#Digital
2018.07.19
Edition No.21 AI-Creating Artificial Intelligence, AutoML
ColumnistChoi Moongyu
Amorepacific Digital IT Innovation Team
What is AutoML?
One of the goals pursued by global IT giants that have preemptively prepared and provided Artificial Intelligence (AI) technologies, such as Google, Amazon Web Services (AWS) and Microsoft, is 'generalization of AI' or 'popularization of AI'. Until now, AI technology was considered high-tech that only a handful of statisticians and technology experts used, and many companies have yet to introduce machine learning. Companies require machine learning models to provide their own unique AI-based services. However, there are not enough skilled resources who can build models, which require a series of complex processes that need a lot of investment and time, while the result is not guaranteed. That is why there is skepticism around AI along with a widening gap between tech companies and non-tech companies in using AI technology. This is where AutoML comes in – it was developed as part of an effort to popularize AI technology to enable developers, data scientists and business analysts to have easy access to AI technology.
-
Challenges of machine learning
Source : https://www.quora.com/What-are-the-machine-learning-APIs-I-can-use-to-analyze-the-correlation-of-data-elements-in-a-file
AutoML – artificial intelligence that creates artificial intelligence
Up until now, businesspersons, data scientists and engineers who develop machine learning codes were involved in the actual development process of a machine learning model. The process included the following steps; Fetch → Clean → Prepare → Train model → Evaluate model → Deploy to production → Monitor. And finally, the training process is iterated to find the "best" result and increase level of accuracy. This was the process of building a machine learning model.
What's special about AutoML is that it helps in using machine learning technology easily. For a machine learning engineer to build a model, one would have to select the best algorithm, training process and variables for training by using training data. And experience and high technological expertise are required to achieve good result. If you use AutoML, however, any person with a certain level of AI development experience can create a model very easily and quickly.
What's special about AutoML is that it helps in using machine learning technology easily. For a machine learning engineer to build a model, one would have to select the best algorithm, training process and variables for training by using training data. And experience and high technological expertise are required to achieve good result. If you use AutoML, however, any person with a certain level of AI development experience can create a model very easily and quickly.
-
Illustration of SageMaker development process
Advantages of AutoML
The secret behind AI that creates AI is using transfer learning and Learning to Learn, which are machine learning methods. The concept is that problem-solving for a new problem-solving can be done easier and faster by automatically using the completed AI model.
-
Illustration of transfer learning
Source : https://docs.google.com/viewer?a=v&pid=sites&srcid=ZGVmYXVsdGRvbWFpbnxsaWZlaXlhZ2l8Z3g6M2FiODZjNmYxZTRmYmY2MQ
So, to summarize the advantages of AutoML;
First is 'excellent accuracy'. AutoML allows machine learning model to learn comparatively easily with better accuracy because it uses an already-trained model to further make the model learn.
The second advantage is 'building speed'. Using AutoML significantly reduces the AI model-building time compared to the previous AI development method, which took at least a few weeks or sometimes even months.
The third advantage is 'easy-to-use interface'. Building machine learning model has become easier by using intuitive graphical user interface (GUI) – and not the difficult programming interface.
First is 'excellent accuracy'. AutoML allows machine learning model to learn comparatively easily with better accuracy because it uses an already-trained model to further make the model learn.
The second advantage is 'building speed'. Using AutoML significantly reduces the AI model-building time compared to the previous AI development method, which took at least a few weeks or sometimes even months.
The third advantage is 'easy-to-use interface'. Building machine learning model has become easier by using intuitive graphical user interface (GUI) – and not the difficult programming interface.
The current status of AutoML
1) Google launched its 'Cloud AutoML' service to assist in building machine learning models. Google CEO Sundar Pichai said "in 5 years, a person who doesn't have any knowledge in coding or computer language will be able to design machine learning programs suited for his or her needs," predicting a rosy future in machine learning.
Source : https://www.blog.google/products/google-cloud/cloud-automl-making-ai-accessible-every-business/
2) Amazon Web Services also released its AI-building platform 'SageMaker', which assists building the development environment for machine learning, provides verified algorithms based on actual Amazon.com experiences and various machine learning templates, and automatically creates easy and reliable services.
Source : https://aws.amazon.com/ko/sagemaker/
3) Microsoft upgraded its previous 'Machine Learning Studio' and introduced 'Azure AI', which is a complete management service that enables users to build AI and services easily.
AutoML, savior to artificial intelligence
The shortfall in talent in AI area is already very severe.
According to an AI status report released in end of 2017, demand for AI talent grew exponentially worldwide. While it is estimated that the number of talent required by AI related companies worldwide is about one million, the number of actual people working in the area is a mere 300,000. And among the 300,000, about 100,000 people are researchers. This means that about 800,000 resources with expertise are required. In addition, the talent shortfall in this area is expected to be chronic, considering that nurturing AI experts is challenging. The supply of talent is about 20,000 a year, a significant shortfall against market demand.
[Source : Tencent's 2017 Global AI Talent White Paper]
Korean companies plan to recruit 9,049 highly-skilled AI talent over the next 5 years, but the number of supply will be a mere 1,781, resulting in companies suffering from a shortfall in highly-skilled talent in the number of 7,268.
[Source : Electronic Times Internet, http://www.etnews.com/20180427000267]
Korean Internet content service company Naver is leading efforts in releasing research performance at academic conferences in and outside Korea, led by its Clova AI research team. It is also increasing efforts in securing talent to improve its capabilities in AI technology, by recruiting more than 1,500 AI talent in the first quarter of this year alone. Samsung, on the other hand, established Global AI Research Centers in the U.S., the UK, Canada and Russia. And it plans to increase AI researchers to more than 1,000 by 2020, based on its AI Center in Korea.
[Source : JoongAng Ilbo, http://news.joins.com/article/22446495]
It is true that recruiting AI talent by non-tech companies is much more challenging as it is difficult to identify and secure well-trained AI talent in the first place. If a company develops its own AI technology or use open source, there is the advantage of being able to internalize the AI technology, but there is also the risk of lagging behind competition by focusing a lot of resources, investment and time in technology development.
Integrating AI in business has become an essential agenda for all companies as we face the era of AI where "AI is Everything". And AutoML will bring huge changes in the overall business landscape as a technology that enables even companies that lack AI development-related talent build AI technology and apply to services.
As Amorepacific's current situation is not so different from that of others, AutoML may be a realistic solution to the problem.
Integrating AI in business has become an essential agenda for all companies as we face the era of AI where "AI is Everything". And AutoML will bring huge changes in the overall business landscape as a technology that enables even companies that lack AI development-related talent build AI technology and apply to services.
As Amorepacific's current situation is not so different from that of others, AutoML may be a realistic solution to the problem.
Amazon Digital Hyecho and AutoML
I had the opportunity to participate in the "Amazon Digital Hyecho" project, the theme selected for 'Theme Hyecho' this March. In mid-May, I came to Seattle, Washington, the city where AI technology is growing at the most dynamic pace, to learn about machine learning and carry out projects to apply my learnings to business with technology experts for about a month and a half.
-
Amazon Headquarters photo / Amazon Front Desk photo
The key agenda here is basic and in-dept learning and practicing of machine learning to improve skills and expertise on building machine learning models using Amazon's AutoML, SageMaker. I was lucky enough to have the opportunity to learn about the AutoML tools of Dimensional Mechanics, Amazon's technology partner currently working on upgrading SageMaker to make it into a stronger "Artificial Intelligence creating Artificial Intelligence". The Knowledge Base of the Artificial Intelligence that makes Artificial Intelligence is called the Oracle. The fact that with more experience, the Oracle becomes smarter and can easily and efficiently solve more complex and difficult problems is the most appealing feature of AutoML from a perspective of a company trying to use it.
-
Dimensional Mechanics logo
-
Neopulse logo
-
Learning from DM Neopulse CEO
-
Neopulse development review
Machine learning development is still very difficult as I just started to learn, but I also experienced first-hand that using AutoML helps to solve simple tasks quickly and easily.
As I write this column, I am preparing for a project to build machine learning model using SageMaker with Amazon's Machine Learning Solution Lab in San Francisco, California.
As I write this column, I am preparing for a project to build machine learning model using SageMaker with Amazon's Machine Learning Solution Lab in San Francisco, California.
-
ML Solution Lab
When I return to Korea, I plan to humbly contribute as Amorepacific's AI Core of Excellence, supporting the application and spread of AI in business, based on my experience with the Amazon Digital Hyecho project. I will share my experiences in using AutoML to contribute to improving Amorepacific's AI capabilities. I thank you for your encouragement and support in me to successfully complete the Amazon Digital Hyecho project.
-
Like
0 -
Recommend
0 -
Thumbs up
0 -
Supporting
0 -
Want follow-up article
0