클라우드 컴퓨팅: 비즈니스와 AI의 혁신

### 클라우드 컴퓨팅의 진화와 영향
* 🌩 클라우드 컴퓨팅은 **2006년 Amazon Web Services(AWS)에서 컴퓨팅 리소스를 서비스로 제공**하여 리소스에 쉽게 액세스하고 종량제 요금제를 사용할 수 있도록 하기 위해 도입한 서비스입니다.
* 💰 AWS는 **400억 달러 이상의 매출**과 전 세계 수백만 명의 고객을 확보하며 컴퓨팅 업계의 리더로 자리매김했습니다.
* 🛠 컴퓨팅에서 머신 러닝에 이르기까지 다양한 영역을 아우르는 **약 175개의 광범위한 AWS 서비스**가 성공의 핵심 요소입니다.
* 🌍 AWS는 **24개 지역 및 가용 영역**으로 구성된 글로벌 인프라를 보유하고 있으며, **5G 네트워크 사업자와 협력**하여 더 빠른 서비스를 제공하고 있습니다.
* 💡 클라우드 컴퓨팅은 다양한 산업을 혁신하고 **기업이 신속하게 서비스를 출시**하고 경쟁 우위를 확보할 수 있도록 지원합니다.
* 🏗 클라우드 서비스는 고객이 서버 관리 없이 애플리케이션 구축에만 집중할 수 있는 **서버리스 아키텍처**로 진화하고 있습니다.
* 🍕 소규모 팀이 서비스를 개발하고 운영하는 **Amazon의 '2피자 팀' 마이크로서비스 아키텍처**는 민첩성과 가용성을 향상시킵니다.
* 🔄 클라우드로의 전환에는 **새로운 기술을 채택하는 것뿐만 아니라 조직의 문화와 구조를 바꾸는 것**도 포함됩니다.

### AI, 로봇 공학, 우주를 통해 미래를 만들어가는 Amazon의 역할
* 🤖 아마존은 이미지 분석, 자동 음악 작곡 등 다양한 애플리케이션을 통해 **"흥미로운 방식으로 인공지능을 배울 수 있는 기회"**를 제공하고 있습니다.
* 📦 아마존 물류센터에 로봇이 도입되면서 운영 효율성을 높이기 위해 **Kiva와 같은 로봇이 널리 사용**되는 등 큰 변화를 가져왔습니다.
* 💻 Amazon CodeGuru와 같은 사기 탐지 및 코드 검토를 위한 Amazon의 AI 학습 모델은 **개발자가 더 나은 코드를 작성**하고 비용을 효과적으로 관리할 수 있도록 지원합니다.
* ☁ AWS는 무료 티어에서 **머신 러닝을 포함한 다양한 클라우드 서비스**를 제공하므로 인공지능 연구에 관심이 있는 사람이라면 누구나 이용할 수 있습니다.
* 🚀 아마존과 제프 베조스의 우주 기업 **Blue Origin은 우주 탐사 비용을 절감하는 방법**을 모색하고 있으며, **Blue Origin은 우주에 대한 접근성을 높이고 영감을 주는 데 중점**을 두고 있습니다.
* 🛰 무게 77g, 한 변 길이 10cm의 소형 위성인 큐브샛은 **스타트업과 개인이 우주 산업에 혁신적이고 영향력 있는 기여**를 할 수 있도록 지원합니다.
* 🌍 "클라우드로 인해 비즈니스 창출 능력이 크게 향상되었습니다." **AWS는 기업과 개인을 우주 산업과 연결하는 데 중요한 역할**을 하고 있습니다.
* 실제로 산업 현장에서는 많은 로봇이 사용되고 있으며, **AWS RoboMaker는 로봇을 쉽게 만들고 운영할 수 있는 클라우드 서비스**를 제공하여 기업들이 효율적으로 프로토타입을 개발하고 테스트할 수 있도록 지원하고 있습니다.

클라우드 컴퓨팅은 쉽게 액세스할 수 있는 리소스와 서비스를 제공하여 비용 절감, 민첩성, 비즈니스 혁신을 가능하게 함으로써 IT에 혁신을 가져왔습니다.

Yes, thank you. Hello, I'm Yun Seokchan, working at AWS, a representative band. I appreciate being invited to such a great place. Today, I would like to introduce the business innovation brought by cloud computing and the future of the underlying technology. I am still with AWS, which stands for Amazon Web Services. I will talk about Amazon's cloud computing business unit. We actually started this cloud computing business in 2006. The reason we started the business is that many companies found it very difficult to purchase servers, place them in data centers, and operate them when providing internet services. Even now, many companies are still doing the same thing. However, in the case of Amazon, while running an online shopping mall, they realized that this was quite difficult.

Managing and operating servers efficiently is very important due to the large amount of traffic coming in. Many customers are experiencing the same issues and are talking about the need for faster operation in setting up and running equipment. Developers want to utilize resources at a low cost, quickly develop something when needed, and are thinking about exploring new possibilities. Therefore, the idea is to turn existing physical equipment like CPUs, memory, and storage into computing resources as a service to sell. This is where virtualization technology comes into play.

Although it existed before, turning this into a business from something expensive is essentially the first time at JW. So if you look here, no one has called it EBS Scars. That is our service name, and now we are selling it by dividing it. So, what we have done is we have made it into a concept to sell. So, cloud computing is a large-scale service that allows us to easily access resources that were difficult to utilize in the past, and use them whenever needed. Also, the cost of using it is not like spending a lot of money upfront when buying a server. You only pay for the time you use it and the amount you use. It's like utilities such as water and electricity in a way. Thankfully, after about 15 years, this business has been doing well, and it seems to have matched well with the trends of the 4th Industrial Revolution era. We have been meeting many demands, and last year, we made around $400.

In terms of revenue, you have also brought about economies of scale, probably by using AWS. This is because of the success of millions of customers worldwide. Therefore, we are currently playing a leading role in the computing industry. One less noticeable thing is that one of the main reasons for the success of AWS is the wide range of services it provides. From CPU, memory, and hard disk to login, when I started with just 8 services, now after 15 years, we have a large number of services. From mobile development to big data analysis, IoT, machine learning, in every area, all the functions and services we think of are built into our modules. Currently, we have around 175 services, so many that even I may not know all of them. So, which services are

It's very important because there are so many, not only things that everyone can easily use, but also a lot of infrastructure worldwide that needs to be operated together. It's not just about developing services, but also providing services. We have 24 visions globally, and in each region, we gather multiple data centers to logically prioritize one, making it easy for users to deploy applications and ensure high availability. We have availability zones to help with that. In the case of Seoul 2, we launched it in 2016. What's interesting is that we have not only transformed these data centers globally but also, recently, with the importance of 5G, we have partnered with external 5G network operators to provide faster services.

We are going to launch a web rent service soon, which is aimed at providing cloud services together. We have partners such as NTT, KDDI, and now Vodafone, as well as SK Telecom in Korea. Many companies consider it very important to connect the cloud when they have a bridge or a fact. At that time, you can directly connect with a dedicated line. People think of the cloud as just floating on the internet, but it's not like that. You can attach it like a plug, so we have 97 dedicated line connection points called Bridge Connect. We also have over 200 CDN pops worldwide to provide CDN services. When distributing content, if there is a CDN pop near your internet, you will receive the content faster. So, it's important to know where the pops are located.

You can receive videos or images much faster. So, we are directly receiving over 200 of them. These are set up with a network of over 100 gigabytes. We are setting up the evening network. If you watch for about five days, you might think, Wow, this innovator not only provides cloud services, but also has a scale of economy that is very impressive. What is cloud computing? Actually, in the past, you had to invest money in advance, but now you can use IT resources anytime, which is an advantage. Secondly, because there are about 170 services that I showed you earlier, you don't need to buy software or launch projects every year to create something new.

Even if it is only 10, there is an advantage that can be used. Also, as a result, you can save costs such as managing IT resources, purchasing hardware, and so on, so you can focus on your business. Through the global infrastructure I just showed you, you can easily access applications from anywhere in the world while sitting in Korea. These are the advantages of cloud computing. These not only save costs but also have a significant impact on the speed of the business. So, briefly, I introduced the current state of cloud computing, and now I will briefly introduce how cloud computing is evolving in terms of graphics, especially how it is influencing the field of artificial intelligence, and how cloud technology will develop in the future. Let's continue to observe.

I will do that. First of all, let's take a look at how cloud computing is currently impacting and influencing various industries. As many of you may know, the IT industry is now breaking down the boundaries of various industries simply with cloud computing. There are various players who are disrupting traditional industries. One of the reasons for this is that these people can easily and quickly make investments that others could not with the tool of cloud computing. Therefore, from retail to accommodation, broadcasting, finance, real estate, and dining, the companies I have shown here are all companies that are based on the cloud and operate cloud-native businesses.

Existing competitors in the industry can be caught at any time. That's right. Companies with traditional people have never seen such competition before. The biggest reason for this is cloud computing. Let me give you an example with MarketCully. In the case of MarketCully, they are in the food delivery business. They play a role in both the food and delivery businesses. In 2015, their revenue was around 29, but by 2018, it had grown to about 50. When they say they have grown like this, it means that their traffic has increased. Instead of using existing infrastructure, they immediately switch to cloud-native and configure applications account-friendly to handle user traffic.

You can immediately bring it up. Otherwise, what will happen? People will erase kids like me and games. They won't use the word cloud directly. That's why the growth of companies like this is based on the cloud. This is a great advantage even for ordinary developers. If they want, they can create and launch a service anytime. For example, in the case of making a corona vaccine, a student developer received 2.4 million traffic in one day. But if they can do this themselves, it would be difficult. Now, through the crowd, you can secure enough traffic. So, not only developers but also startups gain great power in this area. The biggest advantage of the cloud is that it brings about such changes. This is why Lime is coming forward with such changes.

Global enterprises in various regions around the world, including traditional companies with a long history, are planning to go digital. Companies in diverse fields such as finance, automotive manufacturing, and agriculture are already utilizing the cloud. This has been happening for the past 10 years. For example, Netflix, a representative case, currently provides video streaming services worldwide solely on the cloud without their own data centers. Many companies are using services in this manner. Looking briefly at how these companies have transitioned, they have been aggressively moving towards this direction for a long time. For instance, in 2015, Netflix reduced the number of their data centers from 4 to 3 within three years and successfully migrated to the cloud, surpassing the 60% target.

In the case of companies like Coca-Cola, they also convert to Unix Linux or finish a 14-month project and move tasks like reducing data centers and moving everything to the cloud, such as News Cup or Gradian. I would like to mention one good example, which is the company Variation in Jeju. They created a budget for making airplane engines. They are a manufacturer. Where do customers mainly go? Well, it might not have been visible before, but now it seems. However, these companies originally learned to manufacture engines and then just did it. They always provide technical support from the engine to the assembly. Now, after starting digital phones, the business of this company has changed. Now, they are uploading many sensors that were in the engines to the cloud. And this company, which runs these three engines, may look like Airbus or Boeing, but in reality, they are the companies that provide services.

Airlines, you know, four airlines around the world send engine data every time they take off and land their planes. So, they send that data to the cloud for analysis. By doing this, they can identify if there are any issues with the engine, fuel efficiency improvements, or even potential safety concerns in advance. This is why they are engaging in a scrap fashion business. So, in reality, their business used to be more focused on manufacturing in the past, but now they are in the service business. Of course, they are still making engines as they did before. Therefore, you can see that even traditional businesses are being transformed by the cloud. Domestic customers, like those using our servers, such as Samsung Electronics, have been using it since 2012, and this mobile app is used worldwide.

Services are provided to Proud based on scripts, and now they also provide smart IoT services such as saving q electronics, and companies like Korean Air are also making efforts to quickly digitize. Broadcast stations are the same. Whenever there is an issue like the general election, they set up classes and server settings, which are difficult like setting up AO servers, so they are utilizing crowds well. Recently, many financial companies are using it, and many retail companies are also adopting it. Now, in this era of change in competition and markets, not only are they learning from the past but also changing their IT 1 IP glasses to this cloud. Even if the service doesn't work well, there were times when it was okay to respond if you had a crocodile's physical equipment. There was a feeling that it would be fine if the service didn't work well, but now, you don't need that. Just move the virtual server, and with virtual servers, you can just discard them anytime.

We can discard the equipment that can be discarded, set up again, bring it up if necessary, and delete it, creating a situation like that. It's a bit more foreign, modularized, and now it's possible in the container area. Recently, it's changing to an environment called bliss. So, in fact, there are no servers, but from the perspective of using cloud services, it's now being configured to appear as if there are no servers at all in the background. So, cloud services have three main areas. If we classify the services, the first is the ingredients that make it easy for us to cook. These are the services that provide ingredients, which I call ingredient primitives. But now, for those who are good at cooking through these ingredients, they may want to try something else, not just cooking.

I want to eat something that comes out as a product, so now we provide a service that suggests forget about the worries of the Buddha in a way. This is provided to a certain extent like a semi-finished product. Then, because it's difficult for me to manage this, I just want to do a program product. I just want it without the part of management afterward. If I say I want it like this, then Bliss gives the option to distribute it as glasses. So, earlier I mentioned that there are over 170 services, broadly divided into services in various areas such as computing, storage, and all general primitive services. Then there are the Pullum Annie's services and serverless services. So, customers can choose according to the workload they want at any time, making the number of services gradually increase.

You can choose anytime, so in the case of various companies like Samsung Heavy Industries, they are currently working on autonomous ships in Jeju. To provide well-functioning ship services globally, what should we do? If you look here, you can see icons now. We provide services to select the desired functions and assemble them like building blocks or Lego blocks, to create a service that runs in the company. Then, there are services provided by shipowners, followed by service S. We manage these services. You can also structure it like this. Another example is that I am exporting air purifiers to the United States by the company, and if I want to own IoT services here and talk to Mal X to create things like turning on and off through the server.

Actually, Coway is not an IT company, but in order to distribute these things faster, they utilize services based on the server Alice here in various colors, which enables all services to be served. They can configure it to run smoothly for eco-offroad by utilizing IoT services like $5 ODB Lambda. Although it may be controversial, they have to create a bouquet of direct deliveries. Changes in the IT environment like this are not just about changing IP, but actually, it involves creating services and changing services together. One company that has experienced the drawbacks of this is Amazon. When Amazon creates a shopping mall, all functions are not cut into one piece and created that way. Instead, they create it in a way where all these functions are cut into one piece and created together.

Every time we update the functions, it gets slower because we have to wait until yesterday and so on. It keeps getting slower, the deployment gets slower, and the experience of the service going back and forth to the kids gets worse. So, we make a very bold attempt. In 2002, we divided this service into pieces and then separated them, allowing each service to be run by a team separately. This gives complete autonomy and independence to the service. That's why we call it the 2-pizza team. Amazon has all 2-pizza teams now. The meaning of 4 teams is to operate the service by forming teams of such a small size that you can have a meal with two pizzas. So, how are we doing now, Mike? We have these services.

There are thousands of them, and when you go to Amazon sometimes, you might think, There's no 1st place, but that's not it. On the first screen, hundreds of small microservices composed of APIs are called and displayed all at once. So, currently, services are structured and operated in such a way. This is a very suitable development approach in the cloud, right? Therefore, the concept of microservices architecture is to separate things that used to be developed as a whole into services like this. For example, in the case of an online shopping mall, there are services like a cart service, order service, etc., and each service is structured by its own team for development. This is what is currently referred to as the microservices pattern. This is how things are done in the cloud.

It helps a lot in developing by utilizing it well like this. So, when creating like this, which development cycle is the best? There are many methods that can help with that. When splitting this service, each team can develop independently. This increases agility, and then, even if a small part of the service goes down, the entire service does not go down. So, it definitely becomes a service with high availability. I would like to introduce you to something called the The Rule of 4/4/4 which is, no matter what kind of IT system you are building, it closes the way of communication within the company. What does this mean? Every year, we put in new technologies, launch new projects, and create something new, so this is very efficient.

It may seem like going back, but the actual way the communication channels in a company's four organizational cultures move, and capturing them in the same way is quite difficult. So even if we do well with our cloud and phone communication, if the culture or atmosphere within the company is not fixed, the system will still exist in the old ways. Therefore, the transition to the cloud is not just about introducing the cloud, but also about bringing about changes in the company's organization. I would like to emphasize this point. So, for those who want to study a bit more, I recommend looking into patterns that can create new modern apps, such as Michael services, serverless, and so on. Second, let's take a closer look at how cloud computing and artificial intelligence are influencing each other.

Every year, we hold a conference called Re: Malsseumi in the Amazon, where Dr. Nato Messi Tom speaks. He mentioned that the golden age of machine learning and artificial intelligence is being driven by the cloud. Many people are adopting the cloud, which is bringing about significant changes in technology. Issues that were not easily solved in the past, such as image pattern analysis and speech recognition, are being addressed, thanks to technologies like Beam Non. A representative from Amazon.com came up to talk about product prediction and demand forecasting, comparing it to traditional methods.

It's a big improvement by a factor of 15, which sounds like a lie, but in Las Vegas, I met with thousands of people on-site, and there is no exaggeration. In fact, it has now improved by 15 times. One of the biggest reasons for this is when you look at Jeff Dean's slides, in the past, the way they used to approach it was through a method called the neural network approach, which was less efficient than the traditional machine learning methods. There are two reasons why it has improved. One is that the computing power has improved tremendously. So, the computing capacity has increased enough to analyze this data, and the amount of data that can be trained has also increased significantly. In fact, artificial intelligence, or what they call EAI, has been around for about 70 years, but it was not able to advance as much as it is now due to limitations in computing power.

Meeting capacity and data has made a significant improvement. If you think about it as a chess match, in the case of AlphaGo, it made a move on the opposite side of the chessboard where thousands of servers were combined into a computing cluster to train the model. If you were to use computing power, you would need similar computing capacity in terms of cost, such as CPU, memory, and GPU. NVIDIA has created a top-notch card called V100, which is commonly used in AI. Now, you can use it on Amazon EC2. You can use it with server specifications called P3 instances. If you were to purchase it individually, it might be a bit challenging, but roughly speaking, you can buy one for around 13 million won.

But this is not a personal expense to be paid by individuals. If you move this to Seoul in ep.3 SRS, you can use it for 5 balls per hour. This roughly produces computing power equivalent to about 100 units on average. Now, if you need the computing power for 2000 videos, it would cost around $100 when multiplied by 20. This cost is practically negligible, but now, providing a bit more computing capacity allows the model to learn faster. So, by using a lot of computing power and GPUs, the model's learning speed gradually decreases. In this case, we have much more computing power than you might think. Therefore, we have some spare capacity on both sides.

Operating sheep is a bit manufacturing, so it allows customers to use them at a low price, even if they say they can't do it on demand. If you intend to use about 70% of the resources, you can use the computing power with 2000 performances per hour for 30 minutes. And as I mentioned earlier, the more you use, the less time it takes. So if you say pass, the starter company uses this cloud and now we have the most popular artificial intelligence model. During the lesson, when learning, we used a lot of p3 instances by using a lot of clusters, and in just 18 minutes, we pulled the model up to 93%.

It's about raising it, but the problem is that, in terms of what other companies are doing, the actual problem with EB usage is that it could cost up to $40. This is a very big deal. You can easily utilize cloud computing to perform supercomputer-level calculations. These kinds of services not only use server capacity but also provide over 30 different services based on a leader base, using artificial intelligence to provide services. If you expand on this a bit, as I mentioned earlier, it can be divided into nine different areas. If you are a researcher who can use server computing and GPUs well for learning through ProLine, you can use GPUs and test Pro to do so.

However, they said it was a bit difficult to manage such a burden. We, as Seize Maker, completely blew away the service. Earlier, I mentioned the complete service again. Through that, we can save more time. No, it's not that. I want to use only the necessary functions as needed, such as image document processing, voice analysis, fax transmission, book search, personalization recommendations, fraud detection, code reviews, and so on. If you want to use an artificial intelligence service with a balance of 100 different features each month, you can still choose and use it. The service called Money Size Maker really allows you to handle services from model data preparation to model training all at once, like a one-top service, making it possible for you. Oh, like a star. And in the case of GS Caltex, now they are recognizing the vehicles entering this gas station.

Whether a person is refueling or getting a car wash, there are things like that. After the blue moon, now we utilize it for customer marketing. However, as you know, GS Caltex is actually a gas station, right? So, they were good at IT services, not at sales. However, by utilizing things like Sales Make, they can easily have the model data and learn from that model to distribute and provide services. Also, as I mentioned earlier, if individual services are needed, you can choose various services, and among them, the most effective service for business is personalized recommendations. But, as you know, Amazon, for example, has been providing online shopping services for a long time, so they are very good at things like personalized product recommendations and predictions. So, on Amazon.com,

We have simply applied the technologies we use to the services, so now it's called Any Jumper Sunrise 1 Service. In fact, you don't need to do anything inside here. Just input the data, and the results will come out immediately through the API. It's so easy to set up a personalized recommendation service. For example, in the case of Domino's Pizza, we can extract data such as the amount of pizza culture lovers, and then things like weather, matches, and locations, and put them into an Excel file format. By doing this, who likes pizza and why, it all comes out immediately. So, it was easy to implement. Also, when it comes to predictions, it's the same. We are now providing a service called Forecast, using the same technology that Amazon uses for demand forecasting or outing predictions, like in the case of Korean companies such as CJ Logistics.

I also tested this case, and now, based on the existing elliptical map predictions, we have a system that is almost different when compared. The performance is good to that extent. We received evaluations like this. If any of you find this a bit difficult but want to start and have fun, we provide opportunities to learn artificial intelligence in an interesting way. For example, when doing things like image analysis, after the lens, there is reinforcement learning that helps us in four days. Next is the analysis method called 'Gye'. When using that, there are things like automatic music composition, such as Beethoven. Here, I think many of the viewers are male. This is within the team.

Riders wear two cameras and race. In fact, now at Offline 5, we make models like this and put them on the track for autonomous driving through the track. However, it's not like just buying this offline. In our NWS team, we have a service called Gavossi, where you can create models online, test them, participate in virtual racing, and so on. If you are interested, starting like this, you can easily see the announcement. Third, then Amazon was working so hard on AI, so you could see that Amazon was good at this. Actually, stories like this are not well known socially. Other companies are working on AR.

I don't think Oh Jun is doing well. I think Amazon's shopping mall can boast of its clicks. Actually, I've been using AI since I created a shopping mall in 1995. This is the first screen. If you try to shop, there was an agent called Kukdo based on that idea. So, the recommendation agent attached to the engine that recommends books when you buy books has been continuously improved since the beginning. Therefore, in 1999, 35% of the total sales were operated through AI recommendation. How about online shopping malls that do not stop there? It is important for items to be delivered quickly. The moment you put items in your shopping cart, predicting how the items will be delivered to the user is a cool power through reviews and such, which we have been able to recommend from the beginning.

So, how to follow a certain route, where certain items are located, and until when I will go with whom, this was already predicted by Hyundai in 2014. They were making over 500 overseas predictions that week. It's not just important to make predictions like this, but it's really important to go fast, right? It's essential to have the technology to make it go fast, and one of the representative examples is trying to use robots in the delivery center. Through this, there have been some significant changes in the logistics center that we couldn't have imagined before. I will now show you a glimpse of an Amazon logistics center. Inside an Amazon logistics center, it's quite different from what we think of as a logistics center. If you choose rice, rice is gathered like this, if it's not, it gathers in a different way.

There is a robot carrying a box called Bin and delivering it according to the order predicted by customers. You just have to sit and wait. In the past, people used to search for things, so they can quickly do the packing of legs. These robots are strong enough to be widely used in actual centers, especially after the acquisition by the Keyba system in 2011. People might think that Amazon has high standards, but in reality, new trends are emerging. For example, robots like this one come down in two minutes to check in the morning. So, I'm wondering if there are any special clothes, and then the problem.

Now, after checking if there are any clothes, once everything is done like this, yes, now it's all done. I'm the manager who gives instructions like this. There are technicians who repair and manage these things, not just the manager. And there are scholars who optimize each model in the robot for delivery centers, so they work in a state of foreign substances. When we talk about logistics centers, there are many people who think of it as just manual labor. In reality, various experts are working in logistics centers. I will briefly show you an example of utilizing AI in logistics centers. This is to demonstrate how automatic management is done quickly when robots enter the logistics center.

So, in the beginning, you take help or take a picture. So, uh, and when this one goes there and comes back again, it will take another picture. So, it's to detect if there's any change. Each time you remove one, if someone says it's automatically recorded, then it can't be removed automatically. And then it will take a lot of time. So, by using existing computer vision algorithms, we can review these pictures quickly to see how many are in the water. This works very well. But most of the time, these kinds of problems occur - when things stay still or move, when objects are placed incorrectly, or when they fall off, things like that. These are actual friction issues, so they are frequently encountered in the field. Therefore, existing high-quality computer vision algorithms...

Even if you have it, it is difficult to solve such problems easily. To solve such problems, now we use something called CNN, and the performance is very good in solving problems. So, staying or leaving, there was actually a change, and it is easy to determine whether an object is moving or not with binary decision-making. Therefore, it is possible to manage and improve the quality more easily. The interesting thing is that in these logistics centers, high-performance computing images keep coming in. If the images keep coming in, the model cannot learn, right? In that case, how do you... you upload it to the cloud, show it, train it through GPU equipment, and then go to the actual site to make it work.

In this way, even if the computers at existing logistics centers are not very good, they can provide services by utilizing cloud computing to obtain high-quality AI-trained models. We are dividing roles like this and providing services. These things are not only applicable to logistics centers but also to delivery. So, Amazon is currently doing a lot of autonomous delivery. Since 2013, I have been developing prototypes. When you continue like this, you may wonder if deliveries coming from Korea will be possible. One of the advantages of deliveries coming in is that, like in the United States, they can deliver goods to customers in remote areas in just 30 minutes. Of course, in major cities, there may be faster means, but.

To provide such means, there must be a model in front of what comes in, and there must be a model that scans all obstacles in front of the feet. This model is also essential. What I am showing you now is not 11 AT, but actually, it shows the appearances that are being operated. In 2019, we created a new beer design like this, but in order to create more of this, we need to transform 50,000 designs, and then we need to do well in the remaining 3 million scars, 81 times each. Every time we do this ration of 12 layers, we cannot afford to make it new today. Therefore, we use actual AI computing to think about a large amount of what comes in. Actually, before operating it, we use AI computing for wind tunnel experiments.

We conduct all these things and also in terms of graphics, we provide simulations of how they move in the set environment. So when we go to the actual site, we verify that they work well, validate the models, and ensure safety without any issues. Therefore, even the introduction part that is not clearly visible, we quickly catch when they come in. When users come in, they sit here, and if they don't, we don't deliver, but when people leave, then we start the delivery. You can also see the part using the 20g data that has just come in. Many of you may know, but I wanted to introduce shoe information. Shoe information is the most common problem when we go to online shopping malls; it's not shopping that's the problem, but waiting to pay when you leave.

The problem is that it is now easy to do that. It is an offline store that just takes what is done and who took it. What I want is to check without waiting. This is what people think when they open an unmanned store. It's not a monster store, it's for people who want to leave quickly. So, when we use this technology, we utilize not only computer vision technology but also various technologies such as sensing technology. In reality, through these, we are creating something called Jeotsowa Kakao Tech. When we were in beta, when we were making employees, it was the same thing. We launched it in 2018. At that time, you could see people waiting outside. You have to leave when you come out, but when you go in, since the stores are different, many people cannot come in. Now, there are more stores, so it's like this.

You don't have to wait. So now, if you look at the manners, there are hundreds of cameras on the ceiling like this. And what those cameras do is they capture people below. So, when people look down, they see dots. By observing the movements of these dots, the system learns how people move in the space and is guided through models. Not only that, but it can also determine whether a person is picking up, putting down, or stealing an item by analyzing the graph of movements. This is done through models. So, there are those who can carry out such forbidden actions and model behaviors. As I mentioned earlier, they are trained through external clouds like delivery centers and are currently being operated. Currently, sample stores in cities like Chicago, New York, and more are utilizing this technology.

Although it may be a bit difficult for us now, I strongly recommend that you visit our store at least once if you have a chance to travel. Recently, we opened a new grocery store called 로스 리 in Seattle. It's not just a convenience store, but a place to buy groceries. So, we have various types of products at this 로스 리 grocery store. When you visit a grocery store like this, you can find a wide range of fresh produce, or maybe even freshly squeezed juice. The interesting thing is that when you go to a grocery store like this, the items you pick up are automatically calculated for you, making the checkout process more convenient. So, after you finish shopping, the items are automatically calculated, and you can simply go to the green side of the cart.

Just pulling out and going, does it all end in the car? You don't need a Dr. line for this level of student. So, I think you can see it this way. So, even after I show you a lot of recommendations and predictions, if you pay based on that, does Amazon really do well with the technology they have in Europe? For those of you who are asking, I'd like to introduce two interesting things. One is what we call a radio wave detector. It prevents online fraud and helps prevent fraudulent transactions like credit card theft. Amazon is good at this, right? You're probably doing well, but we have a slightly different situation worldwide. But in Korea, the reality is that credit card theft is a bit different, but overseas, this kind of thing happens a lot directly. The cost of building a fraud detection system is quite high.

We provide this through the model that we have developed. Like the Personalized Eyes Pocket I showed you earlier, you just need to input the data, and it immediately distinguishes between positive and negative. Many services like this are being offered, especially for developers who are interested. One of these services is Amazon CodeGuru, which reviews the program code you have created. If your boss at work is not satisfied and you haven't thoroughly reviewed your code, you can use this service. Amazon CodeGuru learns from actual source code on Amazon.com and provides very detailed code reviews. This way, it helps you understand what the problem is with the code and how to operate it.

There is something really important. You need to know if this code actually uses a lot of CPU memory, for example. That's why I have a profiler. It can tell you if 2% of the code really consumes a lot of CPU memory and if it actually incurs a lot of costs. But as I mentioned, we are already using this service on Amazon's existing source. I mentioned this during some kind of Prime Day event. When we inserted these codes into our service, the performance of this code group increased by about 300% in terms of CPU utilization. It means that we were able to reduce the number of servers by a third. In other words, the performance is so good that we can reduce the number of servers to a third and also significantly reduce costs. This not only involves writing good code but also effectively managing costs in terms of actual business.

It's helpful to have someone who can do that for you. So, AWS has a variety of services across different layers. Once you sign up, anyone can use it. You don't need to ask the company to provide the source code every day. You can try out five different services, including machine learning, and they are provided for free under the freetier, so you can use them all for free. Therefore, I recommend trying it out once if you are studying artificial intelligence. Lastly, this cloud computing that influences businesses and various aspects of daily life will greatly impact the future trends of artificial intelligence. Let's take a closer look at how cloud computing, which plays a significant role in this new trend called artificial intelligence, will change in the future. I really like the saying, The future is already here. However, let's look into these aspects a bit and conclude.

I talked about those global leaders who have made the transition. They have been doing it for ten years now. But now, many people are saying that we should do it now because of Corona. Oh, that future is already here, which we didn't know. Yes, that's why we are already in the future. You may not feel it well. So, I would like to give you some examples to help you feel that time. I mentioned that we have an event every year at Amazon, where we ask people to write reviews. Last year, it was held in Las Vegas. Unfortunately, we couldn't proceed as planned. What does it mean to go from Mars to ears? It symbolizes automation, robotics, space, which will greatly impact the business of industries in the future. There are four main themes, especially the theme that Amazon is interested in.

Opinion leaders and 2 compasses provide information together at such events. Here, at the event, I explained some parts in advance about gender identity. So, I would like to explain what changes there are in the OTX and A Space sections. In fact, in the robotics industry, there are many people who don't just like robots to buy toys or manufacture robots. There are also people who operate in a discussion format. Many robots are actually used in industrial sites, and it's not easy to operate them well. If you go to the Aigoo Mars, there are various robots inside the Tank 4D robot itself. So, there are really smart robots using artificial intelligence models.

I am holding an exhibition, but in reality, it is not easy for you to properly receive and understand these clothes. As I mentioned earlier, just coming in, like something we've been doing at 11, continuously changing equipment, solving problems, and making such changes require resources, and continually changing prototypes is important because it costs a lot to keep replacing those clothes. It's not just about capacity, but actually, it's not easy to create an environment that is not used in the field and to operate it. So, there was something called RoboMaker on AWS. What is it? It's a service that makes it easy to create and operate robots in the cloud, allowing you to easily build a 20-layer scene. This is how the robot vacuum cleaner is cleaning this floor.

While driving, we simulate cleaning. So, there is an operating system for 1st Law, which is ROS, the operating system called Opus, and we create program development environments and simulation glasses. We thoroughly test it in the cloud and then create a prototype in the field so that it can be viewed. This is the type of thing we are aiming for. So, you just access it through a web browser and create an environment to realize it at the prototype level. Therefore, companies around the world are really using these rover macros a lot. Currently, companies like Asa always send out Edwards' stuff to explore. They are researching robots not only in schools but also in industrial sites.

Companies are using robots a lot for their purposes, and by operating robots like this, they were able to save a lot of costs before actually operating them. There are cases of using Fit's robots, and in the domestic market, recently, Ewha Brothers has been making a lot of clothes through AJ delivery or making a lot of orders at a restaurant. They are using robots to work in the heat, and even in the case of Doosan Group, when simulating ginger factories, they use robots like Oth. They also use a microphone. At CES in January last year, I first saw Doosan's seeds. They had it and now they are exhibiting it, moving and dancing. All these things that robots do, like dancing, are communicated through makers. Sim Hye-jin and I are also doing exhibitions through this and actually using robots in industrial sites in a proud environment.

You can actually use it as a real weapon. Finally, I will briefly introduce the space industry area. What is this? This is like catching a cloud. You might think of it as a childish thought. Probably, each of you has had dreams since you were young, ignorant and innocent, wanting to jump into a castle or something. In reality, the company that is pioneering space exploration the best now is Sara SpaceX. SpaceX has gone through a lot of trial and error. If you want to reuse rockets, you have to recover them. However, there have been many significant trial and error in doing this. Now, it has been resolved by reusing them. But the videos you see now, even when the rockets explode, people make tens of thousands of mistakes and then succeed separately.

By recycling rockets, we save 10% of the cost of launching the rocket. So nowadays, even NASA doesn't launch rockets. If it's SpaceX, they launch it themselves because it's much cheaper to launch rockets in-house. So now, they are focusing on recycling, recycling, and recycling again, making it easier. A few days ago, our country launched a missile with SpaceX, and now, they mentioned that they have resolved the issue of bearings that used to be separated. For the first time, not only rockets but also bearings can be reused instead of being discarded. Amazon is also interested in launching rockets at 4 p.m., especially since the founder, Jeff Bezos, said something like this: nowadays, space people like him.

We received a lot of great help when we first opened Amazon.com, such as existing telecommunications, server companies, credit card companies, and the cost was low because we did a lot of things with the existing infrastructure. That's why we were able to succeed. But if you think about the space industry, the costs are really high. If a startup says they will launch a satellite, they need at least tens of millions of dollars to start. So, what I'm thinking is, like how the cloud in the existing internet industry adjusted the costs, I hope that investing in the space industry can save costs. That's why I have a company called Blue Origin, and the budget is different from SpaceX.

This is to stack passengers, so we carry passengers four at a time, and then we go up to space, making space look cool like this, and it lifts your spirits. To do that, what should we do? Now, it should become seawater again, right? You can't just use the toilet once and throw it away. Like telling a story again and again, make it recyclable again with 5kg, and then after that, watch a fantastic space lecture by people, and then do business to allow you to lift your spirits like this. But you mentioned that it costs a lot here. We are using all the parts needed here with AWS. We are using HPC technology called hyper-performance computing to simulate many parts through Ethernet and reduce costs in the same way, such as what came in before, like delivery robots.

I would like to talk about one thing that you can now have in your hands, which is the CubeSat. In CubeSat, there is a small satellite weighing 77 grams and 10 cm on each side. It has a camera lens, so you can take pictures of satellites. How amazing is it these days? The camera lens won't rust in space because there is no air in outer space. So, there won't be issues like corrosion. However, if you are a startup and want to make a change in the space industry, it's not an easy task. But here, with CubeSat production, you can make something innovative and impactful. If you want to make something different and meaningful, it's not an easy task, right? But with CubeSat production, you can create something beyond your imagination and make a significant impact, starting from a few million won to tens of millions.

Around the world, thousands of satellites are being created and launched right now. How does a company like SpaceX introduce large satellites like this? There is still space left in orbit, where smaller satellites can be loaded. Then, how about the roots of the trees in space? Yes, they are thrown away like this. These satellites don't stay in orbit for long, about a year. But in the meantime, what needs to be done? So, the cost of launching satellites is very cheap now. Therefore, universities in the United States are conducting research, and even individuals can launch satellites easily. However, the issue is not the satellites themselves but the ground stations that send and receive data. How does a country like Korea deal with space and aviation structures?

When you rent a lab and have something, you should do it this way. Our dish antennas are not installed like that. So many companies or customers say, Oh, there's a problem. What do we do with the satellite now? We need to receive the data, analyze it, and such. That's why we talk about 1WS. So, we received a service called a ground station from this 1WS. This service controls the satellite's attitude and provides the kind of service that can be received. This service is not only available in the United States but also in various data centers around the world. It enables you to use it whenever you want on top of the Earth's space station. This is not just a small dream but companies like Rocket Martin or Dist Glove are already operating satellites and using them for customers. This means that you are closely connected to the space industry. Through this, we can.

The ability to create a business has been greatly lowered by the b, and now the cloud is helping with that. I can explain it like this. Yesterday, at the end of the keynote speech, I would like to conclude by saying this. Through this talk, everyone can dream and anyone can do it. But to do that, you need to stand on the shoulders of giants. Newton talked about that. The reason he made great discoveries was that he could see further by standing on the shoulders of giants. Of course, leaders are leading changes in the world with various technologies and innovations. If you look at the world from the shoulders of the giant technology like AWS Amazon, it will be a better reflection on what you are doing, and also on your thoughts.

I will make sure to finish as I think many ideas will come out. Thank you. [APPLAUSE]

클라우드 컴퓨팅이 가져온  비즈니스 혁신과 기술의 미래  - 윤석찬 (AWS 테크에반젤리스트)

차니 블로그(Channy's Blog) Youtube 채널에 오신 것을 환영합니다. 오픈 웹, 웹 플랫폼, 웹 개발자, 오픈소스 소프트웨어, 빅 데이터 및 클라우드 컴퓨팅 등 다양한 IT 기술 주제에 대해 다루고 있습니다. 

[고지 사항] 이 채널에 있는 영상은 개인적인 의견일 뿐 제가 재직했거나 하고 있는 기업의 공식 입장을 대변하거나 그 의견을 반영하는 것이 아닙니다. 사실 확인 및 개인 투자의 판단에 대해서는 독자 개인의 책임에 있으며, 상업적 활용 및 뉴스 매체의 인용 역시 금지함을 양해해 주시기 바랍니다. 본 채널은 광고를 비롯 어떠한 수익도 창출하지 않습니다.

[Disclaimer] The opinions expressed here are my own and do not necessarily represent those of current or past employers. Please note that you are solely responsible for your judgment on checking facts for your investments and prohibit your citations as commercial content or news sources. This channel does not monetize via any advertising.

Channy Yun - 윤석찬

AWS 클라우드 컴퓨팅의 비즈니스 혁신

아마존의 클라우드 컴퓨팅 사업부인 AWS는 기업들이 인터넷 서비스용 서버를 구매하고 관리할 때 겪는 어려움을 해결하기 위해 2006년에 설립되어 비용 효과적이고 효율적인 솔루션을 제공합니다.

AWS의 성공은 광범위한 서비스, 여러 데이터 센터를 갖춘 글로벌 인프라, 더 빠른 서비스를 위한 5G 네트워크 사업자와의 파트너십에 기인합니다.

클라우드 컴퓨팅은 유연한 IT 리소스 사용, 비용 절감, 애플리케이션에 대한 글로벌 액세스 등의 이점을 제공하여 비즈니스 효율성과 속도를 높여줍니다.

클라우드 컴퓨팅이 다양한 산업에 미치는 영향

클라우드 컴퓨팅은 빠르고 쉬운 투자로 경쟁이 치열해지고 사용자 트래픽을 처리할 수 있게 되면서 기존 산업을 혁신하고 있습니다.

금융, 제조, 농업 등 다양한 분야의 기업들이 디지털 혁신과 글로벌 서비스 제공을 위해 클라우드 기반 운영으로 전환하고 있습니다.

클라우드 서비스를 통해 기업은 빠르게 적응하고 가상 서버를 쉽게 관리하며 서버리스 환경으로 전환하여 궁극적으로 다양한 업계 요구에 맞는 광범위한 서비스를 제공할 수 있게 되었습니다.

클라우드 서비스 및 마이크로서비스의 진화

클라우드 서비스는 컴퓨팅, 스토리지, 기본 서비스 등 170개 이상의 옵션을 제공하여 고객이 필요에 따라 기능을 선택하고 조합할 수 있습니다.

Amazon은 마이크로서비스 아키텍처를 구현하여 독립적인 팀 운영과 높은 서비스 가용성을 가능하게 함으로써 서비스 개발 접근 방식을 혁신적으로 변화시켰습니다.

클라우드로의 전환에는 새로운 기술의 도입뿐만 아니라 진정한 효과를 거두기 위해서는 기업 내 조직의 변화도 필요합니다.

AI 컴퓨팅 성능이 크게 향상되어 컴퓨팅 용량과 더 많은 양의 데이터를 학습할 수 있게 됨에 따라 15배의 성능 향상을 이루었습니다.

특히 NVIDIA V100과 같은 GPU와 Amazon EC2와 같은 클라우드 서비스의 사용으로 높은 컴퓨팅 성능을 효율적이고 경제적으로 활용할 수 있게 되었습니다.

이러한 성능 향상으로 AI 모델의 학습 속도가 빨라졌을 뿐만 아니라 개인화된 추천부터 수요 예측에 이르기까지 다양한 AI 기반 서비스를 제공할 수 있게 되었습니다.

AI는 고객 주문 예측부터 효율적인 배송을 위한 로봇 활용에 이르기까지 물류센터에 혁신을 일으키고 있습니다.

아마존은 물류 운영에 AI와 로봇 공학을 통합하는 데 앞장서며 업계에 큰 변화를 이끌고 있습니다.

컴퓨터 비전 알고리즘과 CNN을 사용하여 물류 센터의 재고 관리 및 품질 관리를 개선하고 있습니다.

아마존의 물류 및 매장 내 AI 및 기술 활용 사례

아마존은 2013년부터 클라우드 컴퓨팅을 사용하여 기존 물류 센터와 자율 배송 및 시제품 개발에 고품질의 AI 학습 모델을 제공하고 있습니다.

또한 원활한 쇼핑 경험을 제공하기 위해 컴퓨터 비전 및 감지 기술을 사용하여 무인 매장에 첨단 기술을 구현하고 있습니다.

아마존은 사기 탐지 시스템과 상세한 코드 검토를 위한 Amazon CodeGuru를 비롯한 다양한 AI 기반 서비스를 제공하며, AI와 기술을 운영에 통합하려는 노력을 보여주고 있습니다.

AWS 서비스가 코드 성능 및 비용 절감에 미치는 영향

프라임데이 이벤트 기간 동안 AWS 서비스를 사용함으로써 코드 그룹의 CPU 사용률이 300% 증가하여 서버 수를 줄일 수 있었습니다.

AWS는 머신 러닝을 포함한 다양한 계층의 서비스를 프리티어 아래에서 제공하므로 비용 효율적인 관리가 가능합니다.

기업들이 클라우드에서 로봇을 제작하고 운영하는 데 널리 사용되고 있는 AWS RoboMaker는 로봇 운영 비용을 크게 절감하는 데 기여하고 있습니다.

로켓 재활용이 우주 산업에 미치는 영향

로켓 재활용은 로켓 발사 비용을 크게 절감하여 특히 SpaceX에서 자체 발사로의 전환을 이끌었습니다.

로켓뿐만 아니라 베어링도 처음으로 재사용할 수 있게 되어 비용 절감과 지속 가능성에 더욱 기여하게 되었습니다.

큐브샛의 생산은 우주 산업에 혁명을 일으켜 스타트업과 개인이 훨씬 저렴한 비용으로 위성을 발사할 수 있는 기회를 제공했습니다.