KR20240063309A - Method of managing artificial intelligence model through artificial intelligence platform based on edge computing - Google Patents

Abstract

The present invention includes the needs information securing step (S110), the planning and design information transmission step (S120), the AI model creation step (S130), the AI model verification step (S140), the AI model execution step (S150), and the AI model prediction result data. Including the transmission step (S160) and the alarm information transmission step (S170), a transaction-related platform is established between the client and the AI model provider, custom-made and provided AI models that meet the client's needs, and edge computing are performed. We are launching a method of operating an AI model marketplace through an edge computing-based artificial intelligence platform that provides prediction results in real time and maintains security.

Description

How to operate an AI model marketplace through an edge computing-based artificial intelligence platform {METHOD OF MANAGING ARTIFICIAL INTELLIGENCE MODEL THROUGH ARTIFICIAL INTELLIGENCE PLATFORM BASED ON EDGE COMPUTING}

The present invention builds a transaction-related platform between a client and an AI model provider, custom-made and provides an AI model that meets the client's needs, provides prediction results in real time through edge computing, and maintains security. This is about how to operate an AI model marketplace through an edge computing-based artificial intelligence platform.

Recently, edge computing technology, which transmits data using edge servers, is becoming active.

That is, edge computing technology may include MEC (Multi-access Edge Computing) or fog computing (FOC), and may be installed in a location geographically close to the electronic device, for example, inside or near a base station. It refers to a technology that provides data to electronic devices through an edge data network or MEC server.

For example, among at least one application installed on an electronic device, an application that requires low latency may transmit and receive data through an edge server installed in a geographically close location, rather than through a server located in an external data network. .

By utilizing this edge computing technology, we aim to build a transaction-related platform between clients and AI model providers to provide technology that can custom-make and provide AI models that meet the needs of clients.

Korean Patent Publication No. 10-2022-0037617 (Video processing method and device using edge computing service, 2022.03.05) Korean Patent Publication No. 10-2022-0074335 (Apparatus and method for counting objects by type using artificial intelligence-based edge computing, 2022.06.03)

The technical task to be achieved by the idea of the present invention is to build an AI model transaction-related subscription platform between the client and the AI model provider, custom-create and provide an AI model that meets the environment and client needs through a survey, and provide edge The goal is to provide a method of operating an AI model marketplace through an edge computing-based artificial intelligence platform that can provide prediction results in real time and maintain security through edge computing on devices.

In order to achieve the above-described object, an embodiment of the present invention includes the steps of performing a survey to secure needs information including the client's requirements and needs from a client terminal to a marketplace platform; Analyzing the needs information in the marketplace platform, performing planning and design of an AI model to solve the needs, and transmitting the needs information and planning design information to an AI model provider terminal; The AI model provider terminal builds a learning data set based on the planning and design information, creates an AI model that performs formalization, text identification, image recognition, video recognition, object detection, prediction, maintenance, and recommendation, Transferring to the marketplace platform; Evaluating and verifying the AI model according to testing standards preset in the marketplace platform; According to the AI model subscription and purchase contract information, an edge device that interfaces with the client terminal, has a sensor, and performs edge computing of the AI model, including an edge analysis module and an edge IoT module, receives the test criteria from the marketplace platform. Downloading and executing the passed AI model; Transmitting prediction result data of the AI model executed on the edge device to the marketplace platform; And providing a method of operating an AI model marketplace through an edge computing-based artificial intelligence platform, including the step of the marketplace platform analyzing the prediction result data and transmitting statistical data and alarm information to the client terminal.

Here, the step of the marketplace platform feeding back consulting information to the client terminal according to the results of the survey may be further included.

In addition, the marketplace platform performs evaluation and verification of the AI model or analysis of prediction result data of the AI model and feeds it back to the AI model provider terminal to perform version management, error correction, and updating of the AI model. It may further include ;.

In addition, the planning and design information may include the background and necessity of the need, the goal, promotion direction and improvement direction of the AI model, the implementation plan of the need, statistical data related to the need, required functions and expected effects.

In addition, it may further include the step of the marketplace platform analyzing subscriber information of the client terminal and matching and recommending an AI model suitable for the environment or situation in which the edge device is deployed.

Additionally, the marketplace platform can encrypt and store the AI model subscription and purchase contract information and the subscriber information based on blockchain.

Additionally, the learning data set may be data collected by the sensor of the edge device and provided to the marketplace platform.

In addition, the AI model provider terminal can construct the learning data set by extracting and processing actual data related to the planning and design information from a cloud server where big data is collected and stored, and labeling the data.

In addition, the AI model provider terminal, through the GAN model, generates similar fictional data that is likely to actually be more than a certain probability from real data associated with the planning and design information extracted from the cloud server, and combines the real data and the fictional data. The learning dataset can be constructed by refining and labeling the data.

In addition, when purchasing the AI model according to the AI model subscription and purchase contract information, the marketplace platform grants ownership to the AI model by matching NFT in a distributed environment using a blockchain network, and the client terminal You can rent, donate, exchange, or resell the AI model whose ownership is granted by the NFT.

In addition, when reselling the AI model, the client terminal and other client terminals corresponding to the resale target are interconnected through a blockchain network to distribute and share transaction history information about the AI model, and the marketplace platform Upon receiving transaction request information for the AI model from the client terminal to the other client terminal, it is determined whether each hash value of the transaction history information for the corresponding AI model matches each other, and the transaction is sent from the marketplace platform to the other client terminal. The AI model and the copyright for resale are transmitted to enable mutual resale transactions between client terminals for the AI model, and the transaction history information may include information on the limit on the number of times the AI model can be resold or the resale validity period.

In addition, the AI model is used for autonomous driving, flight of industrial and military drones, smart factories, smart production lines, smart farms, smart cities, industrial robots, industrial Internet of Things, manufacturing, franchise, distribution, logistics, public safety, security, and mobility. , can be an AI model that performs stereotyping, text identification, image recognition, video recognition, object detection, prediction, maintenance, and recommendation for medical, healthcare, education, or advertising.

In addition, the AI model is downloaded and executed on the first to Nth edge devices, respectively, and the AI model is generated by averaging the weights, bias values, and gradient values transmitted from the first to Nth edge devices to the marketplace platform, respectively. It may be a joint model that updates .

According to the present invention, an AI model transaction-related subscription platform is established between the client and the AI model provider, custom-made and provided AI models that meet the environment and client needs are provided through surveys, and edge computing on edge devices is provided. This has the effect of providing prediction results in real time and maintaining security.

Figure 1 shows a schematic flowchart of a method of operating an AI model marketplace through an edge computing-based artificial intelligence platform according to an embodiment of the present invention.
Figure 2 illustrates the detailed flow chart of Figure 1.
Figure 3 illustrates an implementation diagram of the AI model marketplace operation method through the edge computing-based artificial intelligence platform of Figure 1.
Figure 4 illustrates the resale configuration of the AI model marketplace operation method through the edge computing-based artificial intelligence platform of Figure 1.

Hereinafter, embodiments of the present invention having the above-described features will be described in more detail with reference to the attached drawings.

The method of operating an AI model marketplace through an edge computing-based artificial intelligence platform according to an embodiment of the present invention includes a needs information securing step (S110), a planning and design information transmission step (S120), an AI model creation step (S130), and an AI model. A transaction-related platform is established between the client and the AI model provider, including the verification step (S140), the AI model execution step (S150), the AI model prediction result data transmission step (S160), and the alarm information transmission step (S170). , The goal is to customize and provide AI models that meet the client's needs, provide prediction results in real time through edge computing, and maintain security.

Hereinafter, with reference to FIGS. 1 to 4, the method of operating the AI model marketplace through the edge computing-based artificial intelligence platform of the above-described configuration will be described in detail as follows.

First, in the needs information securing step (S110), a survey sheet is sent from the marketplace platform 120 to the client terminal 110 belonging to the user who wants to use the AI model to conduct a survey on requirements and necessities. By performing it repeatedly or continuously, needs information including the client's requirements and needs is received and secured from the client terminal 110 to the marketplace platform 120.

Thereafter, in the planning and design information transmission step (S120), the marketplace platform 120 analyzes the needs information, performs planning and design of an AI model to solve the needs, and sends the needs information and information to the AI model provider terminal 130. Send planning and design information.

Here, the planning and design information may include the background and necessity of the need, the goal, promotion direction and improvement direction of the AI model, the action plan of the need, statistical data related to the need, required functions and expected effects.

Meanwhile, the marketplace platform 120 may analyze the survey response results and feed back consulting information to the client terminal 110 to derive detailed response results to create an AI model.

In addition, AI models that reflect needs include autonomous driving, flight of industrial and military drones, smart factory, smart production line, smart farm, smart city, industrial robots, industrial Internet of Things, manufacturing, franchise, distribution, logistics, security, etc. It can be an AI model that performs stereotyping, text identification, image recognition, video recognition, object detection, prediction, maintenance, and recommendation for security, mobility, medicine, healthcare, education, or advertising.

Additionally, planning design information may include data design, process design, UI menu, and function design.

Afterwards, in the AI model creation step (S130), the AI model provider terminal 130 selects an AI model that satisfies the needs using planning and design information as a guideline, and directly builds a learning dataset for learning the AI model. Alternatively, an AI model that predicts results targeting needs such as canonicalization, text identification, image recognition, image recognition, object detection, prediction, maintenance, and recommendation using the learning data set provided by the marketplace platform 120. is created and transmitted to the marketplace platform 120.

Here, the learning data set for creating the AI model may be actual data collected by the sensor of the edge device 140 and provided to the marketplace platform 120, and the AI model provider terminal 130 collects big data. You can construct a learning data set by extracting raw data of actual data related to planning and design information from a stored cloud server, processing the raw data into data necessary for creating an AI model, and labeling the data.

Alternatively, since the amount of actual data provided from the sensor of the edge device 140 is relatively small and the prediction performance of the AI model is low, the AI model provider terminal 130 receives information from the cloud server through a GAN (Generative Adversarial Network) model. The predictive performance of the AI model can be improved by generating similar fictional data that is more likely to exist in reality than a certain probability from real data related to the extracted planning and design information, refining the real data and fictional data, and constructing a learning data set by labeling the data. there is.

Meanwhile, the AI model is downloaded and executed on the first to Nth edge devices 140, respectively, and the weight, bias value, and gradient value transmitted from the first to Nth edge devices 140 to the marketplace platform 120, respectively. It may be a joint model that updates the AI model with the average value.

Thereafter, in the AI model verification step (S140), the previously created AI model can be evaluated and verified according to the test criteria preset in the marketplace platform 120 to determine whether needs are met and whether planning and design information is complied with. there is.

Thereafter, in the AI model execution step (S150), according to the AI model subscription and purchase contract information concluded between the client terminal 110 and the marketplace platform 120, it is linked with the client terminal 110 and is equipped with a sensor and edges. The edge device 140, which performs edge computing of the AI model, including the analysis module and the edge IoT module, and predicts the results, receives the AI model that passed the previous test standards from the marketplace platform 120 directly or through the client terminal 110. ) and run it.

Here, the marketplace platform 120 encrypts and stores AI model subscription and purchase contract information and subscriber information based on blockchain to prevent information leakage due to cyber risks such as hacking.

Additionally, when installing or executing an AI model on the client terminal 110, it can be installed or executed after receiving a license consisting of a string containing unique numbers and letters transmitted from the marketplace platform 120.

In addition, the edge analysis module and edge IoT module configure the edge agent, the edge analysis module runs an AI model, the edge IoT module can manage measured values detected from sensors, and edge management manages the edge device (140 ) can be managed.

Thereafter, in the AI model prediction result data transmission step (S160), the prediction result data of the AI model executed on the edge device 140 is transmitted to the marketplace platform 120.

Thereafter, in the alarm information transmission step (S170), the marketplace platform 120 analyzes the prediction result data, generates statistical data and alarm information, and transmits them to the client terminal 110, so that the prediction result exceeds the preset prediction value. You can react and deal with it quickly. Here, the statistical data may be the number of visitors to the unmanned store, the return visit rate, etc., and the alarm information may be a warning about abnormal behavior of visitors to the unmanned store.

Meanwhile, the marketplace platform 120 performs evaluation and verification of the AI model or analysis of the prediction result data of the AI model, and feeds it back to the AI model provider terminal 130 to perform version management, error correction, and update of the AI model. By further including the step (S180), management of the AI model may be performed periodically.

Here, version management can use the Git-based version management tool provided by the marketplace platform 120.

In addition, the marketplace platform 120 stores the subscriber information of the client terminal 110, including field of employment, business field, store location, sales, occupation, hobbies, interests, age, gender, physical information, activity radius, AI model application field, It further includes a step (S190) of analyzing goals, inconveniences, etc., and matching and recommending an AI model suitable for the environment or situation in which the edge device 140 is deployed from a DB storing a plurality of pre-built AI models, An AI model whose predictive performance has already been verified by being applied to the same or similar environment or situation can be provided by immediately transmitting it to the edge device 140 without a separate creation step.

Meanwhile, when purchasing an AI model according to AI model subscription and purchase contract information, the marketplace platform 120 grants ownership to the corresponding AI model by matching an NFT (Non-fungible token) in a distributed environment using a blockchain network. In addition, the client terminal 110 may activate the social circulation of the AI model by renting, donating, exchanging, or reselling the AI model whose ownership is granted by NFT to a third party.

For example, referring to FIG. 4, when reselling an AI model, the client terminal 110 and other client terminals 150 corresponding to the resale object are interconnected through a blockchain network to distribute transaction history information about the AI model. When sharing, the marketplace platform 120 receives transaction request information for an AI model from the client terminal 110 to another client terminal 150, check whether each hash value of the transaction history information for the corresponding AI model matches each other. By making the decision, the AI model and the copyright or license for resale can be transmitted from the marketplace platform 120 to the other client terminal 150 so that a mutual resale transaction can be performed between the client terminals 110 for the AI model.

Here, the transaction history information includes restrictions on the number of times the AI model can be resold or information on the resale expiration date, which can limit indiscriminate resale and determine the resale price by applying depreciation at a specific rate of the AI model according to the resale expiration date. You can also set it.

Additionally, subscription fees, purchase fees, rental fees, and resale fees based on AI model subscription and purchase contract information may be traded through virtual currency.

Meanwhile, another embodiment of the present invention provides a computer program recorded on a recording medium readable by a computer executing the method of operating an AI model marketplace through the above-described edge computing-based artificial intelligence platform.

Therefore, by operating the AI model marketplace through the edge computing-based artificial intelligence platform as described above, a subscription platform related to AI model transactions is established between the client and the AI model provider to respond to the environment and client needs through surveys. A matching AI model can be customized and provided, and prediction results can be provided in real time and security maintained through edge computing on edge devices.

The embodiments described in this specification and the configuration shown in the drawings are only one of the most preferred embodiments of the present invention and do not represent the entire technical idea of the present invention, so various equivalents may be substituted for them at the time of filing the present application. It should be understood that variations and variations may exist.

110: client terminal
120: Marketplace platform
130: AI model provider terminal
140: Edge device
150: Other client terminal
S110: Step to secure needs information
S120: Planning and design information transmission stage
S130: AI model creation step
S140: AI model verification step
S150: AI model execution stage
S160: AI model prediction result data transmission step
S170: Alarm information transmission step

Claims (13)

performing a survey to secure needs information including the client's requirements and needs from the client terminal to the marketplace platform;
Analyzing the needs information in the marketplace platform, performing planning and design of an AI model to solve the needs, and transmitting the needs information and planning design information to an AI model provider terminal;
The AI model provider terminal builds a learning data set based on the planning and design information, creates an AI model that performs formalization, text identification, image recognition, video recognition, object detection, prediction, maintenance, and recommendation, Transferring to the marketplace platform;
Evaluating and verifying the AI model according to testing standards preset in the marketplace platform;
According to the AI model subscription and purchase contract information, an edge device that interfaces with the client terminal, has a sensor, and performs edge computing of the AI model, including an edge analysis module and an edge IoT module, receives the test criteria from the marketplace platform. Downloading and executing the passed AI model;
Transmitting prediction result data of the AI model executed on the edge device to the marketplace platform; and
Including, the marketplace platform analyzing the prediction result data and transmitting statistical data and alarm information to the client terminal.
How to operate an AI model marketplace through an edge computing-based artificial intelligence platform.
According to paragraph 1,
Characterized in that it further comprises a step wherein the marketplace platform feeds back consulting information to the client terminal according to the results of the survey.
How to operate an AI model marketplace through an edge computing-based artificial intelligence platform.
According to paragraph 1,
The marketplace platform performs evaluation and verification of the AI model or analysis of prediction result data of the AI model and feeds it back to the AI model provider terminal to perform version management, error correction, and update of the AI model; Characterized by further comprising:
How to operate an AI model marketplace through an edge computing-based artificial intelligence platform.
According to paragraph 1,
The planning and design information includes the background and necessity of the need, the goal, promotion direction and improvement direction of the AI model, the implementation plan of the need, statistical data related to the need, required functions and expected effects,
How to operate an AI model marketplace through an edge computing-based artificial intelligence platform.
According to paragraph 1,
Characterized in that it further comprises a step of the marketplace platform analyzing subscriber information of the client terminal and matching and recommending an AI model suitable for the environment or situation in which the edge device is deployed.
How to operate an AI model marketplace through an edge computing-based artificial intelligence platform.
According to clause 6,
The marketplace platform is characterized in that the AI model subscription and purchase contract information and the subscriber information are encrypted and stored based on blockchain,
How to operate an AI model marketplace through an edge computing-based artificial intelligence platform.
According to paragraph 1,
The learning data set is characterized in that it is data collected by the sensor of the edge device and provided to the marketplace platform.
How to operate an AI model marketplace through an edge computing-based artificial intelligence platform.
According to paragraph 1,
The AI model provider terminal extracts and processes actual data related to the planning and design information from a cloud server where big data is collected and stored, and constructs the learning data set by labeling the data.
How to operate an AI model marketplace through an edge computing-based artificial intelligence platform.
According to clause 8,
The AI model provider terminal generates similar fictional data that is likely to actually exist above a certain probability from real data associated with the planning and design information extracted from the cloud server through the GAN model, and refines and processes the real data and the fictional data. Characterized by constructing the learning dataset by labeling the data,
How to operate an AI model marketplace through an edge computing-based artificial intelligence platform.
According to paragraph 1,
When purchasing the AI model according to the AI model subscription and purchase contract information, the marketplace platform grants ownership to the AI model by matching NFT in a distributed environment using a blockchain network,
The client terminal is characterized in that it performs rental, donation, exchange, or resale of the corresponding AI model whose ownership is granted by the NFT.
How to operate an AI model marketplace through an edge computing-based artificial intelligence platform.
According to clause 10,
When reselling the AI model,
The client terminal and other client terminals subject to resale are interconnected through a blockchain network to distribute and share transaction history information about the AI model,
When the marketplace platform receives transaction request information for the AI model from the client terminal to the other client terminal, it determines whether each hash value of the transaction history information for the corresponding AI model matches each other, and the marketplace platform By transmitting the AI model and the copyright for resale from the other client terminal, a mutual resale transaction is performed between the client terminals for the AI model,
The transaction history information is characterized in that it includes information on the limit on the number of times the AI model can be resold or the resale validity period.
How to operate an AI model marketplace through an edge computing-based artificial intelligence platform.
According to paragraph 1,
The AI model is used in autonomous driving, industrial and military drone flight, smart factory, smart production line, smart farm, smart city, industrial robots, industrial Internet of Things, manufacturing, franchise, distribution, logistics, public safety, security, mobility, and medical care. , Characterized as an AI model that performs stereotyping, text identification, image recognition, image recognition, object detection, prediction, maintenance, and recommendation for healthcare, education, or advertising.
How to operate an AI model marketplace through an edge computing-based artificial intelligence platform.
According to paragraph 1,
The AI model is downloaded and executed on the first to Nth edge devices, respectively,
Characterized in that it is a joint model that updates the AI model with the average value of the weight, bias, and gradient values transmitted from the first to Nth edge devices, respectively, to the marketplace platform.
How to operate an AI model marketplace through an edge computing-based artificial intelligence platform.

Images