Feature compression methods, devices, equipment, media, and products for edge computing scenarios

Abstract

This application provides a feature compression method, apparatus, device, medium, and product for edge computing scenarios, including: acquiring a set of target features to be processed; sampling the set of target features to be processed according to a Monte Carlo simulation algorithm to generate a feature subset, and calculating the contribution of multiple features based on the feature subset; selecting key features based on the contribution of multiple features through a feature correlation pruning strategy; constructing a feature-compressed set based on the key features, and performing model training operations on an edge computing device based on the feature-compressed set, thereby improving the efficiency of edge computing.

Description

Technical Field

[0001] This application relates to the field of edge computing technology, and in particular to a feature compression method, apparatus, device, medium and product for edge computing scenarios. Background Technology

[0002] Edge computing is a computing model that extends data processing and storage capabilities from centralized data centers to edge devices or nodes closer to the data source, aiming to solve the latency, bandwidth limitations, and data privacy issues inherent in traditional cloud computing models. In edge computing scenarios, the data processed by edge devices typically contains high-dimensional and highly redundant feature sets. Directly embedding and interacting with these features on edge devices would significantly increase model inference time. Therefore, implementing efficient feature compression strategies in edge computing scenarios becomes a key issue in ensuring system stability and operational efficiency.

[0003] In existing technologies, feature compression mainly uses principal component analysis, which preserves the direction of maximum variance through linear transformation to achieve feature dimensionality reduction.

[0004] However, existing methods have high computational complexity and weak ability to handle nonlinear features, resulting in reduced efficiency of edge computing. Summary of the Invention

[0005] This application provides a feature compression method, apparatus, device, medium, and product for edge computing scenarios to solve the problem of reduced efficiency in edge computing in the prior art.

[0006] In a first aspect, embodiments of this application provide a feature compression method for edge computing scenarios, including:

[0007] Obtain the set of target features to be processed;

[0008] The target feature set to be processed is sampled according to the Monte Carlo simulation algorithm to generate a feature subset, and the contribution of multiple features is calculated based on the feature subset.

[0009] Key features are obtained by filtering based on the contribution of the multiple features using a feature relevance pruning strategy;

[0010] A compressed feature set is constructed based on the key features, and model training is performed on the edge computing device based on the compressed feature set.

[0011] In one possible implementation, the step of sampling the target feature set to be processed according to the Monte Carlo simulation algorithm to generate a feature subset, and calculating the contribution of multiple features based on the feature subset, includes: determining the number of samplings for the Monte Carlo simulation using a random algorithm; selecting features from the target feature set to be processed according to the number of samplings to generate a feature subset; and calculating the contribution of multiple features in the feature subset based on the feature subset.

[0012] In one possible implementation, before calculating the contribution of multiple features in the feature subset based on the feature subset, the method further includes: performing statistical analysis on the data distribution of the feature subset within multiple historical time windows to obtain the changing trend of key features; adjusting the Monte Carlo simulation parameters based on the changing trend of the key features; and performing the contribution calculation operation based on the adjusted Monte Carlo simulation parameters.

[0013] In one possible implementation, before sampling the target feature set to be processed according to the Monte Carlo simulation algorithm, the method further includes: acquiring the operating status information of the edge computing device; comparing the operating status information of the edge computing device with a preset resource threshold to determine the resource load status of the edge computing device and generating a resource load judgment result; generating an adjustment strategy for the feature contribution threshold based on the resource load judgment result; adjusting the feature contribution threshold according to the adjustment strategy to obtain the adjusted feature contribution threshold, and updating the feature correlation pruning strategy based on the adjusted feature contribution threshold.

[0014] In one possible implementation, the step of constructing a compressed feature set based on the key features and performing model training on an edge computing device based on the compressed feature set includes: creating a feature interaction graph based on the selected key features and their corresponding contributions, wherein the key features serve as nodes in the feature interaction graph; adjusting the weights of the nodes in the feature interaction graph according to set business objectives to generate adjusted node weights; in response to performing model training on the edge computing device, adjusting the weight update frequency of the feature interaction graph according to the load status of the edge computing device and the priority of the business objectives to obtain an optimized feature interaction graph; constructing a compressed feature set based on the optimized feature interaction graph, and performing model training on the edge computing device based on the compressed feature set.

[0015] In one possible implementation, adjusting the weights of nodes in the feature interaction graph according to the set business objectives to generate adjusted node weights includes: obtaining the filtering period of the feature interaction graph; in response to the triggering operation of the filtering period of the feature interaction graph, traversing the nodes in the feature interaction graph and extracting the historical contribution sequence and target weight value of each node; calculating the filtering score of each node based on the historical contribution sequence and target weight value of each node using a preset scoring model; sorting the nodes according to the filtering scores of each node; filtering the sorted nodes to obtain an update result, and updating the feature interaction graph according to the update result.

[0016] Secondly, embodiments of this application provide a feature compression device for an edge computing scenario, comprising:

[0017] The first acquisition module is used to acquire the set of target features to be processed;

[0018] The sampling module is used to sample the target feature set to be processed according to the Monte Carlo simulation algorithm, generate a feature subset, and calculate the contribution of multiple features based on the feature subset;

[0019] The filtering module is used to filter key features based on the contribution of the multiple features using a feature relevance pruning strategy.

[0020] The construction module is used to construct a feature set after feature compression based on the key features, and to perform model training operations on the edge computing device based on the feature set after feature compression.

[0021] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0022] The memory stores computer-executed instructions;

[0023] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0024] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0025] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0026] The feature compression method, apparatus, device, medium, and product for edge computing scenarios provided in this application introduce a Monte Carlo simulation algorithm to randomly sample the target feature set to calculate the feature contribution. Based on the contribution, a feature correlation pruning strategy is used to select key features. The compressed feature set is then used to perform model training on the edge device. This applies feature evaluation and pruning compression technology based on probability sampling to edge computing scenarios, solving the problem of low efficiency in feature model training under the condition of limited computing and storage resources on edge devices. It achieves intelligent and lightweight compression of the feature set while ensuring the core performance of the model, reducing the complexity and resource consumption of model training, and improving the efficiency of edge computing. Attached Figure Description

[0027] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0028] Figure 1 This is a schematic diagram of the system structure of a computer device provided in an embodiment of this application;

[0029] Figure 2 A flowchart illustrating the feature compression method for edge computing scenarios provided in this application. Figure 1 ;

[0030] Figure 3 A flowchart illustrating the feature compression method for edge computing scenarios provided in this application. Figure 2 ;

[0031] Figure 4 A flowchart illustrating the feature compression method for edge computing scenarios provided in this application. Figure 3 ;

[0032] Figure 5 A flowchart illustrating the feature compression method for edge computing scenarios provided in this application. Figure 4 ;

[0033] Figure 6 A schematic diagram of the feature compression device for the edge computing scenario provided in this application;

[0034] Figure 7 A schematic diagram of the structure of the electronic device provided in this application.

[0035] The accompanying drawings have illustrated specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation

[0036] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0037] Edge computing is a computing model that extends data processing and storage capabilities from centralized data centers to edge devices or nodes closer to the data source, aiming to solve the latency, bandwidth limitations, and data privacy issues inherent in traditional cloud computing models. In edge computing scenarios, the data processed by edge devices typically contains high-dimensional and highly redundant feature sets. Directly embedding and interacting with these features on edge devices significantly increases model inference time. Therefore, implementing efficient feature compression strategies in edge computing scenarios is crucial for ensuring system stability and operational efficiency. Current technologies primarily utilize principal component analysis (PCA) for feature compression, employing linear transformations to preserve the direction of maximum variance and achieve feature dimensionality reduction. However, existing methods suffer from high computational complexity and weak processing capabilities for nonlinear features, leading to reduced efficiency in edge computing.

[0038] To address the aforementioned technical problems, this application proposes the following technical concept: The inventors considered introducing a Monte Carlo simulation algorithm to randomly sample the target feature set, calculate the contribution, and use a feature correlation pruning strategy to select key features based on the contribution. The compressed feature set is then used to perform model training on an edge device. By compressing the feature set, the complexity and resource consumption of model training are reduced, and the efficiency of edge computing is improved.

[0039] Figure 1 This is a schematic diagram of the system architecture of the computer device provided in an embodiment of this application. Figure 1 As shown, the computer device includes: a receiving device 101, a processing device 102, and a display device 103.

[0040] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the feature compression method for edge computing scenarios. In other feasible embodiments of this application, the above architecture may include more or fewer components than illustrated, or combine some components, or split some components, or arrange different components, which can be determined according to the actual application scenario and is not limited here. Figure 1 The components shown can be implemented in hardware, software, or a combination of both.

[0041] In the specific implementation process, the receiving device 101 can be an input / output interface or a communication interface, and can acquire the target feature set to be processed.

[0042] The processing device 102 can construct a feature set with compressed features based on key features, and perform model training operations on the edge computing device based on the feature set with compressed features.

[0043] The display device 103 can be used to display the feature set after the feature compression described above.

[0044] The display device can also be a touch screen, used to receive user commands while displaying the above content, so as to realize the operation interaction with the user.

[0045] It should be understood that the above-mentioned processing device can be implemented by a processor reading instructions from memory and executing those instructions, or it can be implemented by a chip circuit.

[0046] Furthermore, the network architecture and business scenarios described in the embodiments of this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided in the embodiments of this application. As those skilled in the art will know, with the evolution of network architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0047] Figure 2 A flowchart illustrating the feature compression method for edge computing scenarios provided in this application. Figure 1 ,like Figure 2 As shown, the method includes:

[0048] S201: Obtain the set of target features to be processed.

[0049] Specifically, the raw data stream is periodically retrieved through the data acquisition module on the edge computing device.

[0050] In this embodiment, the raw data includes, but is not limited to, timing signals transmitted back by industrial equipment, equipment status identification boxes, and service tags.

[0051] Specifically, the raw data is standardized and cleaned, and the feature engineering library is used to automatically extract time-domain, frequency-domain, and time-series statistical features, which are then aggregated and packaged into a structured set of target features to be processed.

[0052] S202: The Monte Carlo simulation algorithm is used to sample the target feature set to be processed, generate a feature subset, and calculate the contribution of multiple features based on the feature subset.

[0053] Specifically, the number of sampling times for the Monte Carlo simulation is determined by a random algorithm, features are selected from the target feature set to be processed, a feature subset is generated, and the contribution of multiple features in the feature subset is calculated based on the feature subset.

[0054] S203: Key features are obtained by filtering based on the contribution of multiple features through a feature correlation pruning strategy.

[0055] Specifically, the contribution of each calculated feature is compared with the adjusted feature contribution threshold. Features with a contribution below the threshold are removed. The Pearson correlation coefficient matrix between the remaining features is calculated. After contribution filtering and redundancy removal, the key features are obtained.

[0056] S204: Construct a feature set with compressed features based on key features, and perform model training operations on the edge computing device based on the feature set with compressed features.

[0057] Specifically, a feature interaction graph is created based on the selected key features and their corresponding contributions. The weights of the nodes in the feature interaction graph are adjusted according to the set business objectives to generate adjusted node weights. The weight update frequency of the feature interaction graph is adjusted according to the load status of the edge computing device and the priority of the business objectives. A feature set with compressed features is constructed based on the optimized feature interaction graph. Model training is then performed on the edge computing device based on the feature set with compressed features.

[0058] As can be seen from the above embodiments, by introducing the Monte Carlo simulation algorithm to randomly sample the target feature set to calculate the feature contribution, and using the feature correlation pruning strategy based on the contribution to select key features, and using the compressed feature set to perform model training on the edge device, the feature evaluation and pruning compression technology based on probability sampling is applied to the edge computing scenario. This solves the problem of low efficiency in feature model training under the condition of limited computing and storage resources on edge devices. It realizes intelligent and lightweight compression of the feature set while ensuring the core performance of the model, reducing the complexity and resource consumption of model training, and improving the efficiency of edge computing.

[0059] In one embodiment of this application, step S202 includes:

[0060] S2021: Determine the number of samples for the Monte Carlo simulation using a random algorithm.

[0061] Specifically, a baseline number of samplings is set, and the number of samplings is randomly adjusted within a preset range based on the current device usage rate.

[0062] S2022: Select features from the target feature set to be processed based on the number of samplings to generate a feature subset.

[0063] Specifically, a random sampling strategy without replacement is adopted to randomly select a fixed proportion of features from the features of the target feature set to be processed, forming a feature subset.

[0064] S2023: Calculate the contribution of multiple features in the feature subset based on the feature subset.

[0065] In this embodiment, the formula for calculating the contribution value is:

[0066]

[0067] In the formula, N represents the set of all features; i represents the feature index to be calculated, where i∈N; S represents any subset of the feature set N that does not include feature i; This indicates the number of features contained in subset S; Indicates the total number of features; This indicates that after adding feature i to subset S, the subset is used. The function value predicted from the features in the data; This represents the function value used for prediction based on features from subset S; This represents the contribution value of feature i.

[0068] As can be seen from the above embodiments, by determining the number of samplings through a random algorithm, the randomness of sampling and the effectiveness of statistics are ensured. By transforming the abstract Monte Carlo simulation process into a specific, executable, and controllable operation, the operability of data compression in actual deployment is improved.

[0069] Figure 3 A flowchart illustrating the feature compression method for edge computing scenarios provided in this application. Figure 2 ,like Figure 3 As shown, the procedure before step S2023 also includes:

[0070] S301: Perform statistical analysis on the data distribution of the feature subset within multiple historical time windows to obtain the changing trends of key features.

[0071] Specifically, on edge devices, the system maintains a historical window buffer for feature data, storing the data sequence of each feature over historical time periods. For frequently occurring features in multiple generated feature subsets, the system obtains their data changes within the historical time window and analyzes the trend of feature data changes through linear regression or by calculating a moving average.

[0072] S302: Adjust Monte Carlo simulation parameters based on the changing trends of key features.

[0073] Specifically, the sampling strategy of the Monte Carlo simulation is dynamically adjusted based on the changing trends obtained from the analysis.

[0074] For example, for key features whose data distribution changes drastically, the probability of them being included in the feature subset is increased in subsequent sampling by increasing the initial weight of the extracted features.

[0075] S303: Perform contribution calculation based on the adjusted Monte Carlo simulation parameters.

[0076] Specifically, for each generated feature subset, a performance metric for that feature subset on the validation dataset is calculated using a lightweight evaluation model.

[0077] Specifically, the marginal contribution of each feature in the subset is calculated using an algorithm based on perturbation or permutation. The calculation results of all subsets are statistically averaged to obtain the average contribution of each feature in the target feature set to be processed.

[0078] As can be seen from the above embodiments, by analyzing the changing trends of key features in the historical window and optimizing the simulation parameters according to the changing trends, the sampling process can better capture the dynamic evolution of the importance of features, thereby improving the robustness of contribution calculation to data distribution drift and its sensitivity to the current state.

[0079] Figure 4 A flowchart illustrating the feature compression method for edge computing scenarios provided in this application. Figure 3 ,like Figure 4 As shown, the procedure before step S202 also includes:

[0080] S401: Obtain the operating status information of the edge computing device.

[0081] In this embodiment, the operating status information includes, but is not limited to, the real-time utilization of the central processing unit (CPU), the available memory capacity, the current network input / output (I / O) bandwidth utilization, and the remaining space of the storage medium.

[0082] S402: Compare the operating status information of the edge computing device with the preset resource threshold to determine the resource load status of the edge computing device and generate a resource load judgment result.

[0083] Specifically, multiple resource thresholds are set, and the real-time collected operating status information is compared with the thresholds to determine the resource load status of the equipment and generate the corresponding resource load judgment result.

[0084] S403: An adjustment strategy for generating characteristic contribution thresholds based on resource load assessment results.

[0085] Specifically, the feature contribution threshold used for pruning is dynamically adjusted based on the resource load assessment results.

[0086] For example, when the device is in a "heavy load" or "overload" state, the generated adjustment strategy is to "increase the feature contribution threshold by 20%" so that subsequent pruning can be more aggressive, retaining fewer key features and reducing the computing power and memory overhead of model training.

[0087] S404: Adjust the feature contribution threshold according to the feature contribution threshold adjustment strategy to obtain the adjusted feature contribution threshold, and update the feature correlation pruning strategy according to the adjusted feature contribution threshold.

[0088] Specifically, the adjustment strategy is executed, and the adjusted new threshold is synchronized to the pruning strategy module.

[0089] As can be seen from the above embodiments, by evaluating the equipment resource load and dynamically adjusting the feature contribution threshold and pruning strategy according to the equipment resource load, a precise match between compression intensity and equipment carrying capacity is achieved. This prevents performance degradation or task failure caused by over-computation when the equipment is under high load, and ensures that the feature compression and model training process runs adaptively within the resource boundaries that the equipment can bear, thereby improving the practicality and stability of the method in edge environments.

[0090] In one embodiment of this application, step S204 includes:

[0091] S2041: Create a feature interaction graph based on the selected key features and their corresponding contributions, where the key features are nodes in the feature interaction graph.

[0092] Specifically, each key feature is treated as a node, and the initial attributes of the node are its feature name and contribution value. The mutual information or conditional dependency between key features in historical data is analyzed. If the mutual information exceeds the threshold, an undirected edge is established between the corresponding two nodes, and the weight of the edge is initialized to the mutual information value.

[0093] S2042: Adjust the weights of nodes in the feature interaction graph according to the set business objectives, and generate the adjusted node weights.

[0094] Specifically, the filtering period of the feature interaction graph is obtained. In response to the triggering operation of the filtering period of the feature interaction graph, the nodes in the feature interaction graph are traversed, and the historical contribution sequence and target weight value of each node are extracted. The filtering score of each node is calculated and sorted. The sorted nodes are filtered to obtain the update result, and the feature interaction graph is updated according to the update result.

[0095] S2043: In response to the operation of performing model training on the edge computing device, adjust the weight update frequency of the feature interaction graph according to the load status of the edge computing device and the priority of business objectives to obtain an optimized feature interaction graph.

[0096] Specifically, when the system is ready to start model training, it reads the resource load assessment results and the priority of the current business objectives, and formulates a dynamic update strategy.

[0097] For example, if the device load is "light load" and the service priority is "high", then the weight update frequency of the feature interaction graph node is set to "high frequency".

[0098] S2044: Construct a compressed feature set based on the optimized feature interaction graph, and perform model training operations on the edge computing device based on the compressed feature set.

[0099] Specifically, the original data fields corresponding to all nodes in the optimized feature interaction graph are extracted from the original feature matrix to form a feature set with reduced dimensionality and compression.

[0100] Specifically, the training engine on the edge computing device loads the compressed feature set and corresponding label data, starts the training task of the neural network, and applies group regularization constraints to strongly correlated feature groups according to the weights of the edges in the feature interaction graph during training.

[0101] As can be seen from the above embodiments, by transforming feature compression into a context-aware graph optimization process, deep coupling between compression strategy and business requirements and elastic utilization of computing resources are achieved, making the final compressed feature set more adaptable to the real-time changing edge task requirements, and improving the business relevance of the training model and the overall resource utilization efficiency.

[0102] Figure 5 A flowchart illustrating the feature compression method for edge computing scenarios provided in this application. Figure 4 ,like Figure 5 As shown, step S2042 includes:

[0103] S501: Filtering cycle for obtaining feature interaction maps.

[0104] Specifically, the preset filtering cycle is a fixed time interval or event trigger, and the system starts the update and filtering process of the feature interaction graph on a regular or triggered basis.

[0105] S502: In response to the triggering operation of the feature interaction graph's filtering cycle, traverse the nodes in the feature interaction graph and extract the historical contribution sequence and target weight value of each node.

[0106] Specifically, when the screening period arrives, the system traverses all nodes in the graph. For each node feature, it extracts the contribution value sequence within the historical evaluation period and queries the target weight value of the feature under the current target from the preset configuration table according to the current business objective.

[0107] S503: Calculate the screening score for each node based on its historical contribution sequence and target weight value using a preset scoring model.

[0108] In this embodiment, a linear scoring model is used to calculate the screening score.

[0109] S504: Sort the nodes according to the screening score of each node.

[0110] Specifically, after calculating the screening score for all nodes, all nodes are sorted from highest to lowest score.

[0111] S505: Filter the sorted nodes to obtain the updated results, and update the feature interaction graph based on the updated results.

[0112] Specifically, a node retention ratio is set, retaining a specified number of nodes ranked first, and removing the remaining nodes from the graph. The edges associated with the removed nodes are also deleted, resulting in a simplified feature interaction graph that better meets the current business objectives. This simplified graph is then used to replace the original feature interaction graph.

[0113] As can be seen from the above embodiments, by combining historical contribution sequences with target weights for comprehensive scoring and ranking, and continuously eliminating features with decreasing importance, the feature compression set can be continuously evolved with changes in business data and objectives, thus avoiding feature stagnation.

[0114] Figure 6 A schematic diagram of the feature compression device for the edge computing scenario provided in this application is shown below. Figure 6 As shown, the feature compression device 60 for edge computing scenarios provided in this embodiment includes: a first acquisition module 601, a sampling module 602, a filtering module 603, and a construction module 604.

[0115] The first acquisition module 601 is used to acquire the set of target features to be processed.

[0116] The sampling module 602 is used to sample the target feature set to be processed according to the Monte Carlo simulation algorithm, generate a feature subset, and calculate the contribution of multiple features based on the feature subset.

[0117] The filtering module 603 is used to filter key features based on the contribution of multiple features using a feature correlation pruning strategy.

[0118] Module 604 is used to construct a feature set with compressed features based on key features, and to perform model training operations on the edge computing device based on the feature set with compressed features.

[0119] In one embodiment of this application, the sampling module 602 includes:

[0120] The determination unit is used to determine the number of samples in the Monte Carlo simulation using a random algorithm.

[0121] The generation unit is used to select features from the target feature set to be processed based on the number of samplings, and generate a feature subset.

[0122] The calculation unit is used to calculate the contribution of multiple features in the feature subset based on the feature subset.

[0123] In one embodiment of this application, the sampling module 602 further includes:

[0124] The statistical unit is used to perform statistical analysis on the data distribution of feature subsets within multiple historical time windows to obtain the changing trends of key features.

[0125] The first adjustment unit is used to adjust the Monte Carlo simulation parameters according to the changing trends of key features.

[0126] The execution unit is used to perform contribution calculation operations based on the adjusted Monte Carlo simulation parameters.

[0127] In one embodiment of this application, the feature compression device 60 for edge computing scenarios further includes:

[0128] The second acquisition module is used to acquire the operating status information of the edge computing device.

[0129] The comparison module is used to compare the operating status information of the edge computing device with the preset resource threshold, determine the resource load status of the edge computing device, and generate a resource load judgment result.

[0130] The generation module is used to generate adjustment strategies for feature contribution thresholds based on resource load assessment results.

[0131] The adjustment module is used to adjust the feature contribution threshold according to the adjustment strategy of the feature contribution threshold, obtain the adjusted feature contribution threshold, and update the feature correlation pruning strategy according to the adjusted feature contribution threshold.

[0132] In one embodiment of this application, the construction module 604 includes:

[0133] Create a unit to generate a feature interaction graph based on the selected key features and their corresponding contributions, where the key features serve as nodes in the feature interaction graph.

[0134] The second adjustment unit is used to adjust the weights of nodes in the feature interaction graph according to the set business objectives, and generate the adjusted node weights.

[0135] The third adjustment unit is used to respond to the operation of model training on the edge computing device, and adjust the weight update frequency of the feature interaction graph according to the load status of the edge computing device and the priority of business objectives to obtain the optimized feature interaction graph.

[0136] The building unit is used to construct a feature set with compressed features based on the optimized feature interaction graph, and to perform model training operations on the edge computing device based on the feature set with compressed features.

[0137] In one embodiment of this application, the second adjustment unit includes:

[0138] Obtain sub-units to obtain the filtering period of the feature interaction graph.

[0139] Extract sub-units to respond to the triggering operation of the feature interaction graph's filtering cycle, traverse the nodes in the feature interaction graph, and extract the historical contribution sequence and target weight value of each node.

[0140] The calculation subunit is used to calculate the screening score of each node based on the historical contribution sequence and target weight value of each node using a preset scoring model.

[0141] The sorting subunit is used to sort nodes based on the screening score of each node.

[0142] The filtering sub-unit is used to filter the sorted nodes, obtain the updated results, and update the feature interaction graph based on the updated results.

[0143] The feature compression device for edge computing scenarios provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0144] Figure 7 A schematic diagram of the structure of the electronic device provided in this application. Figure 7 As shown, the electronic device 70 provided in this embodiment includes at least one processor 701 and a memory 702. Optionally, the electronic device 70 further includes a communication component 703. The processor 701, memory 702, and communication component 703 are connected via a bus.

[0145] In the specific implementation process, at least one processor 701 executes computer execution instructions stored in memory 702, causing at least one processor 701 to execute the feature compression method for the edge computing scenario described above.

[0146] The specific implementation process of processor 701 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0147] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0148] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0149] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0150] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the feature compression method for the edge computing scenario described above.

[0151] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the feature compression method for the edge computing scenario described above.

[0152] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0153] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0154] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0155] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0156] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0157] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, 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 of the various embodiments of this 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.

[0158] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0159] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A feature compression method for edge computing scenarios, characterized in that, include: Obtain the set of target features to be processed; The target feature set to be processed is sampled according to the Monte Carlo simulation algorithm to generate a feature subset, and the contribution of multiple features is calculated based on the feature subset. Key features are obtained by filtering based on the contribution of the multiple features using a feature relevance pruning strategy; A compressed feature set is constructed based on the key features, and model training is performed on the edge computing device based on the compressed feature set.

2. The method according to claim 1, characterized in that, The step of sampling the target feature set to be processed according to the Monte Carlo simulation algorithm to generate a feature subset, and calculating the contribution of multiple features based on the feature subset, includes: The number of samples for the Monte Carlo simulation is determined using a random algorithm; Features are selected from the target feature set to be processed based on the number of samplings to generate a feature subset; The contribution of multiple features in the feature subset is calculated based on the feature subset.

3. The method according to claim 2, characterized in that, Before calculating the contribution of multiple features in the feature subset based on the feature subset, the method further includes: Statistical analysis was performed on the data distribution of the feature subset within multiple historical time windows to obtain the changing trends of key features; Adjust the Monte Carlo simulation parameters according to the changing trends of the key features; The contribution calculation operation is performed based on the adjusted Monte Carlo simulation parameters.

4. The method according to claim 1, characterized in that, Before sampling the target feature set to be processed according to the Monte Carlo simulation algorithm, the method further includes: Obtain operational status information of edge computing devices; The operating status information of the edge computing device is compared with a preset resource threshold to determine the resource load status of the edge computing device and generate a resource load determination result. An adjustment strategy for the characteristic contribution threshold is generated based on the resource load assessment results; The feature contribution threshold is adjusted according to the adjustment strategy of the feature contribution threshold to obtain the adjusted feature contribution threshold, and the feature correlation pruning strategy is updated according to the adjusted feature contribution threshold.

5. The method according to claim 1, characterized in that, The step of constructing a compressed feature set based on the key features, and performing model training on an edge computing device based on the compressed feature set, includes: A feature interaction graph is created based on the selected key features and their corresponding contributions, wherein the key features serve as nodes in the feature interaction graph; The weights of the nodes in the feature interaction graph are adjusted according to the set business objectives to generate adjusted node weights; In response to the operation of performing model training on the edge computing device, the weight update frequency of the feature interaction graph is adjusted according to the load status of the edge computing device and the priority of business objectives to obtain an optimized feature interaction graph. A compressed feature set is constructed based on the optimized feature interaction graph, and model training is performed on the edge computing device based on the compressed feature set.

6. The method according to claim 5, characterized in that, The step of adjusting the weights of nodes in the feature interaction graph according to the set business objectives to generate adjusted node weights includes: The filtering period for obtaining feature interaction graphs; In response to the triggering operation of the filtering cycle of the feature interaction graph, the nodes in the feature interaction graph are traversed, and the historical contribution sequence and target weight value of each node are extracted. The screening score for each node is calculated based on the historical contribution sequence and target weight value of each node using a preset scoring model. The nodes are sorted according to the screening score of each node; The sorted nodes are filtered to obtain updated results, and the feature interaction graph is updated based on the updated results.

7. A feature compression device for edge computing scenarios, characterized in that, include: The first acquisition module is used to acquire the set of target features to be processed; The sampling module is used to sample the target feature set to be processed according to the Monte Carlo simulation algorithm, generate a feature subset, and calculate the contribution of multiple features based on the feature subset; The filtering module is used to filter key features based on the contribution of the multiple features using a feature relevance pruning strategy. The construction module is used to construct a feature set after feature compression based on the key features, and to perform model training operations on the edge computing device based on the feature set after feature compression.

8. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the feature compression method for edge computing scenarios as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the feature compression method for edge computing scenarios as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the feature compression method for the edge computing scenario as described in any one of claims 1 to 6.

Images