O) interface 1005, which is also connected to a bus 1004. The electronic device 1000 may also include one or more of the following components connected to the input / output (I / O) interface 1005: an input section 1006 including a keyboard, mouse, etc.; an output section 1007 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN card, modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the input / output (I / O) interface 1005 as needed. A removable medium 1011, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1010 as needed so that computer programs read from it can be installed into the storage section 1008 as needed.

[0166] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the task execution method according to the embodiments of this disclosure.

[0167] In this disclosure, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0168] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods provided in the embodiments of this disclosure. When the computer program product is run on an electronic device, the program code is used to enable the electronic device to implement the task execution methods provided in the embodiments of this disclosure.

[0169] When the computer program is executed by the processor 1001, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0170] According to embodiments of this disclosure, program code for executing computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages.

[0171] 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 various embodiments of this disclosure. It should also be noted that in some alternative implementations, the functions indicated in the boxes may occur in a different order than those shown in the drawings.

[0172] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of this disclosure is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A task execution method, comprising: In response to a user's training request for the first model, the first model is trained based on the user's labeled samples to obtain the second model; as well as In response to the application result of the second model meeting the preset conditions, the first model is replaced with the second model, wherein both the first model and the second model are used to assist the demand side in processing business with target type.

2. The method according to claim 1, wherein, Replacing the first model with the second model includes: The model information of the second model is sent to at least one demander that has deployed the first model, so that the at least one demander can replace the first model with the second model based on the model information.

3. The method according to claim 2, wherein, The demand side is associated with at least one provider, which processes at least one business with candidate types, and the labeled samples are determined based on data from the provider.

4. The method according to claim 3, wherein, The provider deploys an intelligent agent; the intelligent agent is used to convert the collected data for the candidate type collected by the provider into data in a preset format and send it to the edge cloud, the preset format being determined based on the provider's device identifier and the candidate type.

5. The method of claim 3, further comprising, before training the first model based on the user's labeled samples in response to a user's training request for the first model to obtain the second model: In response to receiving data in the preset format from the provider, the data in the preset format is stored in a data set; The labeled samples were obtained through the following methods: The data corresponding to the candidate types that match the target type in the dataset are identified as the labeled samples.

6. The method according to claim 1, further comprising: In response to receiving a first request from a provider to be added, the initial model is trained using the training samples indicated in the first request to obtain a third model, wherein the provider to be added has a pre-installed agent and interacts with the edge cloud through the agent; and The model information of the third model is sent to the demand side associated with the provider to be added, so that the demand side can deploy the third model based on the model information, and the provider to be added can use the agent to call the third model from the demand side.

7. The method according to claim 1, further comprising: In response to receiving a service processing result from the at least one agent for the target type of service, a performance evaluation result for the first model is determined based on the service processing result, wherein the service processing result is obtained by the agent processing the pending service data from the at least one provider by invoking the first model deployed by the demand side.

8. The method according to claim 1, wherein, The method is applied to a target edge cloud, which interacts with the central cloud via a virtual local area network; The method further includes: If no server is deployed locally on the target edge cloud, or if a server is deployed locally but is in use, the labeled samples are sent to the central cloud so that the central cloud can use the labeled samples to train the first model and obtain the second model.

9. The method according to claim 8, wherein, The second model is obtained from the central cloud in the following way: Based on resource occupancy information, a target cloud with idle servers is identified among the central cloud and at least one candidate edge cloud. The resource occupancy information characterizes the usage of servers deployed in the central cloud and the usage of servers deployed in the candidate edge clouds. The labeled samples are sent to the target cloud so that the target cloud can use the labeled samples for the target type to train the first model and obtain the second model.

10. The method according to claim 8, wherein, The central cloud is also used to connect the edge cloud to be added to the central cloud via a local area network when a second request is received from the edge cloud to be added; wherein the edge cloud to be added has been pre-associated with at least one provider and at least one demander.

11. The method of claim 8, further comprising, after the application result of the second model meets the preset condition: The model information of the second model is sent to the central cloud so that the central cloud can associate the model information and the target type of the second model and store them in the model set.

12. The method according to any one of claims 1 to 11, wherein, The demand side includes a diagnostic and treatment platform, and the provider connected to the demand side includes diagnostic and treatment equipment. The target type includes a disease type. Both the first model and the second model are used to assist the diagnostic and treatment platform in diagnosing diseases for the disease type.

13. A task execution system, comprising: Central Cloud; At least one edge cloud connected to the central cloud is configured as follows: In response to a user's training request for the first model, the first model is trained based on the user's labeled samples to obtain the second model; In response to the application result of the second model meeting the preset conditions, the first model is replaced with the second model, wherein both the first model and the second model are used to assist the demand side in processing business with target type.

14. The system of claim 13, further comprising: At least one demander connected to the edge cloud, and at least one provider connected to the demander; The provider is configured as follows: Provide collected data for candidate types; The demand side is configured as follows: In response to receiving model information of the second model, the second model is deployed according to the model information to replace the first model.

15. An electronic device comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 12.

16. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 12.

17. A computer program product comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 12.

Images