KR102284440B1 - Method to broker deep learning model transactions perfomed by deep learning model transaction brokerage servers - Google Patents

Abstract

According to an embodiment of the present invention, a method for brokering a deep learning model transaction includes the steps of: collecting purchase information of a purchaser about a deep learning model required by the purchaser from a purchaser terminal; searching for sales information of at least one seller terminal which sells the deep learning model corresponding to the purchase information of the purchaser, and transmitting the sale information of the deep learning model corresponding to the request information of the purchaser to the purchaser terminal; and receiving a model selection signal of the purchaser from the purchaser terminal, and transmitting the model selection signal to the seller terminal corresponding to a purchaser's model selection signal. Therefore, even unskilled people related to deep learning can obtain results they want to predict or analyze through trading of the deep learning models.

Description

How to broker a deep learning model transaction performed by a deep learning model transaction brokerage server

The present invention relates to a method of brokering a deep learning model transaction performed by a deep learning model transaction brokerage server. In more detail, collecting the purchase information of the buyer about the deep learning model requested by the buyer from the buyer terminal, and searching for sales information of at least one seller terminal that sells the deep learning model corresponding to the purchase information of the buyer and transmitting the sales information of the deep learning model corresponding to the buyer's request information to the buyer terminal, and receiving the buyer's model selection signal from the buyer's terminal, and a model selection signal to the seller's terminal corresponding to the buyer's model selection signal It relates to a deep learning model transaction brokerage method comprising the step of sending

Recently, with the rapid development of processors and memories, artificial intelligence (AI) technology has been developed and in the spotlight, and services using artificial intelligence technology have been widely developed in various fields.

In particular, deep learning technology is attracting attention in the field of artificial intelligence technology, and it is showing excellent results in various fields such as data analysis, image recognition, and natural language processing. Deep learning is a field of machine learning, and it is a technique to express data in a form that a computer can process, such as a vector or graph, and build a model that learns it. That is, for a specific learning goal, such as recognizing a face or expression, deep learning focuses on building better representation methods and efficient models for learning, and many of the representation methods in deep learning are inspired by neuroscience. It is based on information processing and communication patterns of the nervous system.

In the early days, the development of artificial intelligence algorithms was conducted somewhat closedly by individual capabilities based on R language and Python language, and the evaluation of performance has been evaluated and verified by individual standards. Therefore, there was a limitation in that artificial intelligence had to be continuously managed and controlled by humans in order to maintain the expected performance and operate stably. In addition, the existing artificial intelligence algorithm has been developed in a somewhat personalized way that analysts apply after design and development verification in a personal development environment in order to meet the unit requirements, understanding and improvement of performance that changes according to the environment, etc. There were also difficulties in proceeding with the process.

Recently, the development of algorithms of open source-based deep learning techniques has increased, and a large number of deep learning algorithms are being developed. For this purpose, it is very difficult to confirm the suitability of the deep learning learning results.

Accordingly, by trading various pre-trained deep learning models, there is a need to develop a technology for a deep learning transaction brokerage method to enable analysis or prediction of data without direct learning.

Republic of Korea Patent Publication No. 10-2018-0066688 (Announcement date: 2018.06.19)

The present invention is to solve the problems described above, and to provide a deep learning model transaction brokerage method in which even non-skilled people related to deep learning can obtain the results that they want to predict or analyze through the transaction of the deep learning model. There is this.

As an embodiment of the present invention, a method of brokering a deep learning model transaction performed by a deep learning model transaction brokerage server may be provided.

The deep learning model transaction brokerage method according to an embodiment of the present invention includes the steps of collecting the purchase information of the purchaser regarding the deep learning model required by the purchaser from the purchaser terminal, and selling the deep learning model corresponding to the purchase information of the purchaser retrieving the sales information of at least one seller terminal, transmitting the sales information of the deep learning model corresponding to the buyer's request information to the buyer terminal, and receiving the model selection signal of the buyer from the buyer terminal, and the model of the buyer It may include transmitting a model selection signal to the seller terminal corresponding to the selection signal.

According to the method of brokering a deep learning model transaction performed by the deep learning model transaction brokerage server of the present invention, even an unskilled person related to deep learning can obtain the desired result to predict or analyze through the transaction of the deep learning model. there is.

1 is a flowchart of a method of brokering a deep learning model transaction performed by a deep learning model transaction brokerage server according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art can easily implement them. However, the present invention may be embodied in many different forms and is not limited to the embodiments described herein. And in order to clearly explain the present invention in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Terms used in this specification will be briefly described, and the present invention will be described in detail.

The terms used in the present invention have been selected as currently widely used general terms as possible while considering the functions in the present invention, but these may vary depending on the intention or precedent of a person skilled in the art, the emergence of new technology, and the like. In addition, in a specific case, there is a term arbitrarily selected by the applicant, and in this case, the meaning will be described in detail in the description of the corresponding invention. Therefore, the term used in the present invention should be defined based on the meaning of the term and the overall content of the present invention, rather than the name of a simple term.

When a part "includes" a certain element throughout the specification, this means that other elements may be further included, rather than excluding other elements, unless otherwise stated. In addition, terms such as "...unit" and "module" described in the specification mean a unit that processes at least one function or operation, which may be implemented as hardware or software, or a combination of hardware and software. . In addition, when a part is "connected" to another part throughout the specification, it includes not only a case in which it is "directly connected" but also a case in which it is connected "with another element in the middle".

1 is a flowchart of a method of brokering a deep learning model transaction performed by a deep learning model transaction brokerage server according to an embodiment of the present invention. Hereinafter, the present invention will be described in detail with reference to the above-described drawings.

As an embodiment of the present invention, a method of brokering a deep learning model transaction performed by a deep learning model transaction brokerage server may be provided. That is, each step (S100-S300) of the method of brokering a deep learning model transaction of the present invention may be performed by a deep learning model transaction brokerage server.

The deep learning model transaction brokerage method according to an embodiment of the present invention includes the steps of collecting the purchase information of the purchaser regarding the deep learning model required by the purchaser from the purchaser terminal (S100), deep learning corresponding to the purchase information of the purchaser Searching for sales information of at least one seller terminal selling a model, and transmitting sales information of a deep learning model corresponding to the buyer's request information to the buyer terminal (S200) and a model selection signal of the buyer from the buyer terminal and transmitting the model selection signal to the seller terminal corresponding to the purchaser's model selection signal (S300).

Hereinafter, each process (S100 - S300) of the deep learning model transaction brokerage method according to an embodiment of the present invention will be described.

First, the step (S100) of collecting the purchase information of the purchaser regarding the deep learning model required by the purchaser from the purchaser terminal may be performed.

The step S100 according to one disclosure may further include a step (S110) of collecting sales information from the seller terminal for the deep learning model in the deep learning model transaction brokerage server.

In the present specification, the deep learning model may be a model obtained by a seller (seller terminal) using a deep learning solution platform. That is, the deep learning model transaction brokerage method according to an embodiment of the present invention may be performed in conjunction with a deep learning solution platform.

The process (S10-S50) in which the deep learning model is obtained through the deep learning solution platform is as follows.

First, in the deep learning solution platform, it is possible to collect input information including tabular data to be analyzed or predicted from the seller terminal, the learning method of the deep learning model, and the learning form of the deep learning model (S10).

Input information according to one disclosure may include table data that is a target of analysis or prediction using a deep learning model. The table data may be data based on a tabular format including at least one or more cells generated by rows and columns. The table data may be in the form of spreadsheet data, CSV (comma-separated value), Excel, HTML (HyperTextMarkup Language), XML (Extensible markup language), PDF (Portable Document Format), etc. there is.

Table data according to one disclosure is data of a column to be analyzed or predicted (hereinafter, an analysis target column) and data of at least one column (hereinafter, a related column) expected to be related to the data of the analysis target column may include. Hereinafter, the column of the table data may include the analysis target column and at least one or more related columns. The analysis target column may be determined by the analysis target column determination signal collected from the seller terminal in the deep learning solution platform, the analysis target column is determined from the table data by the analysis target column determination signal, and a table excluding the analysis target column The remaining columns of data may be determined as related columns.

Input information according to one disclosure may include information about a learning method of a deep learning model. The learning method may include information for learning to proceed according to a learning method preferred by the seller based on information on learning speed, information on accuracy of learning, information on error rate of learning, and the like.

The input information according to one disclosure may include information about a learning form of the deep learning model. The learning form may include learning form information such as binary classification, multi-class classification, regression, prediction, segmentation, and detection. That is, the learning form information may be input and collected by the user so that the learning result appears according to the learning form preferred by the seller.

According to one disclosure, the input information may further include other types of data as well as structured data such as the aforementioned table data. For example, the input information may include time series data, image data, natural language data, voice data, etc., and may generate a deep learning model by receiving various data such as the above from a deep learning solution platform.

Next, the step (S20) of collecting pre-processing information including whether the table data is pre-processed for each column, whether it is used for learning the deep learning model, and the pre-processing method from the seller terminal in the deep learning solution platform may be performed.

The pre-processing information according to the disclosure may include information on whether pre-processing is input to the deep learning solution platform through the seller terminal with respect to whether pre-processing is performed for each column. Since table data of input information is configured in a table form, a column indicating an attribute of data for one column may be included. That is, the information on whether to perform the pre-processing may indicate information on whether to perform the pre-processing for each column of table data. In other words, in S20, the pre-processing may be performed by separately selecting only the columns that the seller wants to pre-process from among the columns of the input table data. Through the above process, it is possible to select the presence or absence of pretreatment according to the column differently, so that the learning result desired by the seller can be derived.

The preprocessing information according to the disclosure may include information on whether or not to be used for learning the deep learning model for each column. That is, the column used for training of the deep learning model by the seller in the deep learning solution platform can be selected. In other words, a learning result desired by the seller may be derived by excluding the column necessary for learning from the table data.

The pre-processing information according to the disclosure may include information about a pre-processing method. The pre-processing method may include data cleansing, normalization, de-identification, and data replacement. The data cleansing may refer to at least one data operation such as data deduplication, error removal, and invalidation removal. For example, the data cleansing may refer to a data operation of deleting rows having a predetermined value in which the number of data is less than a predetermined number, or deleting rows outside a predetermined ratio of the overall standard distribution. The de-identification may be a data operation of deleting a column in which a predetermined data value occupies a predetermined ratio or more of the total number of data. The data replacement may refer to a data operation of deleting rows having empty values or filling empty values with a predetermined value. The seller may choose to perform the pre-processing in a manner selected by the seller from among the pre-processing methods including data cleansing, normalization, de-identification, and data replacement in the deep learning solution platform.

In S20 according to an embodiment of the present invention, in the deep learning solution platform, the seller may determine whether to perform pre-processing for each column of table data and determine whether to use it for learning for each column. In addition, the pretreatment method may be selected by the vendor. That is, in the deep learning solution platform, by specifying only the columns used for learning among the table data input by the seller, the deep learning model can be trained so that the learning result desired by the seller can be derived. In addition, by performing only preprocessing on a specific column, the deep learning model can be trained in the same direction as the seller wants.

Next, a plurality of deep learning models prepared in advance in the deep learning solution platform are trained based on the collected input information and preprocessing information, and the learning state, accuracy, and error rate for each of the plurality of deep learning models according to the learned result and providing learning result information including the degree of similarity to the seller terminal (S30) may be performed.

S30 according to an embodiment is a step (S31) in which learning is performed based on the collected input information and preprocessing information for a plurality of pre-prepared deep learning models and a plurality of deep learning models according to the result learned in S31. It may include a step (S32) of providing learning result information including the learning state, accuracy, error rate, and similarity for each of the subjects.

In S31 according to an exemplary embodiment, the plurality of pre-prepared deep learning models may include models in which learning has been performed in advance based on a pre-prepared database. The plurality of deep learning models may include deep learning models having different structures in which at least one or more neural networks are combined. For example, the deep learning model may include a multi-layer perceptron, a recurrent neural network, a convolutional neural network, a generative adversarial network, etc., as well as a combination of the neural networks or It may have been modified and created. That is, in S31 , learning may proceed according to the input information and pre-processing information for the plurality of deep learning models that have been trained and prepared in advance.

In S32 according to an embodiment, the learning result information may include a learning state, accuracy, error rate, and similarity for each of a plurality of deep learning models. The learning state may indicate a learning result for indicating whether learning is completed or learning has failed according to a plurality of deep learning models. The accuracy may indicate the accuracy of a deep learning model that has been trained. The error rate may indicate a rate of errors generated during training of the deep learning model. The similarity (Dice) may represent a sample coefficient for measuring the similarity between the training data and the result data of the deep learning model. The seller may select a deep learning model to be used by the seller for data analysis or prediction of the analysis target column by using the learning result information including the learning state, accuracy, error rate, and similarity.

That is, after step S32, the step (S40) of collecting seller selection information for the deep learning model that the seller will use for analysis or prediction of the data of the analysis target column from the seller terminal in the deep learning solution platform may be performed. The deep learning model in which the seller selection information is collected may be referred to as a selection model hereinafter.

The learning result information according to the start may further include detailed learning information. The detailed learning information may include importance information, detailed analysis information, loss function information, and other information. The precise analysis information may include statistical information on data for each column. The loss function information may include information about a loss function used for training a deep learning model. The other information may include statistical information used to measure the reliability of the deep learning model.

The importance information according to one disclosure may be information for indicating a degree of association between an analysis target column and at least one related column according to a result learned for each deep learning model. That is, the seller can check the most important related column for analyzing or predicting the data of the analysis target column through the importance information. The importance information may be displayed by quantifying the degree of association for each association column. Alternatively, the importance information may be displayed by visualizing (eg bar graphing) the degree of association for each association column. That is, in the deep learning solution platform, the importance information may be displayed in various ways for the seller to determine the correlation between the analysis target column and the related column. In other words, in the deep learning solution platform, the seller can check the most important (most relevant) related column for analysis or prediction of the analysis target column for each deep learning model that has been trained according to the above-described process through the importance information.

According to one disclosure, after the seller selection information is collected from the seller terminal in the above-described S40 in the deep learning solution platform, an analysis request signal may be collected from the seller terminal in the deep learning solution platform, and in the collected analysis request signal Correspondingly, the prescriptive analysis result information may be provided from the deep learning solution platform to the seller terminal (S50). That is, the analysis request signal may be a signal from a seller terminal requesting a prescriptive analysis result using the selection model. Here, prescriptive analytics is one of the business analysis methods that use information to identify phenomena based on given data, predict future events, and determine appropriate actions. It can be an analysis method that presents the best alternative by efficiently allocating limited resources with ideal results. In the deep learning solution platform, prescriptive analysis result information can be generated (S51) by deriving an optimal decision using the learning result of the deep learning model. The selection model in which the prescription analysis result information is generated may be referred to as a prescription analysis model.

According to one disclosure, the seller may sell the prescription analysis model in the deep learning model transaction brokerage server after the prescription analysis result information is generated according to the processes of S51-1 to S51-3 below. That is, only sales information of the prescription analysis model in which the prescription analysis result information is generated according to the following processes S51-1 to S51-3 may be collected from the deep learning model transaction brokerage server. That is, the sales information according to the disclosure may include information on whether or not the prescriptive analysis result information is generated.

Hereinafter, the process of generating the prescriptive analysis result information (S51-1 to S51-3) will be described in detail. Also, in the following description, n may be a natural number.

First, a final related column among at least one or more related columns may be generated ( S51 - 1 ). The final association column may be determined according to a difference in the degree of association. More specifically, if the value of the nth correlation difference satisfies Equation 1 below in a relationship from the first correlation difference to the maximum value among the n-1 th correlation difference values (n≥3), among at least one or more association columns From the first association column to the nth association column may be determined as the final association column.

[Equation 1]

Here, d _n is the nth correlation difference, d _x is the maximum value (n≧3) among values of the n−1th correlation difference from the value of the first correlation difference, and r _f may be a reference ratio.

Here, the reference ratio may be a predetermined value. In addition, the nth association column is the nth association column when the degree of association with the analysis target column is listed in descending order (in descending order) from the greatest value (the value of the degree of association of the first association column) according to the importance information. (from the highest correlation column to the nth highest correlation column) may be referred to. That is, the first association column may be an association column in which the degree of association with the learning result analysis target column of the deep learning model is greater than that of the remaining associated columns except for the first association column. The nth correlation difference may be a value obtained by dividing an absolute value of a difference between the degree of association of the nth association column and the degree of association of the n+1th association column by the degree of association of the n+1th association column as shown in Equation 2 below.

[Equation 2]

Here, v _n may be an nth degree of association.

For reference, the final related column was composed of at least three columns to define the relationship with the analyte column (n≥3).

After the final related column is determined as in step S51-1, an approximate expression for analysis or prediction of the data of the column to be analyzed may be generated as shown in Equation 3 below (S51-2).

[Equation 3]

Here, J is the data of the column to be analyzed, p is the number of the last related column, d ^f _k is the degree of correlation of the last related column, and a _k may be the final correlation coefficient. (k=1, 2, ..., p)

After the approximate expression is generated as in S51-2, the final association coefficient can be derived by substituting it into the approximate expression based on the relationship between the data of the final association column and the data of the analysis target column. That is, the input information includes table data, and since the table data includes the data of the last related column and the data of the analysis target column, the final related column data and the analysis target column data are substituted into the approximate expression to obtain the final result. The approximation can be completed by deriving the correlation coefficient.

Next, prescriptive analysis result information may be provided (S51-3) based on the data value of the analysis target column derived through the approximation equation for arbitrary data for the final related column.

In the deep learning solution platform according to the disclosure, the prescriptive analysis result information prepared based on the approximation formula as in the process of S51-1 to S51-3 described above is provided, so that the information most relevant to the data to be analyzed or predicted A solution consisting only of can be presented to the seller. That is, the seller may use the importance information to determine the data related to the data that the seller directly analyzes or predicts, but uses the prescriptive analysis result information according to the process of S51-1 to S51-3 described above. This can be used as a more efficient problem solution by providing a relationship between the data to be analyzed or predicted and the data that is actually related. In other words, if the prescriptive analysis result information according to the above-described processes S51-1 to S51-3 is used, errors that may occur when the seller determines the importance information according to the deep learning learning result can be reduced.

Next, searching for sales information of at least one seller terminal that sells the deep learning model corresponding to the purchase information of the buyer, and transmitting the sale information of the deep learning model corresponding to the request information of the buyer to the buyer terminal (S200) ) can be performed.

The sales information according to the start of work may further include a sales amount.

The sales amount according to the start of work may include a model use amount and a brokerage fee.

The model usage amount according to the start may be determined from the seller setting information collected from the seller terminal. The seller setting information may include desired selling price information of the prescription analysis model determined by the seller.

The brokerage fee according to the initiation may be determined as in Equation 4 below based on the seller learning index.

[Equation 4]

where Wcom is the brokerage fee, f is the reference cost, Q _sell is the seller learning index, q _p is the platform learning factor, N _acc is the seller’s cumulative learning times, and N _tot is the seller’s learnable number of times there is.

The reference cost may be predetermined by the deep learning model transaction brokerage server. The platform learning factor indexes the degree to which the learning of the deep learning model is not completed when the seller terminal learns the deep learning model using the deep learning solution platform during the reference period, and may be predetermined by the deep learning solution platform.

The cumulative number of learning may represent the number of times the seller terminal trained the deep learning model using the deep learning solution platform during the reference period. The learnable number represents the number of times that the seller terminal can learn the deep learning model using the deep learning solution platform during the reference period, and the learnable number may be determined by the seller grade. The reference period may be predetermined in the deep learning model transaction brokerage server. The seller grade is the number of times the seller terminal sold the prescription analysis model using the deep learning model transaction brokerage server by the deep learning solution platform, and the total sales amount of the prescription analysis model by the seller terminal using the deep learning model transaction brokerage server And it may be determined based on the total number of times the seller terminal trained the deep learning model using the deep learning solution platform during the entire period.

Finally, receiving the model selection signal of the purchaser from the purchaser terminal, and transmitting the model selection signal to the seller terminal corresponding to the model selection signal of the purchaser (S300) may be performed.

The deep learning model transaction brokerage method according to the start may further include a step of concluding a sale contract according to the transmission of the model selection signal of S300, and calculating a transaction cost according to the conclusion of the sale contract. However, since this is not a matter to be protected by the core of the present invention, it will be omitted below.

The user terminal according to one disclosure may include a seller terminal and a buyer terminal. Also, hereinafter, a user may refer to a seller or a buyer.

According to one disclosure, the user terminal may execute various types of applications, and may display the currently running application by visual or auditory display and provide it to the user. The user terminal may include a display module for visually displaying an application, and may include an input module for receiving a user's input, a communication module, a storage module storing at least one program, and a control module.

The user terminal according to one disclosure may be a mobile terminal such as a smart phone or a tablet PC, and according to an embodiment, a fixed device such as a desktop may also be included. Here, the user terminal includes a mobile phone, a smart phone, a laptop computer, a tablet PC, a wearable device, for example, a watch-type terminal (smart watch), a glasses-type terminal (smart glass) )) may be included.

The user terminal according to the present disclosure may download and install various applications by accessing an app store or a play store. Here, the application may be a web browser application that outputs content such as games, news, photos, and videos provided online, or a dedicated application for providing each content.

The user terminal according to one disclosure may be connected to a deep learning model transaction brokerage server through a communication network. Here, the communication network may include a wired network and a wireless network. More specifically, the communication network may include various networks such as a local area network (LAN), a metropolitan area network (MAN), and a wide area network (WAN). The communication network may also include the well-known World Wide Web (WWW). However, the communication network according to the present invention is not limited to the above-listed networks, and may include a well-known wireless data network, a well-known telephone network, a well-known wired or wireless television network, and the like. That is, the communication network may be any network method for connecting the user terminal and the deep learning platform providing apparatus.

The deep learning model transaction brokerage server according to one disclosure may be connected to the user terminal through an application, and may provide various information related to the transaction of the deep learning model in response to a request of the user terminal.

The processor of the deep learning model transaction brokerage server according to one disclosure can collect the purchase information of the buyer about the deep learning model required by the buyer from the buyer terminal, and sell the deep learning model corresponding to the purchase information of the buyer Search for sales information of at least one seller terminal, and transmit sales information of a deep learning model corresponding to the buyer's request information to the buyer terminal, receive a model selection signal of the buyer from the buyer terminal, and select the model of the buyer A model selection signal may be transmitted to the seller terminal corresponding to the signal.

The apparatus for providing a deep learning platform according to one disclosure may include a memory for storing one or more instructions and a processor for executing one or more instructions stored in the memory.

According to one disclosure, the processor may control at least one other component (eg, hardware or software component) of the device for providing a deep learning platform connected to the processor by, for example, driving software (eg, a program), Various data processing and operations can be performed. The processor may load and process commands or data received from other components into a volatile memory, and store resultant data in a non-volatile memory. According to an embodiment, the processor operates independently of the main processor (eg, a central processing unit or an application processor), and additionally or alternatively, uses less power than the main processor, or a coprocessor (eg, specialized for a specified function) : graphic processing unit, image signal processor, sensor hub processor, or communication processor). Here, the auxiliary processor may be operated separately from or embedded in the main processor.

According to one disclosure, the memory may store a program for processing and controlling the processor, and may store data input to or output from the apparatus of the present invention. Programs stored in the memory may be classified into a plurality of modules according to their functions, where the plurality of modules are software, not hardware, and are functionally operated modules. According to one disclosure, the programs stored in the memory may be classified into a plurality of modules according to their functions, where the plurality of modules are software rather than hardware, and may mean modules that operate functionally.

As an embodiment of the present invention, a computer-readable recording medium in which a program for implementing the above-described method is recorded may be provided.

That is, the above-described method can be written as a program that can be executed on a computer, and can be implemented in a general-purpose digital computer that operates the program using a computer-readable medium. In addition, the structure of the data used in the above-described method may be recorded in a computer-readable medium through various means. A recording medium for recording an executable computer program or code for performing various methods of the present invention should not be construed as including temporary objects such as carrier waves or signals. The computer-readable medium may include a storage medium such as a magnetic storage medium (eg, a ROM, a floppy disk, a hard disk, etc.) and an optically readable medium (eg, a CD-ROM, a DVD, etc.).

The above description of the present invention is for illustration, and those of ordinary skill in the art to which the present invention pertains can understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present invention. will be. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a dispersed form, and likewise components described as distributed may be implemented in a combined form.

Although the preferred embodiment of the present invention has been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements by those skilled in the art using the basic concept of the present invention as defined in the following claims are also provided. is within the scope of the That is, the scope of the present invention is indicated by the following claims rather than the above detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be interpreted as being included in the scope of the present invention. do.

Claims (1)

In the method of brokering a deep learning model transaction performed by a deep learning model transaction brokerage server,
collecting the purchase information of the purchaser regarding the deep learning model required by the purchaser from the purchaser terminal;
Searching for sales information of at least one seller terminal that sells the deep learning model corresponding to the purchase information of the buyer, and transmitting the sale information of the deep learning model corresponding to the request information of the buyer to the buyer terminal; And
Receiving the model selection signal of the purchaser from the purchaser terminal, and transmitting the model selection signal to the seller terminal corresponding to the model selection signal of the purchaser;
The deep learning sales information is,
Includes result information learned in response to the buyer's request information for a plurality of pre-prepared deep learning models,
The buyer's model selection signal is based on the learned result information,
Deep learning model trading brokerage method.

Images