인공지능 분야 천상계 대한민국 단 4명 뿐인 캐글 그랜드 마스터 인터뷰

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
Hello everyone. It's JoCoding who makes coding videos that everyone can try. I have a special guest today. It's LEE YOOHWAN, the Grandmaster of Kaggle, who is top 0.1% tier in AI field. -Nice to meet you today. -Hello, nice to meet you as well. Nice to meet you. I've prepared some questions our subscribers asked on my community. Please briefly introduce yourself. Hello. My name is LEE YOOHWAN. I research and try to service AI field at Kakao Brain. You're working for Kakao Brain now. Right. Where were you before? I worked at Korea Atomic Energy Research Institute. -What did you -Yes. do there? There is a technology called 'Anomaly detection'. It's detecting abnormal symptoms in advance. I researched in that field for safer use of atomic energy. I've also researched combining visual information with our linguistic information to answer some questions. What do you do at Kakao Brain? It's actually just my second day at work. AI would learn from data given by us. How it learns is very important. So I'll probably research in that field. I've heard that you are in researching field at AI, Yes. -I've also heard that field is very diverse. -Right. Very diverse. There are data scientists, data engineer, and researcher as well. -Right -How can we differentiate them? It's a good question and a hard question to answer. It's difficult to differentiate clearly because it varies by companies. And it varies by countries as well. When there are research area and the product area, it varies accordingly. Also, how much of coding is required also differentiates the fields. The scientists who research are developing the core learning method to AI and figure out how to solve some troubles. Engineers would solve some troubles as well but scientists can talk about it related to research. Engineers would approach from the product side. There are actually research engineers as well. So, they think about how to bring what they've researched into the product. And there are data engineers as well. They would prepare piling data and decide whether to give to the production or researchers Fields covered by data scientists are very massive. It shares work with data analysts as well. And some of the research area and contribute to product. -They do everything then. -Yes. It can't be clearly separated. But when the corporate gives out rough plan about what to do, it can be somewhat categorized. You can refer to that as well. It's the job where you need to do a lot. Right. From the core technology of AI, Yes and to the front use of AI. You know very well. -You are in the research area right now, -Right. How's the working style in that field? If I think about my former job, It's rather flexible. Research is something that overcomes the limits of time and space. For example, if I come home from work, Let's say that it takes me about 40 minutes for commuting. If I think about it during that time, it's also part of the research, right? Or I can come up with some idea while doing an interview, I'll try to focus on the interview though, Research was like that to me. That's why I said it's somewhat flexible. There should be some advantages of research field in AI. -It's about learning and you have to wait after pressing the enter key. -Right. So we have that spare time. Short time. How long does learning process usually takes? It varies. Smaller models will end in few seconds. Larger models can take up few days. There is no set time for it. Right. -But we have to keep thinking about it -Right. It can be a tough job. Right. It's more like task oriented. Task oriented. Until we solve this problem, we have to solve this no matter the ways, time, and means. How is the demand in data science and AI fields? It's massive. For example, if you look at LinkedIn sites, there are lots of job posts in the related field. Always coming up with new jobs as well. Do people must have Master's degree or Doctor's degree to be in AI and data science field? Is that right? I always say this. If you are skilled, nothing matters. You can go everywhere. That's set for sure. Usually, Those degrees are about researching. If you want to work in research field, you need some experience. Right. So, when I talk to people who ask me those question, I always ask back. What do you want to do? What are you really aiming for? It's better to think in that way. If one wants to write a nice thesis related to AI, or make a new model and new methodology, Then, going for degrees would be nice. If one wants to actually develop something, or if one prefers engineering field, Starting right away would be better. It's important to find what one wants to do first. There are some people who didn't majored in this field but trying to switch over since there are more demand in the field. Right. How should they prepare? Well, we should clarify the meaning of 'Those who didn't major in this field'. Those who didn't major in this field. Normally it refers to those who didn't major in computer science. -I actually majored in chemical engineering as well. -Didn't major in computer science. Then, should I be called as so? I got my doctor's degree in chemical engineering but I am still somewhat related to this field. Reason why I'm talking about this is because people think that they won't be able to utilize their major. IT companies are leading the AI field at the moment. It's applied to all industries. Which means that you can combine your knowledge into this field. So, I ask a question to them again. What do you want to do? If you want to be in the core IT field, you should learn that field. This AI field is not about AI only. What will we develop with? -With computer -Right. Then, we have to learn about Linux, Bash, Server management and memory management. Which all leads to IT field. So, you have to know about not only the AI field but also IT related fields as well. If you want to utilize your major into it, then the field of study would shrink a little bit. Since we have nice tools nowadays. Then, how long would it take for them to switch career to work as data scientists or some jobs related to AI? At least half a year? At least half a year. I've been studying AI for about 4 years now. 4 years. I've studied hard in that long period. So, it's not that easy to switch career. It's not something that can be done in several weeks. If someone did so, I'd like to meet him and ask him about what I've done wrong. What did I do wrong to take it 4 years. -One would need at least half a year of hard learning to switch their career. -Right. Then one can have a possibility of working in AI field. You've put together my words very well. -Right, hard learning -I like that skill. Hard learning. Statistical math would be important -in data science and AI field. -Right. -As well as coding -Yes. Which one do you think is more important? Which one... They're all important -All of them -Yes, but What I'd like to mention is that -it'll be different depending on what you want to do. -Depending on what you want to do. There's a popular Venn diagram in this field. With Domain knowledge, Computer knowledge, and Mathematical knowledge. Those who fulfill all 3 are hard to find. -When we play the soccer game you can see that stats, -Right. Studying all 3 at once will be really hard. I think one should decide on which field to study first. -But you'll have do all three in the end -In the end. Yes. I can say this for sure. You'll have to do all in the end. I've tried to avoid math because I didn't like it but I'm doing it now. -You'll encounter that moment in the end. -Math as well. I really wanted to talk about this. I've been participating in Kaggle Korea community and saw some people who majored in Liberal Arts studying for several months and gain medals from the contest. I've seen several of them so it's not something that's impossible. But when I talked with them, I realized it must've been a hard work for them. I wish people will be hopeful about it. Everyone can do it. -Half a year is at least. -At least. Any other important required skills? One would be communication. -I've mentioned about several fields it has. -Right. Several people work in one project. So we have a lot of discussions. Having a communication is very important. Some would think about preparing for data science and AI fields. Yes. Will you recommend how to study for this field? I usually go head first. I like to study several times. Then I buy a beginner's book and read it several times. And I move on to a little more difficult book after I understand the book. After doing this 3 times, you will definitely gain confidence. You can't just read a book. You need to practice. That means your own project. In my case, it was a competition. Study from the easy ones, but the most efficient way is to go with theory and practice as possible. Is there a book or something you would recommend, a lecture or a book? The first lecture I studied was Professor Sung-Hoon Kim's Deep Learning for Everyone. And then Lee Yoohan. I'll stop here. There is a book called Machine Learning Using Python Libraries. Even though that's a beginner's book, the content is vast, full, and really good. The practice is also well done. -After reading it once, you will have a sense of what to read next. -Will we know? You will also know if this is for you or not. This is not for me. Then you have to make a decision. If you want to make a model well and it's the deep learning side. Then you need to buy that book. There are so many books for deep learning. Hands-on machine learning, founder of Keras. There are a lot of machine learning books. But what kind of framework will you choose? Whether you're going to use TensorFlow or a Pytorch? Then that's what you need to study. But if you want to study the data, then you have to study data engineering too. Actually, I don't know deeply about database SQL. But you can get information in other people's blogs That will be helpful. How can we differentiate between machine learning and deep learning? I will explain it in a simple way. Imagine there is a big circle called 'machine learning, and deep learning is inside that circle. In the past, when we made a machine do something, we typed, 'if + action'. If you keep typing the conditions, the indent gets long. That's how it worked. Then people decided to create something that could create conditions through education. In this way, you can learn based on machine learning and data to create something from it. Then how will we get there? Right. We can simply draw a line. A simple line. Or we can draw a curved line, too. This is about algorithmic. One of those methods is Artificial Neural Network. Let's create something using the reaction and activation of human neurons. Deep learning is like a deepened neural network. It will be easy if you understand it that way. Let's talk about Kaggle. -It's my favorite. -Your specialty. I feel more relaxed now. It feels like home. You are a Kaggle master. You must know about Kaggle really well. But there must be many people who don't what it is. Can you tell us what Kaggle is? It is a game using data. It's a game. There is a ranking system. In LOL, it's called a 'tier'. Kaggle is a machine learning competition broadcasting platform. It’s a company that holds competitions instead? It was founded in 2010 and then acquired by Google in 2017. There are many elements of Kaggle, but the most important part is the competition. It's competitions. So, companies provided data and prizes, and with this data, we gave this quiz. That's it's a competition to create artificial intelligence that can solve this quiz. And the prize is given to the person who made the best artificial intelligence. You are the Grand Master of Kaggle. Right. -There are only four people in Korea? -Yes, only four people. That's great. It's 0.1% globally, only four people in Korea. Do you know why it is 0.1%? I heard about 5 million people have subscribed to Kaggle. However, if you participate in the competition even once, you can earn ranking points. That means about 150,000 people have the ranking points. My rank is 60 something and that's why people say I'm in 0.1%. It's too much pressure for me to be honest. It sounds like I'm boasting. What are the qualifications to be the grandmaster? After the competition, we get points depending on the rank. But medals are usually given in percent. So, within 1% of the total participants, They would give a gold medal. But to be a Grandmaster, you need 5 gold medals. You can do Kaggle by yourself or as a team, but you need a gold medal that you won by yourself. You need 5 gold medals in total and must have a gold medal won by yourself to be a Grandmaster. Wow. Unbelievable. I have someone that great here. -You are greater. -No. It's an honor for me as a beginning YouTuber. We'll talk about YouTube as well later. Okay. You must have lots of prizes then. What were the typical questions and is there anything left in your mind? First thing that comes up on my mind is when I won the first medal. It was predicting the characteristics of a molecule. Molecules have something called coupling constant J. I majored in chemical engineering. So, I got my first gold medal from there. I came in third. And I got 3 consecutive gold medals. The second one was done by myself. Right, I've cried when I won that medal. Then, there was COVID 10 Vaccine contest last October. We here a lot of mRNA vaccine on the news. The weakness with this is that it's too weak. RAN would keep collapsing. So the issue was trying to prohibit the collapsing and transport well. Problems you see nowadays is from wrong storage method and how to solve that problem is a big challenge. It'd be nice to predict with AI. Vaccine is about designing. It's really effective but there are some that would collapse easily. If there is one that's effective and won't collapse easily, we should making that one. So, there was a contest to find that model and I won gold medal by myself. I cried again. Kept crying. Because I thought about those hard days staying up all nights. It must've been like Olympics. People would participate from all over the world -and winning a gold medal there by yourself seems really hard -Right. It was really worthwhile for me. -Since COVID 19 was a big issue as well. -Right. Will the model you made in the contest be actually used at those companies producing vaccines? I'm not too sure on that. The contest was hosted by DAS lab in Stanford University. There are several contests at Kaggle and that one was a research contest. It's classified as a research. The purpose is really nice. Those who loves data from all over the world would research about this topic for a month. -It's awesome -Yes. It's like crowd based research. The methodology that was submitted is all open to the public. The best thing about Kaggle is sharing. People would voluntarily open up their source code. Won't they have looked it up? I'm not too sure. But it was meaningful and I keep having interest in the field even after the contest. I've heard that the hosting organization is proceeding with follow up research. Really? Yes. I've heard from my friend who is working in the mRNA industry that the topic itself was very revolutionary in that contest. He was surprised at how much AI can be adapted into such fields. It can be said that you helped out in ending the COVID 19. It's nice of you to say so. I really hate COVID 19. Seriously. I got married last year and I went through hard times because of COVID 19. I had that anger. So I kept thinking about how to put end to it. Then, I saw this contest and thought I should participate. Since I've studied AI for 3 years. I thought I should contribute whatever I can. It was really meaningful. My wife really liked it as well. -Must have been proud of you -She let me do what I wanted to do and I just kept working with codes. She forgave me for being in front of the computer throughout the weekends. She complimented me. I was really happy. Thank you. The end of current pandemic will be faster thanks your effort. Lots of people are going through hard time and I felt good about being able to help them out with what I can do. Kaggle means a lot more than just another line in my resume for me. Right. Doing something with AI I think it's a chance to try that I can do something meaningful, and it's fun. It truly is. - So Kaggle holds a contest. - Yes - And gives out reward money too. - Of course. How big is the sum? It varies from contest to contest. - $500K - $500K. - So that would be about 600 million won? - 600 million won. There are contests like that. - And usually they're at 25? 30K dollars. - 25K dollars. About that. So how much have you won? I've placed third twice. 3rd twice. And when I came second place, I had 5 people in the team. So dividing up the money between 5 people left me a small sum, but with that money, I paid for my dorm and bought a new laptop. It's very constructive, right? Do you get a lot of offers? Yes. A lot. Really a lot. If you came first place in a contest, then it's really good, but it's hard to say how. - So if you win a contest and you get a lot of offers, - Yeah. How big would your salary be? If I'll share some public information, - It'll start in the ten thousands - ten thousands and sometimes 6 digits. - 6 figures. - Yes. - In dollars - Yes. Really. So at Kaggle, what can beginners try? It's a machine learning contest, So it would be better if they joined after reading some elementary books, some easy books I shared before and get used to machine learning. If you look, Kaggle has some tutorial contests too. Like the Titanic contest. But the tutorial contests, anyone can do them. But the actual contests are really hard. But those are what you need to do if you want to build your skillset up. So do the tutorial intensely, shortly. - Shortly - Yeah, just intensely and participate in the real games to learn about what you lack. It'll tell you cold-heartedly by giving you a rank. If you get 700 out of 1000, then you're like, oh I should study, or not. I'm curious. How long did it take you to start Kaggle and win your first prize? How long did it take? I think about 7-8 months. - You tried for 7-8 months. - Yes. - So you work as one of the admin staff for Kaggle Korea. - Yes. Can you introduce what you do? Yes. So it's a non-profit Facebook community. And I really want to accentuate the name. When you say Kaggle Korea, Kaggle's the company and Korea implies that it's branch. But we're not an official branch. But we didn't use it without permission either. We contacted the Kaggle HQ and told them Korea has these groups, and most of them add Korea in their names. And we asked them if we could do the same, and they said yes. But run it as non-profit. So we're a Facebook group mainly about Kaggle, and because Kaggle deals with deep learning, machine learning and AI, we cover all those topics too and we're a non-profit community. What type of activities take place in the community? The slogan we put forward is 'a community that studies'. So we study. - We ran Kaggle study groups. - Study. When we were at our peak, we had about 23 groups running. - 23 study groups. - Yes. And we co-hosted a contest with Google Korea. But because of COVID, there are little offline activities these days. Yeah, so it's been hard lately. And our team has been discussing a lot amongst ourselves. What could we do? How can you join Kaggle Korea? Just search for it. If you request to join, one of our admin staff - will accept - your request. Are there any qualifications? Qualifications? I don't have any, so anyone can join. - Oh, there is one requirement. - What is it? You have to have internet. - So, Yoohan, you work in AI. - Yes. Does it mean you have a really good computer? Yes, because you need a computer. I worked for a company, so I usually used the company server. Not my personal one. When I worked in a lab, I used the lab computer to study so being a server computer, the specs were really good. And the CPU and the code was top-tier too. Not the Intel CPU that most computers have, but the really good ones. - Cores for calculation. - Yes, cores for calculation. I used laptops with those. And the gadgets you plug into the PCs, I also used really high-quality gadgets. Because you need GPU calculation to create deep learning models, I needed to have a good GPU. So I used 1080Ti and the company server had v100 or Nvidia products that were good. So those who want to study AI would be really interested, - but graphic cards and other parts tend to be expensive. - Of course. For those who can't necessarily afford it, Will there be an alternative? Yes there is. Your questions are so good. For example, there are programs that loan equipment, run by government agencies such as NIA and NIPA. - Rent? - Yes. But you need to write an application. And a plan on what I intend to do. Then Google Colab. - Colab. - Colab. Colab is good. I think you can use it if you pay about $10 a month. If you use the free version, I think the runtime has a limit. I heard Colab was really good. - If you use the paid plan. - Yes, the paid plan. So I recommend that. Is it enough to train your AI? I mean the more, the better. But recently, there was a Korean who beat that. There is someone who came first place in the recently completed Riiid contest or the Landmark contest of last year. I saw him during Riiid, and he used Colab. - With just Colab. - Just Colab. I really reflected on myself. People like him exist. I think more equipment, the better. More you have, more you can experiment, so it's better. I mean that's right too. But it gave me chills. I respect him so much. He's amazing. Kaggle is ranked 7th in the world. - 7th in the world? - 7th. - That's amazing. - He is a great person. You know, they say experts don't blame the tools. I'm not an expert. - I'm - No, I didn't mean that. Of course, bad carpenters will blame the tools. So for a beginner who's just started playing with AI, Colab will be enough? Yes, it'll be enough. - You can be the first in the world using that. - You can be the first in the world. I heard that in the field of AI, people share what they've learned with other people. Yes. Do you have a platform where you share your insights? Well, there's Kaggle Korea. There's a way to share your own code and thoughts within Kaggle. You can get a medal for sharing. - For sharing? - Yeah, you get likes for it. So there's that. And I've been running a small YouTube channel. So there are videos on the tutorial contests or Titanic that I've shot. And recently, basic things like Python. - I'm married, so with my wife. - With your wife. - She's in music. She covers music and child raising. - She's never done coding before. - Oh, so for the first time. With people like her, - Showing them that they can do it too. - Yes, so going through it step by step. It's fun. It brings you closer. I watched it too, you guys were so sweet. It's a cover. That's real life. We're usually very sweet with each other. Alrighty. So for those of you who want to learn python or Kaggle, you can watch videos on Yoohan Lee channel. We'll link it. - I also have another question. - What? The future of AI. One of my living role models is Elon Musk. And iron man, who's now passed away. I truly respect them. And seeing them, I picture Jarvis happening. - Jarvis. - Don't you think so? How fun would that world be to live in? I really hope that if you have to build an AI, people dreams of that and works towards that. How helpful would that be? It helps humanity. I think many of our viewers today would be interested in becoming data scientists or work with AIs. Can you give those people some words of advice? I want to tell them to not give up. There's something that I've said which I really like. My award is everything I've learned. I studied for three years doing that. Winning money is good. And I used that money to pay for my dorm or buy a laptop. But at the end of the day, what's important is that me, as a person, has the skills and I've built that up. So sometimes, the road ahead of you seems to stretch forever and you can't really see where it leads but don't give up. If you really want to do this, then you should be expecting that. I really encourage you to try everything. Try everything for 6 months. Even if that turns out to be something you didn't want, that 6 months can change your life. Yeah. If there's anything else, you have a lot of subscribers so you must have questions? Please leave them in the comments, and I will reply. - Thank you for being part of our interview. - Thank you for inviting me. If you found this helpful, please like, subscribe and click the alarm button. I'll return with a more helpful video. - Thank you. - Thank you.
Info
Channel: 조코딩 JoCoding
Views: 164,081
Rating: undefined out of 5
Keywords: 캐글, 인공지능, AI, 딥러닝, 머신러닝, 데이터사이언티스트, 데이터엔지니어, 데이터사이언스
Id: tu6b3xbTj6M
Channel Id: undefined
Length: 32min 58sec (1978 seconds)
Published: Sun Feb 21 2021
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.