인공지능 데이터 라벨링 전문가 교육 프로그램

과학기술정보통신부는 2023년 인공지능 데이터 라벨링 전문가 양성을 위한 필수 교육을 실시할 예정입니다.

### 2023 인공지능 데이터 라벨링 전문가 교육 개요
* 📚 과학기술정보통신부가 한국정보화진흥원 주관으로 **'2023년 인공지능 데이터 라벨링 전문가 양성과정 필수 교육'**을 시작합니다.
* 🎓 첫 번째 교육에서는 **인공지능 학습 데이터 구축 지원 사업의 이해, 인공지능 학습 데이터 라벨링 전문가 양성과정 안내, 인공지능 학습 데이터 구축 능력 국가직무능력표준 기반 경력개발 경로 안내** 등 전체 커리큘럼을 소개합니다.
* 🛠 인공지능 학습용 데이터 구축 사업은 **인공지능 분야의 글로벌 경쟁력 강화와 생태계 및 데이터 인프라 확보**를 통한 정상화 추진을 목표로 합니다.
* 📊 인공지능 학습을 위해 수집되는 데이터에는 **이미지, 동영상, 텍스트, 음성, 다각도 객체 이미지, 센서 데이터** 등 6가지 유형이 있습니다.
* 🏗 데이터 구축 프로세스에는 **미션 정의, 데이터 수집 및 수집, 데이터 정제, 데이터 라벨링, 데이터 트레이닝**이 포함됩니다.
* 🖼 데이터 라벨링 유형에는 **경계 상자, 다각형, 폴리라인, 키 포인트, 직육면체, LiDAR 센서의 PCD 데이터, 전사** 등이 있습니다.
* 📝 교육 과정에는 **기본 이미지 및 동영상 라벨링**부터 **전문가 과정, 데이터 관리자 과정**까지 다양한 온라인 및 오프라인 교육 프로그램이 포함되어 있습니다.
* 📈 인공지능 학습 데이터 구축 능력 국가직무능력표준(NCS)은 **필요한 직무 능력을 정의**하고 수행 근거를 통해 제시하여 **낮은 수준부터 높은 수준까지 경력 개발 경로**를 제공합니다.

The Ministry of Science and ICT, sponsored by the Korea Information Society Development Institute, will begin the essential education of the Artificial Intelligence Data Labeling Specialist Training for 2023. Hello, I am Jung Yong-hoon from the media group Saranggasu, in charge of the essential course lectures for the Artificial Intelligence Data Labeling Specialist Training. The lecture content of the essential education consists of two main topics: introducing the education process of the Artificial Intelligence learning data labeling specialist training and providing basic education on Artificial Intelligence to understand what it is and how it is being utilized. The first session of education introduces the entire curriculum, including: understanding the support project for constructing Artificial Intelligence learning data, guidance on the specialist training course for Artificial Intelligence learning data labeling, and guidance on career development paths based on the National Competency Standards for Artificial Intelligence learning data construction abilities. This is the order of the three topics.

It is composed. The goal of this first session is as follows: First, you can understand the purpose, types, and construction process of supporting projects for artificial intelligence data. Second, you can understand the labeling techniques for artificial intelligence data. Third, you can understand the educational process of specialized training for artificial intelligence data labeling. Lastly, you can understand the career development path through the NCS competency units, which are national job competency standards. Now, let's start by introducing the curriculum of the first session, beginning with an explanation of the supporting project for building artificial intelligence learning data. The project for building artificial intelligence learning data is one of the national policy projects being implemented by the government. It aims to enhance global competitiveness in the field of artificial intelligence, a core resource for digital transformation, and to promote normalization by securing an ecosystem and data infrastructure. In other words, it is a project to secure learning data that can be the foundation for the development of the artificial intelligence field across society.

Opening up a lot of data can be seen as providing advanced artificial intelligence and digital platform government infrastructure. The data construction project, which began in 2017, has transitioned and expanded into a national government-led artificial intelligence learning data construction project under the Smart Growth 4.0 policy, starting with data construction to support the Fourth Industrial Revolution. Currently, with the emergence of GPT in 2023, the flow of artificial intelligence learning data construction projects is facing a change towards new ultra-large AI data construction. So, what kind of data is being constructed for artificial intelligence learning? Let's introduce six types of collected data. The first and second types, image and video data, refer to the form of photos or videos, allowing for the detection of hazardous conditions of workers in danger zones and emotion recognition through facial expressions. Additionally, it detects and tracks objects in videos to predict situations, among other things.

Analyzing human and object movements can also be utilized for diagnosing diseases in human and animals. The third is text, which means developing services used in messengers utilizing letters and character data, as well as automatic translation of foreign languages and automatic summarization of documents on web pages, being used in artificial intelligence services to find numerous clickbait articles. The fourth, widely used, is voice data. When recording this voice, it is displayed as letters and characters, executing commands when given through voice, and even engaging in conversations with artificial intelligence through voice. Next, capturing images of an object from various angles to compile them at once is a type of data that expresses three-dimensional depth and texture. The last type of data is sensor data. Using lidar sensors, it calculates the distance information of the signal reflected from the object that was emitted. In such cases, the point referred to as a point that comes out is imported, and we commonly refer to that form as.

Autonomous driving or representing three-dimensional spatial information outdoors has been widely used. Dozens of data are obtained to create data for artificial intelligence learning, as explained earlier. It seems like you are seeing the process on the screen to define the mission, acquire data, extract necessary information from the acquired data, define content to be taught to artificial intelligence from meaningful static data through data labeling, and finally teach the constructed artificial intelligence learning data to artificial intelligence. Through this series of processes, we will prepare to develop intelligent artificial intelligence services. Now, let's provide more detailed explanations for each step. In the first step of the construction process, the mission definition stage, it is the process of defining what data will be created by whom and how. It mainly refers to activities in the design and planning stages, such as business execution plans, construction plans, quality management.

You will be writing various plans such as project proposals. The second stage of construction is data acquisition and collection. It is the process of acquiring and collecting data defined in the mission definition stage among all types of data such as image, video, audio, and text, as well as all data that can be obtained in Ilsan. When acquiring this data, it is essential to ensure that there is no violation of personal information or copyright laws, which have been strengthened recently. As you can see in the image, recording audio, capturing specific situations on video or in photos, scanning documents or specific texts are ways to obtain and collect data for artificial intelligence learning. The third stage of construction is data refining. It can be described as the process of extracting only meaningful data necessary for artificial intelligence learning from randomly imported data. Data collected in the form of videos, as seen on the screen, is part of this process.

If we assume this video, we can see that it is composed of many consecutive still images. Among countless consecutive still images, it is possible to refine and extract meaningful images that can be used as artificial intelligence training data. The most labor-intensive and time-consuming process in the process of constructing artificial intelligence data is data labeling. Artificial intelligence labelers or artificial intelligence labeling may have been introduced through various media at least once. If one image has been refined, the next step is to create material that serves as a role of guiding artificial intelligence, telling it what it is, like a teacher to artificial intelligence. Now, what types of labeling are there? Many labeling workers may have experienced at least one of the types of labeling introduced. The first type of labeling is the most common.

Bounding boxes are widely used, as you can see in the video of Hwayeong. Bounding boxes are a type of labeling where objects are selected in a rectangular shape, which is relatively easy. One drawback is that even though you select the actual object within the rectangle, there are often many empty areas included that are not related to the object, which can be visually noticeable. The next labeling type is polygons. As you see in the video, polygons are a labeling type that accurately selects the shape of the object. One drawback is that it takes relatively longer to select objects, but compared to bounding boxes, it allows for more precise selection of objects, making it advantageous for creating good training data for artificial intelligence. The next labeling type is polyline. Although polylines, consisting of line segments, are not widely used as other labeling types, they are used in the autonomous driving field for lane or sidewalk distinction.

After that, the next labeling type is key points. It is also called the skeleton type because it defines joint points such as those of the human body or animals' faces to analyze movements and perceive states and situations through changes in the angles of each joint. Recently, it has been widely used in the sports field. The next labeling type is cuboid. This cuboid type, which is currently the most widely used in the autonomous driving field, extracts three-dimensional spatial information from images of the cuboid plane. The task involves connecting two rectangles to create space as seen on the screen. Next is the labeling type of PCD data by a LiDAR sensor. This PCD data stands for Point Cloud Data, meaning points gathered in a cloud-like form, hence the term cloud data. Therefore, it is referred to as point cloud data. The screen shows labeling of PC data in the autonomous driving field.

It's the appearance of labeling, moving the mouse up, down, left, and right in a space of 3,000 won to select objects in the vehicle. The last labeling type is called transcription, which labels voice into text. First, let's show the video on the screen. When I was young, in elementary and middle school, my favorite subject was history, and I can proudly say that. Also, I used to enjoy reading books by Choe Mi-ro. As you can see, we are working on transferring speech into text. Please note that this labeling method is different from visual labeling, focusing on spelling, spacing, and standard language usage before transferring it into text. You may notice that it is a different type of labeling technique for artificial intelligence learning data construction.

The final step in the process is data training. As mentioned earlier, in order to inform artificial intelligence, training data is created, and artificial intelligence is educated for a certain period to acquire cognitive abilities such as perception, identification, judgment, etc. through artificial intelligence. In this way, the process of educating artificial intelligence for a certain period can be said to be the process of building training data for artificial intelligence to learn something. Those of you who are interested in and participating in the artificial intelligence training data construction project can be considered as teachers who are building better data to educate artificial intelligence. Next is an introduction to the educational process, the second step, which is the introduction of a specialized training course for labeling artificial intelligence training data. First, comparing the 2022 and 2023 educational programs, how have the educational programs improved and what kind of educational...

Let's find out about the newly established courses. First, I will explain about the required curriculum. The 2022 required curriculum included explanations about understanding artificial intelligence and supporting the establishment of artificial intelligence learning data. However, there was no guidance on what educational courses students should take for career development, which I think caused difficulties for many students. This year, in the 2023 curriculum, guidance on educational courses for career development has been added through the ongoing required education. Secondly, we have separated the Data Expert courses by difficulty level. In 2022, education was conducted through a single course such as Multi-Model Course, LIDAR Sensor Course, and Autonomous Driving Integration Sensor Course. However, in 2023, after the basic course explaining fundamental concepts in the field, we added advanced courses such as development directions and connections to other fields as advanced courses.

Considering the similarity of education, this year we have integrated the autonomous owner limit process into a difficulty-based arrangement, which can be called a trend for 2023, related to super AI vision technology and image caption education. Furthermore, a new education program has been established. Lastly, the graduation criteria have been slightly changed compared to last year, and we will guide you through this in the final chapter. The online education program for 2023 is as shown on the screen. First, in the mandatory education, it is composed of an orientation to the education process and small-scale education to help understand artificial intelligence. The next step, the data labeler education program, is divided into three levels for each of the most commonly used image and audio-text fields in the data labeling field, and in the introductory course, only theoretical education is provided. In the basic and advanced courses, practical data labeling processes are also included.

Next, it is structured into expert training courses in the field of data labeling, which includes an advanced level data expert course and a data manager course for managing data. The data expert course consists of four areas: multimodal vision technology, autonomous driving, image captioning, with both basic and advanced levels having practical assignments. The data manager course consists of two areas: quality management and security management, with practical assignments in quality management training. Lastly, for the Artificial Intelligence Mobility Data Construction Project, there is a managerial level course called the Construction Project Manager course, which consists of data planning and project management, without practical training. The offline training course is structured similarly to the online training course, but the main difference is that in the case of training with practical assignments, learners can receive real-time feedback on the assignments directly from the instructor and at the training location.

Let's explore detailed educational content by course from the next chapter. It is recommended to take all courses sequentially rather than being mandatory. If you are looking to develop a career in the field of data labeling, it is also recommended to gradually learn about it. All educational programs were created based on the Quality Management Guide and Construction Guide 3.0 version of the Korea Artificial Intelligence Society. First, let's look into the field of image and video. This is a recommended course for those who wish to engage in activities related to data labeling in the field of images and videos. Through this course, basic labeling techniques and concepts in the field of image and video will be taught. Next, similar to the field of image and video, there is also a high demand for education in the field of speech and text. This course covers the most fundamental educational content in the field of speech and text, suitable for those interested in activities involving data and labeling in the field of speech and text.

I recommend this training to those who are interested. Through this education, we provide basic labeling techniques and concepts in the field of speech text. Next, for those who want to engage in expert labeling activities, we recommend taking both the image video field and the speech text field data labeling education after completing the expert course. In the case of multi-model education, a fusion of image video and speech text is provided, covering basic concepts and expert labeling techniques in the field of multi-model. This year, a newly established vision technology education focuses on artificial intelligence estimating and analyzing objects or things, rather than the general field of image videos. It is an advanced stage of the image video field, providing education on the domain and labeling techniques of this field. It is education on autonomous driving. The basic principles of autonomous driving artificial intelligence.

Education is provided on the labeling method of point cloud data, which represents spatial information necessary for the learning of autonomous driving artificial intelligence. Another newly established training this year is on image and video captioning, which is widely used in the field of ancient AI, to help artificial intelligence perceive situations like humans by explaining objects or situational relationships in images or videos. Next is training for data managers, skilled data labelers, or artificial intelligence training data managers. Among them, in quality management education, training is conducted for high-quality data quality management to improve the performance of artificial intelligence. In security management education, training related to laws and regulations such as personal information or sensitive information is provided within the category of security laws, educating on the security scope of data necessary for artificial intelligence learning. Lastly, artificial intelligence passenger...

I will become a project manager for data construction support projects. Education on overall understanding of data construction support projects, from data planning to project management, will be provided. Detailed information on all processes, including data labeler construction project management, can be guided through subject-specific education. Let's find out about the application process for online and offline education registration. All education registrations are done through the education site, so please make sure to refer to it. Now, let's first look into the procedure for online education registration. First, online registration will be conducted in a total of 9 sessions, with a 2-week interval between each session. During the first week of registration, you can select the desired educational program and apply for registration. Some registrations may end early due to limited capacity.

During the registration period, education courses that do not close will have an additional week for registration. After the first week of registration deadline, the following week, for two weeks, you can take the education courses on the desired date and time. However, please note that after these two weeks, due to the end of the course period for the registered courses, you will no longer be able to attend the classes. Sometimes, students ask if they won't be able to register again for the course if they couldn't attend the classes due to personal reasons during the education period. Registration is always possible again, so there is no need to worry too much. Next, let's look into the registration process for offline education courses. Offline classes are conducted nationwide, taking into account the location and schedule.

The location of the training venue, training schedule, and training process will be announced as a notice on the training website one month before the training. Just like online training, offline training also has a first-come, first-served deadline. You can apply up to a week before the training, but unlike online training, if the number of participants does not reach the scheduled number for offline training, the regular opening may not take place, and we will inform the students separately. Also, offline training will only start on specific dates, not on a rolling basis, so please make sure to attend according to the training schedule you have applied for. Finally, let's look at the criteria for completing the course. As mentioned earlier, once you enroll in the selected course, you will first take a self-assessment quiz before the course.

Quizzes are not for evaluating the level of students, but for self-assessing the understanding of the subject. It is not related to completion, so please don't feel pressured and just enjoy it. After completing the self-check quiz, you will watch educational videos on the subject. Once the viewing is done, you will take another quiz at the same level as the self-check quiz. This is designed for students to compare their knowledge before and after the course. After finishing the self-paced training, a satisfaction survey about the course will be conducted to reflect your opinions for improvement. Not all courses have practical assignments. In courses with practical assignments, you must submit them for evaluation by the instructors, leading to course completion.

You will be able to determine whether or not it is a development path for job careers through the third chapter in the first session curriculum, which is the National Competency Standards (NCS) for Artificial Intelligence Learning Data Construction Ability. NCS is a system that defines the level of job and abilities that can be fostered through education, training, or certification using NCS in fields such as industrial site-centered talent development. If there are requirements for job abilities or personnel in the industrial field as seen on the screen, the national competency standard NCS defines the required job abilities in the industrial field. Through activities such as education, training, or certification acquisition, individuals can be nurtured with the required job abilities demanded in the field, allowing them to demonstrate those abilities again in the industrial field. This system explains how individuals can then utilize those abilities in the industrial field.

In October 2022, an NCS capability unit called artificial intelligence learning data construction related to one of the Earth's capabilities was developed. It defines the necessary job skills and presents job skills through performance evidence. For more details, please refer to the URL address on the screen. The artificial intelligence learning data construction NCS capability unit is classified into the largest category, information communication, and the subcategory information technology, and further classified into the smallest category, artificial intelligence, specifically artificial intelligence learning data construction. By clicking through the classification system, you can see nine artificial intelligence learning data construction NCS capability units. These units present overall job skills related to artificial intelligence learning data construction. By clicking on the preview button on the right, you can see more detailed performance evidence.

There are numbers, and the numbers indicated for each ability level are higher for advanced job skills. So, how can we find a career development path through NCS? As you can see on the screen, we have established a career development sequence from a low level to a high level. First of all, our education process starts with basic courses such as Data Labeler Course, Data Analyst Course, Data Manager Course, and finally, the Business Manager Course. We have defined jobs in each field and created a path on the screen for career professionals to advance to a higher level through learning. With this, we have explained all the essential training courses and introduced the first session. Thank you to all the students who have listened until the end. In the second session, we will meet again in the basic education on studying artificial intelligence.

[2023] 인공지능 학습용 데이터 라벨링 전문교육 - 필수(1/2)

한국지능정보사회진흥원(NIA)에서 운영하는 AI Hub(www.aihub.or.kr) 공식 유투브 채널입니다.

NIA AI Hub

인공지능 데이터 라벨 제작의 필수 교육

과학기술정보통신부가 한국정보화진흥원 주관으로 2023년 인공지능 데이터 라벨링 전문가 양성을 위한 필수 교육인 '인공지능 데이터 라벨링 전문가 양성과정'을 개설합니다.

교육은 인공지능 학습용 데이터 라벨링 전문가 양성과정 소개, 인공지능 기초 교육, 경력 개발 경로 안내 등으로 구성됩니다.

이 사업은 이미지, 영상, 텍스트, 음성, 3D 객체, 센서 데이터 등 다양한 형태의 인공지능 학습용 데이터를 확보해 지능형 인공지능 서비스를 개발하는 것을 목표로 합니다.

구축 과정은 미션 정의, 데이터 수집, 정보 추출, 데이터 라벨링, 구축된 데이터를 AI에 학습시키는 과정으로 구성되며, 이를 통해 AI 서비스 개발을 준비합니다.

인공 지능을 위한 데이터 수집 및 라벨링 계획 수립

데이터 수집 단계에서는 개인정보 보호 및 저작권법을 준수하면서 일산에서 다양한 유형의 데이터를 수집합니다.

데이터 정제 단계에서는 AI 학습을 위한 의미 있는 데이터를 추출하는 데 중점을 두며, 특히 동영상에서 정지 이미지를 정제하는 데 중점을 둡니다.

데이터 라벨링에는 바운딩 박스, 폴리곤, 폴리라인, 키 포인트, 큐로이드, PCD 데이터와 같은 일반적인 유형이 포함되며, 각 유형은 AI 학습에서 특정 용도로 사용되며 폴리곤은 정확한 물체 선택을 제공하고 키 포인트는 스포츠 분야에서 널리 사용됩니다.

인공지능 학습 데이터 구축의 진화

인공지능 학습 데이터를 구축하는 과정에는 시각적 라벨링, 전사 등 다양한 유형의 라벨링 기법이 사용됩니다.

2023년 교육 프로그램은 경력 개발 과정에 대한 안내를 추가하고 데이터 전문가 과정을 난이도별로 구성하는 등 개선되었습니다.

또한 인공지능 비전 기술에 대한 교육 프로그램을 새롭게 도입하고 졸업 기준도 일부 변경했습니다.

데이터 라벨링 전문가 교육 과정

교육은 멀티모달 비전 기술, 자율주행, 이미지 캡션 등을 다루는 고급 데이터 전문가 과정과 데이터 관리자 과정으로 구성됩니다.

이 과정은 한국인공지능학회의 품질 관리 가이드와 구축 가이드 3.

0 버전을 기반으로 한 실습 과제를 제공합니다.

이 프로그램에는 품질 관리 및 보안 관리 교육도 포함되어 있으며, 실습 과제 중 학습자를 위한 실시간 피드백에 중점을 두고 있습니다.

데이터 구축 지원 프로젝트의 교육 등록 및 과정 이수 절차

데이터 기획, 프로젝트 관리 등 데이터 구축 지원 사업에 대한 전반적인 이해를 위한 교육이 제공됩니다.

온라인 교육 등록은 2주 간격으로 총 9개 세션으로 구성되며, 과정 선택 및 신청이 가능하고 재등록도 가능합니다.

오프라인 교육 과정은 전국적으로 진행되며, 선착순 마감일과 구체적인 시작 날짜가 정해져 있습니다.

과정을 완료하려면 자가 평가 퀴즈, 교육 동영상 시청, 경우에 따라 평가를 위한 실습 과제를 제출해야 합니다.

인공지능 경력 개발을 위한 국가 역량 표준

국가직무능력표준(NCS)은 산업 현장에서 요구되는 직무 능력과 개인이 이러한 능력을 습득하고 발휘할 수 있는 방법을 정의합니다.

2022년 10월, 인공지능 학습 데이터 구축에 대한 새로운 NCS 능력 단위가 개발되어 세부 직무 능력과 수행 근거를 제공합니다.

NCS는 데이터 라벨러와 같은 기초 과정부터 비즈니스 관리자와 같은 고급 과정까지 경력 개발 경로를 제공하여 전문가가 경력을 발전시킬 수 있도록 안내합니다.