Incremental learning method and device based on cloud edge collaboration architecture, equipment and medium

Abstract

The application provides an incremental learning method and device based on a cloud-edge collaborative architecture, equipment and a medium. The method comprises the following steps: training a cloud-end basic model based on original data collected from each edge node; training an edge model corresponding to each edge node based on self-owned data of each edge node and the cloud-end basic model; screening incremental data generated by each edge node to obtain an incremental data set, wherein the incremental data is generated by the edge node based on the edge model; and performing incremental learning training on the cloud-end basic model through the incremental data set to obtain an updated cloud-end basic model. Through the cloud-edge collaborative architecture, the application solves the problem that three important indicators, i.e., model generalization ability, model accuracy and reasoning time delay, are difficult to meet business requirements at the same time, and ensures that the cloud-end basic model can maintain good generalization ability in multiple scenarios through a cloud-end training strategy.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an incremental learning method, apparatus, device, and medium based on a cloud-edge collaborative architecture. Background Technology

[0002] Artificial intelligence has permeated all areas of industrial production. Currently, computer vision technology still faces numerous technical bottlenecks in industrial quality inspection. The high-speed operation of various machines in industrial production processes poses significant challenges to the deployment of AI algorithms and inference latency. Currently, solutions for balancing model generalization ability and accuracy can be summarized into four categories: 1. Cloud training and cloud inference; 2. Cloud training and edge inference; 3. Edge training and edge inference; 4. Cloud-edge collaborative federated learning.

[0003] Each of the four methods described above has its limitations. Firstly, cloud-based training and inference architectures suffer from significant network latency in inference services, typically failing to meet most industrial-grade requirements. Secondly, cloud-based training and edge inference architectures cannot iterate and update the algorithm model in real-time based on business data. Since the statistical characteristics of the target variable usually change over time, the model may drift and degrade after a period of use. Thirdly, edge-based training and inference architectures struggle to fully utilize common data across different scenarios. Training models using edge computing power is inefficient, and the models are prone to overfitting, exhibiting poor generalization performance and robustness, making it difficult to support training models with massive amounts of data. Fourthly, while cloud-edge collaborative federated learning architectures can fully utilize data resources from both the cloud and edge, they are currently often based on simple model integration using expert systems, resulting in relatively poor interpretability of prediction results and difficulty in guaranteeing generalization ability in new scenarios. Summary of the Invention

[0004] This invention provides an incremental learning method, apparatus, device, and medium based on a cloud-edge collaborative architecture to solve the aforementioned technical problems in existing solutions for balancing model generalization ability and accuracy.

[0005] This invention provides an incremental learning method based on a cloud-edge collaborative architecture, comprising:

[0006] The cloud-based basic model is trained based on the raw data collected from each edge node;

[0007] Based on the proprietary data of each edge node and the cloud-based basic model, the edge model corresponding to each edge node is trained.

[0008] The incremental data generated by each edge node is filtered to obtain an incremental dataset, wherein the incremental data is generated by the edge node based on the edge model;

[0009] The cloud-based basic model is incrementally trained using the incremental dataset to obtain an updated cloud-based basic model.

[0010] According to the incremental learning method based on a cloud-edge collaborative architecture provided by the present invention, the cloud-based basic model trained based on raw data collected from each edge node includes:

[0011] The weighted focusing loss function is used as the loss function for neural network optimization. The model is trained based on gradient descent and raw data collected from each edge node to obtain the cloud-based basic model.

[0012] According to the incremental learning method based on a cloud-edge collaborative architecture provided by the present invention, the step of training an edge model corresponding to each edge node based on the proprietary data of each edge node and the cloud-based basic model includes:

[0013] The cloud-based basic model is then distributed to each of the edge nodes.

[0014] Freeze the backbone network of the cloud-based basic model and use the proprietary data of each edge node as the model training input;

[0015] The cloud-based basic model is subjected to small-sample transfer learning based on the transfer learning method and the focus loss function to obtain the edge model corresponding to each edge node.

[0016] According to the incremental learning method based on a cloud-edge collaborative architecture provided by the present invention, after training the edge model corresponding to each edge node based on the proprietary data of each edge node and the cloud-based basic model, the method includes:

[0017] The data generated by the edge nodes within the first time period is used as the model input, and the edge model is updated through an incremental learning method.

[0018] The incremental data generated by the edge node within the second time period is uploaded to the cloud server, where the second time period is a positive integer multiple of the first time period.

[0019] According to the incremental learning method based on a cloud-edge collaborative architecture provided by the present invention, the step of filtering the incremental data generated by each edge node to obtain an incremental dataset includes:

[0020] The incremental data generated by each edge node is filtered using a reinforcement learning strategy;

[0021] The incremental dataset is obtained by dividing the filtered incremental data using a clustering algorithm.

[0022] According to the incremental learning method based on a cloud-edge collaborative architecture provided by the present invention, the step of incrementally learning and training the cloud-based basic model using the incremental dataset to obtain an updated cloud-based basic model includes:

[0023] The cloud-based basic model is trained and updated using the incremental dataset and the incremental learning method to obtain the updated cloud-based basic model.

[0024] The present invention also provides an incremental learning device based on a cloud-edge collaborative architecture, comprising:

[0025] The cloud-based basic model determination module is used to train and obtain the cloud-based basic model based on the raw data collected from each edge node;

[0026] The edge model determination module is used to train and obtain the edge model corresponding to each edge node based on the proprietary data of each edge node and the cloud-based basic model.

[0027] An incremental dataset determination module is used to filter the incremental data generated by each edge node to obtain an incremental dataset, wherein the incremental data is generated by the edge node based on the edge model;

[0028] The model training and update module is used to perform incremental learning training on the cloud-based basic model using the incremental dataset to obtain the updated cloud-based basic model.

[0029] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the incremental learning method based on the cloud-edge collaborative architecture as described above.

[0030] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the incremental learning method based on the cloud-edge collaborative architecture as described above.

[0031] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the incremental learning method based on the cloud-edge collaborative architecture as described above.

[0032] The incremental learning method, apparatus, device, and medium based on a cloud-edge collaborative architecture provided by this invention trains a cloud-based basic model based on raw data collected from each edge node. Then, based on the proprietary data of each edge node and the cloud-based basic model, an edge model corresponding to each edge node is trained. Subsequently, the incremental data generated by each edge node is filtered to obtain an incremental dataset. The incremental data is generated by the edge nodes based on the edge model. The cloud-based basic model is incrementally trained using the incremental dataset to obtain an updated cloud-based basic model. Through the cloud-edge collaborative architecture, the problem that it is difficult to simultaneously meet business requirements for three important indicators—model generalization ability, model accuracy, and inference latency—is solved. The cloud training strategy ensures that the cloud-based basic model can maintain good generalization ability in multiple scenarios. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0034] Figure 1 This is one of the flowcharts of the incremental learning method based on cloud-edge collaborative architecture provided by the present invention;

[0035] Figure 2 This is a schematic diagram illustrating the steps of the incremental learning method based on a cloud-edge collaborative architecture provided by the present invention;

[0036] Figure 3 This is the second flowchart of the incremental learning method based on cloud-edge collaborative architecture provided by the present invention;

[0037] Figure 4 This is a schematic diagram of the incremental learning device based on a cloud-edge collaborative architecture provided by the present invention.

[0038] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0040] The following is combined with Figures 1-3This invention describes an incremental learning method based on a cloud-edge collaborative architecture.

[0041] Please refer to Figure 1 This invention provides an incremental learning method based on a cloud-edge collaborative architecture, comprising:

[0042] Step 100: Train the cloud-based basic model based on the raw data collected from each edge node;

[0043] Specifically, raw data is collected from each edge node and added to the dataset to obtain the initial dataset D0 for model training. The initial dataset is stored in the cloud node. Then, the weighted focusing loss function proposed in this embodiment is used as the loss function for neural network optimization, and the model is trained based on the gradient descent method to obtain the initialized cloud base model, which is the cloud base model in this embodiment.

[0044] Step 200: Based on the proprietary data of each edge node and the cloud-based basic model, train the edge model corresponding to each edge node;

[0045] Specifically, the cloud-based basic model is distributed to each edge node. First, the backbone network of the cloud-based basic model is frozen. Then, through transfer learning, the data from each edge node is used as the input for training the cloud-based basic model, and the focusing loss function is used as the loss function for model training. Small-sample transfer learning is performed on the cloud-based basic model to obtain multiple edge models, namely the edge models in this embodiment, such as... Figure 2 As shown, Figure 2 The edge server in this embodiment refers to the edge node.

[0046] Step 300: Filter the incremental data generated by each edge node to obtain an incremental dataset, wherein the incremental data is generated by the edge node based on the edge model;

[0047] Specifically, incremental model learning is performed using production data collected from edge nodes, with two time periods T set respectively. edge and T cloud T cloud =nT edge Where n is a positive integer, in each time period T edge At the end, the incremental data d generated by the edge nodes edge As input to the edge model, the edge model is updated through an incremental learning method at each time period T. cloudAt the end of the time period, the incremental data generated by each edge node during that period is uploaded to the cloud server. The cloud server uses a reinforcement learning strategy to filter the incremental data uploaded by the edge servers, and uses a clustering algorithm to divide the incremental data uploaded by each edge server into several categories, thereby selecting the dataset that is most helpful in improving the model's generalization ability, which is the incremental dataset in this embodiment.

[0048] Step 400: Incremental learning training is performed on the cloud-based basic model using the incremental dataset to obtain the updated cloud-based basic model.

[0049] Specifically, using the selected incremental dataset, the cloud-based foundation model is trained and updated based on incremental learning to obtain the updated cloud-based foundation model. The selected incremental dataset serves as the training data for the cloud-based foundation model. Using incremental learning, the backbone network of the cloud-based foundation model is frozen, and finally, the updated cloud-based foundation model is obtained through gradient descent algorithm.

[0050] In this embodiment, a cloud-based basic model is trained based on the raw data collected from each edge node. Then, an edge model corresponding to each edge node is trained based on the proprietary data of each edge node and the cloud-based basic model. Subsequently, the incremental data generated by each edge node is filtered to obtain an incremental dataset. The incremental data is generated by the edge nodes based on the edge model. The cloud-based basic model is incrementally trained using the incremental dataset to obtain an updated cloud-based basic model. Through the cloud-edge collaborative architecture, the problem that the three important indicators of model generalization ability, model accuracy, and inference latency are difficult to meet business requirements at the same time is solved. The cloud training strategy ensures that the cloud-based basic model can maintain good generalization ability in multiple scenarios.

[0051] In one embodiment, the incremental learning method provided in this application may further include:

[0052] Step 110: The weighted focusing loss function is used as the loss function for neural network optimization. The model is trained based on the gradient descent method and the raw data collected from each edge node to obtain the cloud-based basic model.

[0053] Specifically, raw data is collected from each edge node and added to the dataset to obtain the initial dataset for model training. The initial dataset is stored in the cloud node. Then, the weighted focusing loss function proposed in this embodiment is used as the loss function for neural network optimization. The model is trained based on the gradient descent method and the raw data collected from each edge node to obtain the initialized cloud base model, which is the cloud base model in this embodiment. The cloud base model mainly includes a one-stage object detection algorithm, such as the YOLO series algorithm. The cloud base model mainly includes a backbone network, a neck network, and a head network. The backbone network is used to extract and aggregate image features at different granularities of the image. The neck network is used to integrate the extracted image features and pass the image features to the head network. The head network is used to make predictions based on the image features and generate results such as bounding boxes, categories, and confidence scores.

[0054]

[0055] The weighted focusing loss function is shown in Equation 1, where WFL k (p t p represents the loss function for class t on the k-th marginal model. t This represents the confidence level of the prediction for category t. The higher the confidence level, the less loss is incurred. This represents the average macroscopic F1 score of each marginal model. k This represents the macroscopic F1 score of the k-th edge model. A higher F1 score for an edge model indicates a smaller difference in loss between easy and difficult samples, and vice versa. By introducing a weighted focusing loss function as the model training loss function, we can balance the impact of easy and difficult samples, addressing the sample imbalance problem within edge nodes, and also balance the imbalance problem between edge nodes. This allows the cloud-based base model to focus more on low-accuracy edge node data during training.

[0056] This embodiment proposes a new weighted focusing loss function that comprehensively considers the impact of performance imbalance between edge nodes and sample imbalance within edge nodes on the model loss, thereby improving the generalization ability of the cloud-based basic model.

[0057] Please refer to Figure 3 In one embodiment, the incremental learning method based on a cloud-edge collaborative architecture provided in this application may further include:

[0058] Step 201: Distribute the cloud-based basic model to each of the edge nodes;

[0059] Step 202: Freeze the backbone network of the cloud-based basic model and use the proprietary data of each edge node as the model training input.

[0060] Step 203: Perform small-sample transfer learning on the cloud-based basic model based on the transfer learning method and the focus loss function to obtain the edge model corresponding to each edge node.

[0061] Specifically, the cloud-based basic model is distributed to each edge node. First, the backbone network of the cloud-based basic model is frozen. Then, using transfer learning, the data from each edge node is used as the input for training the cloud-based basic model, and the focusing loss function is used as the loss function for model training. Small-sample transfer learning is performed on the cloud-based basic model to obtain m edge-end models. This refers to the edge model in this embodiment. The transfer learning method in this embodiment mainly trains the model by freezing the parameters of the backbone network of the cloud-based basic model, and learns the neck network and head network in the above embodiment through gradient descent.

[0062] This embodiment employs transfer learning and incremental learning methods, which solves the problem of insufficient computing resources at the edge on the one hand, and the problems of model drift and degradation on the other.

[0063] In one embodiment, the incremental learning method based on a cloud-edge collaborative architecture provided in this application may further include:

[0064] Step 210: Use the data generated by the edge node within the first time period as the model input, and update the edge model through an incremental learning method;

[0065] Step 220: Upload the incremental data generated by the edge node within the second time period to the cloud server. The second time period is a positive integer multiple of the first time period.

[0066] Specifically, incremental model learning is performed using production data collected from edge nodes, with two time periods T set respectively. edge and T cloud That is, the first time period and the second time period in this embodiment, T cloud =nT edge Where n is a positive integer, and the time interval for incremental learning of each edge model is T. edge The data sample set d generated by the k-th edge node in the t-th time period. t,k As the data input for incremental learning of the edge model corresponding to the k-th edge node, the edge model As a pre-trained model, incremental learning is performed on the frozen edge model backbone network and neck network based on the gradient descent method to obtain a new model. After time period T cloud Then, the incremental data generated by each edge node during that time period is uploaded to the cloud server.

[0067] This embodiment uses an incremental learning method to upload incremental data generated by edge nodes to a cloud server, which helps to further improve the model's generalization ability.

[0068] In one embodiment, the incremental learning method based on a cloud-edge collaborative architecture provided in this application may further include:

[0069] Step 310: Filter the incremental data generated by each edge node using a reinforcement learning strategy;

[0070] Step 320: The filtered incremental data is divided using a clustering algorithm to obtain the incremental dataset.

[0071] Specifically, the cloud server uses a reinforcement learning strategy to filter the incremental data uploaded by the edge server, where the dataset uploaded by m edge nodes in the Tth time period is (D T1 D T2 ,...,D Tm Image features extracted from the backbone network of the cloud-based basic model are used to cluster each dataset separately using clustering algorithms such as DBSCAN (Density-Based Spatial Clustering of Applications with Noise). Each dataset is divided into C subsets, resulting in a total of C×m subsets. In this way, the incremental data that is most helpful in improving the model's generalization ability is selected to form a dataset, namely the incremental dataset in this embodiment.

[0072] This embodiment uses reinforcement learning technology to filter the data uploaded from the edge, retaining only the data that significantly improves the model's generalization ability, which helps to further enhance the model's generalization ability.

[0073] In one embodiment, the incremental learning method based on a cloud-edge collaborative architecture provided in this application may further include:

[0074] Step 410: Train and update the cloud-based basic model using the incremental dataset and the incremental learning method to obtain the updated cloud-based basic model.

[0075] Specifically, the results of the above data subset selection are used as the state function s for reinforcement learning. t Adding or deleting certain subsets is used as the action function a. tThe selected dataset is used as the training data for the cloud-based basic model. Incremental learning is performed on the cloud-based basic model, and the change in the model's F1 score is used as the reward function r for reinforcement learning. t The optimal data subset combination is searched using reinforcement learning algorithms such as Policy Gradient and Actor-Critic (a combination of Q-Learning and Policy Gradient algorithms). The selected data is used as the training data D1 for the cloud model, and D1 is used as the training data for the cloud base model. Incremental learning method is used to freeze the backbone network of the cloud base model. Finally, the updated cloud base model is obtained through gradient descent algorithm.

[0076] This embodiment trains and updates the cloud-based basic model using incremental datasets and incremental learning methods, thereby improving the generalization ability of the cloud-based basic model.

[0077] The incremental learning device provided by the present invention is described below. The incremental learning device described below can be referred to in correspondence with the incremental learning method described above.

[0078] Please refer to Figure 4 The present invention also provides an incremental learning device based on a cloud-edge collaborative architecture, comprising:

[0079] The cloud-based basic model determination module 401 is used to train the cloud-based basic model based on the raw data collected from each edge node.

[0080] The edge model determination module 402 is used to train and obtain the edge model corresponding to each edge node based on the proprietary data of each edge node and the cloud-based basic model.

[0081] The incremental dataset determination module 403 is used to filter the incremental data generated by each edge node to obtain an incremental dataset, wherein the incremental data is generated by the edge node based on the edge model;

[0082] The model training and update module 404 is used to perform incremental learning training on the cloud-based basic model using the incremental dataset to obtain the updated cloud-based basic model.

[0083] Optionally, the cloud-based basic model determination module includes:

[0084] The cloud-based basic model determination unit is used to use the weighted focusing loss function as the loss function for neural network optimization. The model is trained based on the gradient descent method and the raw data collected from each edge node to obtain the cloud-based basic model.

[0085] Optionally, the edge model determination module includes:

[0086] The cloud-based basic model distribution unit is used to distribute the cloud-based basic model to each of the edge nodes;

[0087] The backbone network freezing unit is used to freeze the backbone network of the cloud-based basic model and use the proprietary data of each edge node as the model training input.

[0088] The edge model determination unit is used to perform small-sample transfer learning on the cloud-based basic model based on the transfer learning method and the focus loss function to obtain the edge model corresponding to each edge node.

[0089] Optionally, the incremental learning device includes:

[0090] The edge model update module is used to take the data generated by the edge node in the first time period as the model input and update the edge model through an incremental learning method.

[0091] The incremental data upload module is used to upload the incremental data generated by the edge node within a second time period to the cloud server, wherein the second time period is a positive integer multiple of the first time period.

[0092] Optionally, the incremental dataset determination module includes:

[0093] An incremental data filtering unit is used to filter the incremental data generated by each edge node through a reinforcement learning strategy;

[0094] The incremental dataset determination unit is used to divide the filtered incremental data using a clustering algorithm to obtain the incremental dataset.

[0095] Optionally, the model training update module includes:

[0096] The model training and update unit is used to train and update the cloud-based basic model using the incremental dataset and the incremental learning method to obtain the updated cloud-based basic model.

[0097] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include a processor 510, a communications interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communications interface 520, and the memory 530 communicate with each other through the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute an incremental learning method.

[0098] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several 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 methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0099] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the incremental learning method based on the cloud-edge collaborative architecture provided by the above methods.

[0100] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the incremental learning methods based on the cloud-edge collaborative architecture provided by the above methods.

[0101] The device embodiments described above are merely illustrative. 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 modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0102] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An incremental learning method based on a cloud-edge collaborative architecture, characterized in that, include: The cloud-based basic model is trained based on the raw data collected from each edge node; Based on the proprietary data of each edge node and the cloud-based basic model, the edge model corresponding to each edge node is trained. The incremental data generated by each edge node is filtered to obtain an incremental dataset, wherein the incremental data is generated by the edge node based on the edge model; The cloud-based basic model is incrementally trained using the incremental dataset to obtain an updated cloud-based basic model. The incremental dataset obtained by filtering the incremental data generated by each of the edge nodes includes: The incremental data generated by each edge node is filtered using a reinforcement learning strategy; The filtered incremental data is divided using a clustering algorithm, and the filtered incremental data of each edge node is divided into multiple data subsets. The results of the data subset selection are used as the state function of reinforcement learning, the addition or deletion of certain data subsets are used as the action function, the selected dataset is used as the training data of the cloud-based basic model, incremental learning is performed on the cloud-based basic model, the change of the model's F1 score is used as the reward function of reinforcement learning, the optimal combination of data subsets is searched through the reinforcement learning algorithm, and the selected data is used as the training data of the cloud-based basic model, i.e., the incremental dataset.

2. The incremental learning method based on a cloud-edge collaborative architecture according to claim 1, characterized in that, The cloud-based basic model trained based on raw data collected from each edge node includes: The weighted focusing loss function is used as the loss function for neural network optimization. The model is trained based on gradient descent and raw data collected from each edge node to obtain the cloud-based basic model.

3. The incremental learning method based on a cloud-edge collaborative architecture according to claim 1, characterized in that, The edge model corresponding to each edge node, trained based on the proprietary data of each edge node and the cloud-based basic model, includes: The cloud-based basic model is then distributed to each of the edge nodes. Freeze the backbone network of the cloud-based basic model and use the proprietary data of each edge node as the model training input; The cloud-based basic model is subjected to small-sample transfer learning based on the transfer learning method and the focus loss function to obtain the edge model corresponding to each edge node.

4. The incremental learning method based on a cloud-edge collaborative architecture according to claim 1, characterized in that, After training the edge model corresponding to each edge node based on the proprietary data of each edge node and the cloud-based basic model, the process includes: The data generated by the edge nodes within the first time period is used as the model input, and the edge model is updated through an incremental learning method. The incremental data generated by the edge node within the second time period is uploaded to the cloud server, where the second time period is a positive integer multiple of the first time period.

5. The incremental learning method based on a cloud-edge collaborative architecture according to claim 4, characterized in that, The step of incrementally training the cloud-based basic model using the incremental dataset to obtain the updated cloud-based basic model includes: The cloud-based basic model is trained and updated using the incremental dataset and the incremental learning method to obtain the updated cloud-based basic model.

6. An incremental learning device based on a cloud-edge collaborative architecture, characterized in that, include: The cloud-based basic model determination module is used to train and obtain the cloud-based basic model based on the raw data collected from each edge node; The edge model determination module is used to train and obtain the edge model corresponding to each edge node based on the proprietary data of each edge node and the cloud-based basic model. An incremental dataset determination module is used to filter the incremental data generated by each edge node to obtain an incremental dataset, wherein the incremental data is generated by the edge node based on the edge model; The model training and update module is used to perform incremental learning training on the cloud-based basic model using the incremental dataset to obtain the updated cloud-based basic model. The incremental dataset determination module is specifically used for: The incremental data generated by each edge node is filtered using a reinforcement learning strategy; The filtered incremental data is divided using a clustering algorithm, and the filtered incremental data of each edge node is divided into multiple data subsets. The results of the data subset selection are used as the state function of reinforcement learning, the addition or deletion of certain data subsets are used as the action function, the selected dataset is used as the training data of the cloud-based basic model, incremental learning is performed on the cloud-based basic model, the change of the model's F1 score is used as the reward function of reinforcement learning, the optimal combination of data subsets is searched through the reinforcement learning algorithm, and the selected data is used as the training data of the cloud-based basic model, i.e., the incremental dataset.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the incremental learning method based on the cloud-edge collaborative architecture as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the incremental learning method based on the cloud-edge collaborative architecture as described in any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the incremental learning method based on the cloud-edge collaborative architecture as described in any one of claims 1 to 5.

Images