KR-102963405-B1 - System and method for improving artificial intelligence performance of edge devices

KR102963405B1KR 102963405 B1KR102963405 B1KR 102963405B1KR-102963405-B1

Abstract

An artificial intelligence performance improvement system for an edge device according to one embodiment of the present invention comprises an edge device that performs deep learning-based inference operations, a model management server that manages a model installed on the edge device based on the inference results of the edge device, and an MLOps (Machine learning operations) platform that builds a model for improving the performance of the edge device, wherein the model management server monitors the inference operations of the edge device and, if performance degradation below a preset reliability level occurs, provides a previously built model to replace the model installed on the edge device, and the MLOps platform can build a new model corresponding to the environment in which the edge device operates and the inference task performed by the edge device.

Inventors

김병희
이종국

Assignees

주식회사 써로마인드

Dates

Publication Date: 20260512
Application Date: 20250724
Priority Date: 20241206

Claims (16)

Edge device that performs deep learning-based inference operations; A model management server that manages a model mounted on the edge device based on the inference results of the edge device; and A machine learning operations (MLOps) platform for building a model to improve the performance of the edge device; comprising, The above model management server is, If the inference operation of the edge device is monitored and performance degradation below a preset reliability level occurs, a pre-established model is provided to replace the model installed on the edge device, and Based on reinforcement learning, the threshold value of the model used in the edge device is adjusted and transmitted to the edge device, and The above MLOps platform is, A new model corresponding to the environment in which the edge device operates and the inference task performed by the edge device is constructed, and The edge device updates the model used in the edge device based on the adjusted threshold received from the model management server, and The above model management server is, Using the environment information, network status, inference task characteristics, and performance indicators of the above edge device, the threshold adjustment amount of the above model, the timing of model replacement, and whether to perform an update are determined according to a reinforcement learning-based policy, and The above policy applies even when it is determined that a new model needs to be built due to changes in the operating environment and inference tasks of the edge device, and The above MLOps platform is, An artificial intelligence performance improvement system for an edge device, comprising building a model corresponding to the environment in which the edge device operates and the inference task, and generating a model provided according to the policy application result performed by the model management server.
In paragraph 1, The above edge device is, An artificial intelligence performance improvement system for an edge device, wherein an environment profile containing information about the internal and external environments of the edge device is transmitted to the model management server along with a request from the model.
In paragraph 2, The above model management server is, An artificial intelligence performance improvement system for an edge device, wherein one of the models loaded in a model database is selected by referring to the environment profile received from the edge device and transmitted to the edge device.
In paragraph 1, The above edge device is, If the reliability of the computation result according to the above deep learning-based inference computation is below a preset threshold, An artificial intelligence performance improvement system for an edge device, wherein the above calculation result and environment information of the edge device during inference calculation are transmitted to the model management server.
In paragraph 4, The above model management server is, An artificial intelligence performance improvement system for an edge device, which compares the calculation result received from the edge device with the calculation result calculated from the same model provided to the edge device and loaded in the model database by referring to previously stored performance data for each edge device.
delete
In paragraph 1, The above edge device is, If the performance of the model does not improve even after the application of the adjusted threshold value, metadata including the environment profile of the edge device is transmitted to the model management server, and The above model management server is, An artificial intelligence performance improvement system for an edge device, which builds a state-specific model of the edge device based on received metadata.
In Paragraph 7, The above model management server is, An artificial intelligence performance improvement system for edge devices that references metadata for multiple edge devices when constructing the above-mentioned state-specific model.
(a) a step in which an edge device performs a deep learning-based inference operation; and (b) a step in which a model management server manages a model mounted on the edge device based on the inference result of the edge device; wherein The above model management server is, If the inference operation of the edge device is monitored and performance degradation below a preset reliability level occurs, a pre-established model is provided to replace the model installed on the edge device, and Based on reinforcement learning, the threshold value of the model used in the edge device is adjusted and transmitted to the edge device, and A new model is received from an MLOps platform that builds a new model corresponding to the environment in which the edge device operates and the inference task performed by the edge device, and is provided to the edge device. The edge device updates the model used in the edge device based on the adjusted threshold received from the model management server, and The above model management server is, Using the environment information, network status, inference task characteristics, and performance indicators of the above edge device, the threshold adjustment amount of the above model, the timing of model replacement, and whether to perform an update are determined according to a reinforcement learning-based policy, and The above policy applies even when it is determined that a new model needs to be built due to changes in the operating environment and inference tasks of the edge device, and The above MLOps platform is, A method for improving the artificial intelligence performance of an edge device, comprising constructing a model corresponding to the environment in which the edge device operates and the inference task, and generating a model provided according to the result of applying a policy performed on the model management server.
In Paragraph 9, The above edge device is, A method for improving the artificial intelligence performance of an edge device, wherein an environment profile containing information about the internal and external environments of the edge device is transmitted to the model management server along with a request from the model.
In Paragraph 10, The above model management server is, A method for improving the artificial intelligence performance of an edge device, wherein one of the models loaded in a model database is selected by referring to the environment profile received from the edge device and transmitted to the edge device.
In Paragraph 9, The above edge device is, If the reliability of the computation result according to the above deep learning-based inference computation is below a preset threshold, A method for improving the artificial intelligence performance of an edge device, wherein the above calculation result and environment information of the edge device during inference calculation are transmitted to the model management server.
In Paragraph 12, The above model management server is, A method for improving the artificial intelligence performance of an edge device, wherein the method involves comparing the calculation result received from the edge device with the calculation result calculated from the same model provided to the edge device and loaded in the model database, by referring to previously stored performance data for each edge device.
delete
In Paragraph 9, The above edge device is, If the performance of the model does not improve even after the application of the adjusted threshold value, metadata including the environment profile of the edge device is transmitted to the model management server, and The above model management server is, A method for improving the artificial intelligence performance of an edge device, wherein a state-specific model of the edge device is constructed based on received metadata.
In paragraph 15, The above model management server is, A method for improving the artificial intelligence performance of an edge device, wherein, in constructing the above-mentioned state-specific model, metadata for a plurality of edge devices is referenced.

Description

System and method for improving artificial intelligence performance of edge devices The present invention relates to a system and method for improving the artificial intelligence performance of an edge device. As artificial intelligence (AI) technology becomes more widespread, there is an increasing number of cases where AI models are installed on multiple edge devices to perform computations independently. However, conventional technology adopts a method of deploying the same AI model in bulk without considering the unique environment in which individual edge devices are situated. Since each edge device is placed in an operating environment that changes from moment to moment, such as indoor or outdoor installation locations, lighting, temperature, and humidity, this uniform approach has the problem of hindering flexible response to environmental changes. Consequently, performance degradation of AI models occurred due to changes in various operating environments and usage patterns, and there were technical limitations in detecting this in real-time and responding appropriately. In other words, the absence of a system to dynamically generate and deploy customized AI models optimized for the unique characteristics of each edge device and changing operating environments made it difficult to ensure the reliability and stability of AI services. Meanwhile, the adoption of On-device AI technology is increasing to eliminate data processing latency caused by cloud servers and to ensure real-time responsiveness by directly deploying AI models on edge devices to perform computations. Since edge devices have limited computing resources—such as CPU, memory, and power—compared to central servers, lightweight AI models that reduce model size and optimize computational load are inevitably used. However, these lightweight models have clear performance limitations due to structural simplification, which leads to a decline in data processing accuracy. Consequently, in edge device environments, there was a problem of a high incidence of false negatives—where specific objects or situations are not detected—and false positives—where normal situations are incorrectly identified as abnormal. FIG. 1 is a diagram illustrating the configuration of an artificial intelligence performance improvement system for an edge device according to one embodiment of the present invention. FIG. 2 is a diagram illustrating the detailed configuration of an artificial intelligence performance improvement system for an edge device according to one embodiment of the present invention. FIG. 3 is a diagram illustrating the flow of a method for improving the artificial intelligence performance of an edge device according to an embodiment of the present invention. The objectives and effects of the present invention, and the technical configurations for achieving them, will become clear by referring to the embodiments described in detail below in conjunction with the accompanying drawings. In describing the present invention, if it is determined that a detailed description of known functions or configurations may unnecessarily obscure the essence of the invention, such detailed description will be omitted. However, this is not intended to limit the invention to specific embodiments, and it should be understood that it includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the invention. Furthermore, the terms described below are defined considering their functions in the present invention, and these may vary depending on the intentions or practices of the user or operator. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. Therefore, such definition should be based on the content throughout this specification. The terms used in this application are used merely to describe specific embodiments and are not intended to limit the invention. The singular expression includes the plural expression unless the context clearly indicates otherwise. In this application, terms such as "comprising" or "having" are intended to specify the presence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof. Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the present invention pertains. Terms such as those defined in commonly used dictionaries should be interp