KR102700445B1 - System for generating automatically machine learning model pipeline based on data - Google Patents

Abstract

The present invention relates to a system for automatically generating a data-based machine learning model pipeline, comprising: a data collection device which collects data uploaded by a user; a data analysis device which performs preprocessing on the data collected from the data collection device, recommends user-popular operators among user-generated operators previously generated by experts or non-experts and stored in advance according to preset user popularity based on the number of times other users have used them, and recommends similar operators corresponding to a categorized category or the same category among the recommended user-popular operators; an operator loader device which loads information data on the recommended similar operators using a separate operator database based on the similar operators recommended from the data analysis device; and an operator configuration device which automatically configures a pipeline so that a specific machine learning model can be generated based on the information data on the similar operators loaded from the operator loader device, thereby enabling non-experts to easily configure machine learning model development (ML Dev) and machine learning model operation (MLOps) at the same time.

Description

{SYSTEM FOR GENERATING AUTOMATICALLY MACHINE LEARNING MODEL PIPELINE BASED ON DATA}

The present invention relates to a system for automatically generating a data-based machine learning model pipeline.

The amount of big data that has been growing exponentially in recent years has gone beyond zettabytes and even reached yottabytes.

Rather than manually analyzing such massive data, research has been actively conducted recently on automating machine learning model development (Machine Learning Developments, ML Dev) to automatically analyze it, and machine learning model operations (Machine Learning Operations, ML Ops) to operate the developed machine learning models are gradually being introduced in each service.

However, in the case of machine learning model development (ML Dev), data analysis is difficult for non-experts, and even for experts, it is a task that requires a lot of manual work, so work efficiency may decrease. Therefore, it is necessary to develop a machine learning model creation automation and operation system that can provide services to non-experts up to the stage of automatic machine learning model development and operation by integrating machine learning model development (ML Dev) and machine learning model operation (ML Ops).

On the other hand, the variety of machine learning platforms and tools has increased, which has increased the range and variety of choices, but this has also caused problems such as increased complexity and a lack of uniformity. Due to these problems, the field of machine learning is becoming a field for experts, and the barrier to entry for non-experts is rising, which is causing many difficulties for those trying to operate services.

Domestic Publication Patent No. 10-2021-0055934 (Published on May 18, 2021)

The present invention has been made to solve the above-mentioned problems, and the purpose of the present invention is to provide a system for automatically generating a data-based machine learning model pipeline that allows even non-experts to easily configure machine learning model development (ML Dev) and machine learning model operation (ML Ops) at once by utilizing operators created in advance by experts and/or non-experts.

In order to achieve the above-mentioned object, one aspect of the present invention provides a system for automatically generating a data-based machine learning model pipeline, which includes: a data collection device that collects data uploaded by a user; a data analysis device that performs preprocessing on the data collected from the data collection device, recommends user-popular operators among user-generated operators previously generated by experts or non-experts and stored in advance according to a preset user popularity based on the number of times other users have used them, and recommends similar operators that fall into a category identical to or similar to a corresponding classified category among the recommended user-popular operators; an operator loader device that loads information data on the recommended similar operators using a separate operator database (DB) based on the similar operators recommended from the data analysis device; and an operator configuration device that automatically configures a pipeline so that a specific machine learning model can be generated based on the information data on the similar operators loaded from the operator loader device, thereby generating a pipeline for the specific machine learning model.

Here, the data analysis device preferably includes a data preprocessing module that performs preprocessing on data collected from the data collection module; a data category classification module that receives data on which preprocessing has been performed from the data preprocessing module and classifies which category the data on which preprocessing has been performed belongs using a YOLO model that has been labeled and learned in advance; a popular operator recommendation module that receives data classified into categories from the data category classification module and recommends user-popular operators based on preset user popularity according to the number of times other users have used the operator, among user-generated operators that have been previously created and stored by experts or non-experts; and a similar operator recommendation module that receives recommended user-popular operators from the popular operator recommendation module and recommends similar operators corresponding to a category identical to or similar to the classified category among the recommended user-popular operators based on the recommended user-popular operators, and also extracts preset hyper parameter values previously used by other users for the recommended similar operators.

Preferably, the data preprocessing module can automatically perform image preprocessing tasks such as adjusting the sharpness of the image data or deleting or enlarging the image of a portion to be classified in order to accurately classify the category of the image data among the data collected from the data collection module.

Preferably, a hyper parameter modification module may be further included to provide an input window for modifying the hyper parameters so that the preset hyper parameter values of the similar operators recommended from the similar operator recommendation module can be modified by the user.

Preferably, the operator can modularize the step-by-step execution tasks of the data workflow as a single process for an expert or non-expert to develop a specific machine learning model.

Preferably, the system may further include an operator tester device that receives information data of a pipeline for a specific machine learning model generated from the operator configuration device and tests the data analysis device, the operator loader device, and the operator configuration device's work processes to be automatically performed sequentially based on the data collected from the data collection device.

Preferably, the operator tester device may provide a user interface environment setting for custom hyperparameter changes so that the hyperparameter values for the operator of a specific machine learning model being tested can be changed by the user.

Preferably, an operator storage device may be further included that receives result information data of a specific machine learning model for which test execution has been completed from the operator tester device, and manages the hyperparameter values assigned to the operator together with the operator for the specific machine learning model for which test execution has been completed based on the data, and stores them in a separate operator database (DB).

Preferably, the operator storage device can provide a service to automatically distribute, through wired or wireless communication, the operator for a specific machine learning model for which the test execution has been completed, together with the hyperparameter values assigned to the operator, to a specific server that must operate the specific machine learning model.

According to the system for automatically generating a data-based machine learning model pipeline of the present invention as described above, there is an advantage in that even non-experts can easily configure machine learning model development (ML Dev) and machine learning model operation (ML Ops) at once by utilizing operators created in advance by experts and/or non-experts.

In addition, according to the present invention, there is an advantage in that the data and model performance differences occurring between machine learning model development (ML Dev) and machine learning model operation (ML Ops) can be minimized, and data analysis and service development can be popularized through the development of machine learning model operation (ML Ops) that can be automatically operated even by non-experts.

In addition, according to the present invention, there is an advantage in that data analysis is possible even without having the level of knowledge of an expert in collecting a large amount of data and developing a machine learning model based on the collected data, and this can help popularize service development.

In addition, according to the present invention, experts can maximize the efficiency of work by reducing the hassle of having to do work that they have already done, and even if the work is not done by the expert himself, but is done by another expert, the expert can retrieve and use the work, so there is an advantage in that the efficiency of the work can be increased.

In addition, according to the present invention, in the case of non-experts, a separate learning time is required to acquire the knowledge to operate machine learning, but when using this system, it can be expected that the learning time will be significantly reduced compared to the existing learning time, and also, by reducing the learning time when using this system to apply machine learning to a service, better quality services can be expected, which has the advantage of influencing the global market.

FIG. 1 is an overall block diagram illustrating a system for automatically generating a data-based machine learning model pipeline according to one embodiment of the present invention.

The above-mentioned objects, features and advantages are described in detail below with reference to the attached drawings, so that those with ordinary knowledge in the technical field to which the present invention pertains can easily practice the technical idea of the present invention. In explaining the present invention, if it is judged that a detailed description of a known technology related to the present invention may unnecessarily obscure the gist of the present invention, a detailed description thereof will be omitted.

Terms including ordinal numbers such as first, second, etc. may be used to describe various components, but the components are not limited by the terms. The terms are only used for the purpose of distinguishing one component from another. For example, the first component may be referred to as the second component, and similarly, the second component may be referred to as the first component, without departing from the scope of the present invention. The terminology used in this application is only used to describe specific embodiments and is not intended to limit the present invention. The singular expression includes the plural expression unless the context clearly indicates otherwise.

The terms used in the present invention are selected from the most widely used general terms possible while considering the functions of the present invention, but they may vary depending on the intention of engineers working in the field, precedents, the emergence of new technologies, etc. In addition, in certain cases, there are terms arbitrarily selected by the applicant, and in this case, the meanings thereof will be described in detail in the description of the relevant invention. Therefore, the terms used in the present invention should be defined based on the meanings of the terms and the overall contents of the present invention, rather than simply the names of the terms.

When a part of the specification is said to "include" a component, this does not mean that it excludes other components, but rather that it may include other components, unless otherwise specifically stated. In addition, terms such as "part", "module", etc., used in the specification mean a unit that processes at least one function or operation, which may be implemented by hardware or software, or a combination of hardware and software.

Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings. However, the embodiments of the present invention exemplified below may be modified into various other forms, and the scope of the present invention is not limited to the embodiments described below. The embodiments of the present invention are provided to more completely explain the present invention to those skilled in the art.

The combination of each block of the attached block diagram and each step of the flowchart may be performed by computer program instructions (execution engine), and these computer program instructions may be installed in a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus, so that the instructions, which are executed by the processor of the computer or other programmable data processing apparatus, create a means for performing the functions described in each block of the block diagram or each step of the flowchart. These computer program instructions may also be stored in a computer-available or computer-readable memory that can be directed to a computer or other programmable data processing apparatus to implement the function in a specific manner, so that the instructions stored in the computer-available or computer-readable memory can produce an article of manufacture including an instruction means for performing the functions described in each block of the block diagram or each step of the flowchart.

And, since the computer program instructions can also be installed on a computer or other programmable data processing equipment, a series of operational steps are performed on the computer or other programmable data processing equipment to create a computer-executable process, so that the instructions that perform the computer or other programmable data processing equipment can also provide steps for executing the functions described in each block of the block diagram and each step of the flowchart.

Additionally, it should be noted that each block or each step may represent a module, segment, or portion of code that includes one or more executable instructions for performing specified logical functions, and that in some alternative embodiments, the functions mentioned in the blocks or steps may occur out of order. For example, two blocks or steps depicted in succession may in fact be performed substantially concurrently, or the blocks or steps may be performed in the reverse order of their corresponding functions, as desired.

FIG. 1 is an overall block diagram illustrating a system for automatically generating a data-based machine learning model pipeline according to one embodiment of the present invention.

Referring to FIG. 1, a system for automatically generating a data-based machine learning model pipeline according to an embodiment of the present invention largely includes a data collection device (100), a data analysis device (200), an operator loader device (300), and an operator configuration device (400). In addition, the system for automatically generating a data-based machine learning model pipeline according to an embodiment of the present invention may additionally include an operator tester device (500), an operator storage device (600), and the like. Meanwhile, since the components illustrated in FIG. 1 are not essential, the system for automatically generating a data-based machine learning model pipeline according to an embodiment of the present invention may have more or fewer components than the components illustrated in FIG.

Hereinafter, components of a system for automatically generating a data-based machine learning model pipeline according to one embodiment of the present invention will be specifically examined.

The data collection device (100) performs the function of collecting data uploaded by the user and transmitting it to the data analysis device (200).

The data analysis device (200) reads and analyzes data uploaded by a user, and in particular, performs preprocessing on data collected from a data collection device (100), classifies categories of the preprocessed data, recommends user-popular operators based on preset user popularity according to the number of times other users have used them among user-generated operators previously created and stored by experts or non-experts, and also performs a function of recommending similar operators that fall into categories identical to or similar to the classified categories among the recommended user-popular operators.

This data analysis device (200) comprises, as illustrated in FIG. 1, a data preprocessing module (220), a data category classification module (240), a popular operator recommendation module (260), and a similar operator recommendation module (280).

Here, the data preprocessing module (220) performs a function of performing preprocessing on data collected from the data collection device (100) and transmitting the data to the data category classification module (240).

In addition, the data preprocessing module (220) can automatically perform image preprocessing tasks such as adjusting the sharpness of the image data or deleting and/or enlarging the image for the portion to be classified in order to accurately classify the category of the image data collected from the data collection device (100).

In addition, the preprocessing task of the data preprocessing module (220) can utilize the YOLO (You Only Look Once) model applied to the data category classification model (240) described later.

The data category classification module (240) receives preprocessed data from the data preprocessing module (220) and performs the function of classifying which category the preprocessed data belongs to using a YOLO (You Only Look Once) model that has been labeled and trained in advance.

In addition, the data category classification module (240) classifies a category, such as fish classification, animation character classification, or shoe brand classification, to determine which operator it is related to, and thereby increase classification accuracy.

Meanwhile, the YOLO model is a technology that detects a semantic region within an image (or video) in real time, for example, in the field of machine learning or deep learning. The semantic region is an image pattern for an object such as a person's face, a car, a tree, glasses, etc. that can be included in any image.

The above YOLO model can be configured to include a step of generating learning data in pairs of semantic regions to be recognized and classification information (i.e., labeling) therefor using a machine learning algorithm in the form of supervised learning, a step of generating learning model data using the thousands or more of the generated learning data as input parameters, and an object classification step of extracting learned semantic regions from an arbitrary (or specific) image using the generated learning model data.

In addition, as an example of a field utilizing the above YOLO model, in order to distinguish the type of an arbitrary vehicle driving on the road using a digital camera, more than thousands of training data consisting of pairs of vehicle images and classification information for the vehicle images (e.g., trucks, buses, passenger cars, etc.) are prepared, and model data is generated by executing the YOLO model machine learning function using the training data as an input parameter, and when the classification function of the YOLO model is executed using the generated model data and an arbitrary image as input parameters, the results of extracting the semantic domain can be provided in the form of a bounding box and classification information for the bounding box.

In addition, the above YOLO model has high performance and accuracy and fast execution speed in the field of machine learning image recognition, so it can be used in fields such as self-driving cars and real-time object recognition such as person detection in images.

In addition, in the object classification step of the above YOLO model, the target reliability of the initial detection object for accurate detection of the semantic region is specifically set, and the input image is divided into bounding box units of a certain size. This object detection process calculates the reliability of the detection candidate object that can be included in the bounding box for all the divided bounding boxes, and if the calculated detection candidate object reliability is higher than the target reliability of the initial detection object, the corresponding bounding box is extracted. If the bounding box is not extracted through the object detection process, the size of the bounding box is adjusted and the object detection process is repeatedly executed.

That is, the YOLO model described above has the features of viewing the entire image only once, integrating individual elements of object detection into a single neural network, and recognizing objects in real time. While other object recognition models combine and use various preprocessing models and artificial neural networks, the YOLO model processes everything with a single artificial neural network, so it can quickly recognize objects in images.

Then, the YOLO model divides each image into SxS grids and calculates the reliability of the grid. The reliability reflects the accuracy of object recognition in the grid. Initially, a bounding box that is far from object recognition is set, but by calculating the reliability and adjusting the location of the bounding box, a bounding box with the highest object recognition accuracy can be obtained. To calculate whether an object is included in the grid, the object class score is calculated. As a result, a total of SxSxN objects are predicted. Most of these grids have low reliability. To increase the reliability, the surrounding grids can be merged.

After that, you can set a threshold to remove unnecessary parts. The above YOLO model is very fast with simple processing. Compared to other existing real-time vision technologies, it shows about twice the performance, because it classifies by looking at the entire image at once.

The popular operator recommendation module (260) receives categorized data from the data category classification module (240) and, based on this, recommends popular user operators among user-created operators previously created and stored by experts and/or non-experts, based on preset user popularity according to the number of times other users have used them.

In addition, the similar operator recommendation module (280) receives user popular operators recommended by the popular operator recommendation module (260) and, based on the received popular operators, recommends similar operators that fall into the same or similar category as the classified category among the recommended user popular operators. In addition, it also performs a function of extracting preset hyper parameter values previously used by other users for the recommended similar operators.

That is, similar operators extracted through the similar operator recommendation module (280) also include hyper parameter values that other users have previously used, so that they can be used immediately without separate settings by the user. However, in case of low accuracy, hyper parameter values can be arbitrarily modified.

At this time, the hyper parameter value is a value reflected in the learning process of the machine learning model, and in simple terms, it is a value directly set by the developer of the machine learning model. In other words, the hyper parameter value is configured as a value that is set for data learning and tuned while repeating learning and verification until the learned model produces the desired performance. These hyper parameters may include, for example, a learning rate, number of hidden layers, batch size, number of epochs, type of loss function, C, gamma, depth, etc.

Meanwhile, although not shown in the drawing, the data analysis device (200) may further include a hyper parameter modification module that provides an input window for modifying the hyper parameters so that the preset hyper parameter values of similar operators recommended from the similar operator recommendation module (280) can be modified by the user.

On the other hand, the operator mentioned throughout the present invention refers to a process for experts and/or non-experts to develop a specific machine learning model. In other words, it refers to the step-by-step execution of data workflow, such as data preprocessing/postprocessing modules. And, by assigning hyper parameters to the operator, the accuracy of the test results after learning can be adjusted.

In addition, the operator applied to the present invention may include functions such as data collection, machine learning model development algorithm, machine learning model verification algorithm, and machine learning model deployment. In this way, since the operator created in advance by experts and/or non-experts can be reused, machine learning model development (ML Dev) and machine learning model operation (ML Ops) can be easily configured at once.

The operator loader device (300) performs a function of loading information data on similar operators recommended from the data analysis device (200) using a separate operator database (DB) (350).

The operator configuration device (400) automatically configures a pipeline so that a specific machine learning model can be generated based on information data about similar operators loaded from the operator loader device (300), and performs the function of generating a pipeline for the specific machine learning model.

In addition, although not shown in the drawing, a power supply device is further included to supply power required for each of the aforementioned device components, i.e., a data acquisition device (100), a data analysis device (200), an operator loader device (300), an operator configuration device (400), an operator tester device (500), and an operator storage device (600). Although it is preferable to implement a power supply device that converts commercial alternating current (AC) power (e.g., AC 220 V) into direct current (DC) and/or alternating current (AC) power for continuous power supply, the present invention is not limited thereto, and may be implemented by including a conventional portable battery.

In addition, the power supply device may include a power management unit (not shown) that protects components from external power shock and performs the function of outputting a constant voltage. The power management unit may include an ESD (Electro Static Damage) protector, a power detector, a rectifier, and a power circuit breaker.

Here, the ESD protector is configured to protect the electrical components from static electricity or sudden power shock. The power detector is configured to send a blocking signal to the power circuit breaker when a voltage outside the allowable voltage range is input, and to transmit a boost or lower signal to the rectifier according to a voltage change within the allowable voltage range. The rectifier is configured to perform a rectification operation of boost or lower voltage according to a signal of the power detector so as to minimize fluctuations in the input voltage and supply a constant voltage. The power circuit breaker is configured to block power supplied from a battery according to a blocking signal transmitted from the power detector.

The various embodiments described herein may be implemented in a computer-readable recording medium or a similar device, for example, using software, hardware, or a combination thereof.

In terms of hardware implementation, the embodiments described herein can be implemented using at least one of Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), Field Programmable Data Arrays (FPGAs), Processors, Controllers, Micro-Controllers, Microprocessors, and electrical units for performing functions. In some cases, such embodiments can be implemented by at least one of the device components described above.

In a software implementation, embodiments such as procedures or functions may be implemented with separate software modules that perform at least one function or operation. The software code may be implemented by a software application written in a suitable programming language. In addition, the software code may be stored in a separate storage device (not shown) and executed by at least one of the device components described above.

The storage device may include at least one type of storage medium among, for example, a Flash Memory type, a Hard Disk type, a Multimedia Card Micro type, a card type memory (e.g., an SD or XD memory, etc.), a Random Access Memory (RAM), a Static Random Access Memory (SRAM), a Read-Only Memory (ROM), an Electrically Erasable Programmable Read-Only Memory (EEPROM), a Programmable Read-Only Memory (PROM), a magnetic memory, a magnetic disk, and an optical disk.

Additionally, the operator tester device (500) receives information data of a pipeline for a specific machine learning model generated from the operator configuration device (400) and performs a function of testing the work processes of the data analysis device (200), the operator loader device (300), and the operator configuration device (400) to be automatically performed sequentially based on the data collected from the data collection device (100).

Additionally, the operator tester device (500) may provide a user interface environment setting for changing hyper parameters customized for the user so that the hyper parameter values for the operator of the specific machine learning model being tested can be changed by the user.

That is, the user can check the results of the machine learning model for which the test execution is completed through the operator tester device (500), and if not satisfied, the user can change the hyper parameter values that can be assigned to the operator and perform the test again. When a satisfactory result is obtained by continuing to change the hyper parameter values in this way, the operator for the corresponding data and the hyper parameter values assigned to the operator are stored in a separate operator database (DB) (350) through the operator storage device (600) described below, and the stored operator can be reused when the operator is recommended and automatically configured thereafter.

In addition, the operator storage device (600) receives result information data of a specific machine learning model for which test execution has been completed from the operator tester device (500), and based on this, manages the hyper parameter values assigned to the operator along with the operator for the specific machine learning model for which test execution has been completed so that they are stored in a separate operator database (DB) (350).

In addition, the operator storage device (600) can perform a function of providing a service to automatically distribute the hyper parameter values assigned to the operator for a specific machine learning model for which the test has been completed, together with the operator for the specific machine learning model, to a specific server (20) that must operate the specific machine learning model via wired and/or wireless communication via a communication network (10).

At this time, the communication network (10) is a communication network that is a high-speed backbone network of a large communication network capable of providing large-capacity, long-distance voice and data services, and may be a next-generation wireless communication network including WiFi, WiGig, Wibro (Wireless Broadband Internet, Wibro), WiMAX (World Interoperability for Microwave Access, Wimax), etc. for providing the Internet or high-speed multimedia services.

The above Internet refers to a worldwide open computer network structure that provides various services existing in the TCP/IP protocol and its upper layer, such as HTTP (Hyper Text Transfer Protocol), Telnet, FTP (File Transfer Protocol), DNS (Domain Name System), SMTP (Simple Mail Transfer Protocol), SNMP (Simple Network Management Protocol), NFS (Network File Service), NIS (Network Information Service), etc., and provides an environment that allows an operator storage device (600) to be connected to a server (20). Meanwhile, the Internet may be a wired or wireless Internet, or may be a core network integrated with a wired public network, a wireless mobile communication network, or a portable Internet.

If the communication network (10) is a mobile communication network, it may be a synchronous mobile communication network or an asynchronous mobile communication network. As an example of the asynchronous mobile communication network, a WCDMA (Wideband Code Division Multiple Access) type communication network can be mentioned. In this case, although not shown in the drawing, the mobile communication network may include, for example, an RNC (Radio Network Controller). Meanwhile, although the WCDMA network is mentioned as an example, it may be a next-generation communication network such as a cellular-based 3G network, LTE network, 4G network, 5G network, or other IP-based IP network. This communication network (10) performs the role of mutually transmitting signals and data between the operator storage device (600) and the server (20).

A system for automatically generating a data-based machine learning model pipeline according to one embodiment of the present invention as described above analyzes data to develop and operate a machine learning model, performs standardization of a platform for developing machine learning models for users who wish to do so, and lowers the entry barrier for non-experts through a standardized pipeline, so that anyone can develop and operate a machine learning model.

In addition, the system of the present invention generates operators that match the level of knowledge and competency possessed by experts and/or non-experts, and the generated operators can be used as is by other experts and/or non-experts. Therefore, most of the detailed work and creation of operators that require specialized knowledge will be generated by experts while developing the model.

On the other hand, even if a non-expert wants to develop something, they need to create an operator, but if there is an operator that has already been created in advance by an expert to process a similar type of data pipeline, they can load the operator and use it by adjusting the detailed settings.

Once machine learning development is completed, the development is automatically deployed to the server that needs to operate it through an automated system module. All processes in the middle are automated through machine learning model development (ML Dev) and machine learning model operation (ML Ops), lowering the entry barrier for non-experts and enabling experts to work faster.

That is, experts and/or non-experts modularize various detailed tasks for creating machine learning models and create operators in advance. Information about the created operators is stored in a separate database (DB) and can be retrieved and used whenever needed.

Among them, the present invention focuses on machine learning model development (ML Dev). In other words, it focuses on the process of combining multiple operators to create a machine learning model, testing it, and the stage just before deploying it to operation.

Although a preferred embodiment of a system for automatically generating a data-based machine learning model pipeline according to the present invention has been described above, the present invention is not limited thereto, and various modifications may be made and implemented within the scope of the claims, the detailed description of the invention, and the attached drawings, which also belong to the present invention.

100 : Data collection device,
200: Data analysis device,
300: Operator loader device,
400: Operator Configuration Device,
500 : Operator Tester Device,
600 : Operator Storage Device

Claims (9)

A data collection device that collects data uploaded by users;
A data analysis device that performs preprocessing on data collected from the above data collection device, classifies the categories of the preprocessed data, recommends user-popular operators among user-generated operators previously created and stored by experts or non-experts based on preset user popularity according to the number of times other users have used them, and recommends similar operators that fall into categories identical to or similar to the classified categories among the recommended user-popular operators;
An operator loader device that loads information data on similar operators recommended from the above data analysis device using a separate operator database (DB);
An operator configuration device that automatically configures a pipeline so that a specific machine learning model can be generated based on information data about similar operators loaded from the operator loader device, and generates a pipeline for the specific machine learning model;
An operator tester device that tests the work processes of the data analysis device, the operator loader device, and the operator configuration device to be performed sequentially and automatically based on the data collected from the data collection device; and
An operator storage device that receives result information data of a specific machine learning model for which test execution has been completed from the operator tester device and manages the hyper parameter values assigned to the operator together with the operator for the specific machine learning model for which test execution has been completed based on the data, and stores them in a separate operator database (DB).
The above operator is a modularized version of the step-by-step data workflow tasks that experts or non-experts can perform as a process for developing a specific machine learning model.
The above data analysis device automatically performs image preprocessing tasks such as adjusting the clarity of the image data or deleting or enlarging the image of a portion to be classified in order to accurately classify the category of the image data collected from the data collection device, and classifies which category the preprocessed data belongs to using a YOLO model that has been labeled and learned in advance, and also extracts preset hyper parameter values previously used by other users for the recommended similar operators, and provides an input window for modifying the hyper parameters so that the preset hyper parameter values of the recommended similar operators can be modified by the user.
The YOLO model applied to the above data analysis device is configured to include a step of generating learning data in pairs of semantic areas to be recognized and classification information therefor using a machine learning algorithm of a supervised learning method, a step of generating learning model data using the generated learning data as an input parameter, and an object classification step of extracting a semantic area learned from a specific image using the generated learning model data.
The object classification step of the above YOLO model sets the target reliability of the initial detection object for accurate detection of the corresponding semantic region, divides the corresponding specific image into bounding box units of a certain size, and in the corresponding object detection process, calculates the reliability of the detection candidate object that can be included in the bounding box for all the divided bounding boxes, and if the calculated detection candidate object reliability is higher than the target reliability of the initial detection object, the corresponding bounding box is extracted, and if the bounding box is not extracted through the corresponding object detection process, the size of the bounding box is adjusted and the object detection process is repeatedly executed.
The above operator tester device provides a user interface environment setting for changing hyperparameters customized for the user so that the hyperparameter values for the operator of the specific machine learning model being tested can be changed by the user.
The above operator storage device is a system for automatically generating a data-based machine learning model pipeline, characterized in that it provides a service to automatically distribute, through wired or wireless communication, the operator for a specific machine learning model for which the test execution has been completed, and the hyperparameter values assigned to the operator to a specific server that must operate the specific machine learning model. delete delete delete delete delete delete delete delete

Images