Mr. Yutaka Matsuo, Chairman of the Japan Deep Learning Association (JDLA), took the stage at a seminar on the 9th at the 5th AI/ Artificial Intelligence EXPO [Spring] held from April 7th to 9th, 2021. did.
Despite being held offline, more than 1,000 people attended the lecture and it received a great deal of attention.
With the theme of “the front line of AI ( deep learning ) utilization in the DX era,” Mr. Matsuo introduced various business cases and the latest technology utilizing deep learning, and explained the importance of AI and deep learning in DX. , explained the new policy “DL for DX” announced by JDLA on March 30.
In this article, I will report on the seminar that Mr. Matsuo took the stage.
Contents
- Deep learning moves from technology to application phase
- Deep Learning Now – Concern about big delays in utilization in Japan
- How to digitize real data is important
- It is necessary to strengthen the literacy of management and general employees
- New AI Basic Course “AI For Everyone” Starts May 6, 2021
Deep learning moves from technology to application phase
Compared to 3-4 years ago, when deep learning technology itself was the focus of attention, Mr. Matsuo says that in the past 1-2 years, the focus has been on how to use deep learning in business and DX initiatives. I said that it is changing to
Mr. Matsuo also said that there are many cases of DX in Japan, and he picked up the following cases.
|
Deep Learning Now – Concern about big delays in utilization in Japan
In an introduction to the latest technological trends in deep learning, Mr. Matsuo said, “In 2012, the accuracy of deep learning image recognition improved significantly, resulting in a breakthrough, and in 2018, the field of natural language processing made great progress.” and explained that the accuracy has improved significantly.
He also mentioned that the general-purpose language model “GPT-3” developed by OpenAI in the United States in 2020 had a major impact on the natural language processing field, and is expected to be used not only for natural language processing but also for images and videos. I said
GPT-3 can also be used to create tools such as inputting English sentences and outputting them in HTML. AI uses text data on the web to learn natural language processing. In other words, I am also learning the relationship between HTML code on the web and natural language. So you can type instructions in English and instantly get an HTML command that executes them on the web. It’s like automated programming, and programmers may soon be out of business.
(Omitted)
” NeRF ” enables accurate image representation. Conventional pixel images become blurry when magnified and the details become unclear, but “NeRF” learns how an object looks from all angles with a neural network, so it is always clear regardless of “where you look at it”. image can be displayed. I believe that NeRF technology is similar to how the human brain perceives images.
It is still only partially known, but I think it will be used practically in various places in the future.
NeRF
Image generation technology developed by a research team consisting of the University of California, Berkeley, Google Research, the research division of Google, and the University of California, San Diego. By learning photos taken from various angles using a neural network, it is possible to view objects three-dimensionally and realistically from any viewpoint.
Mr. Matsuo expressed concern that while advanced development of GPT-3 is progressing in English-speaking and Chinese-speaking countries, Japan will fall behind in its utilization.
Since GTP-3 requires learning with a huge model, it is said that it costs several hundred million yen or more for one learning, and capital strength is required. Japan tends to start losing when it comes to the battle for capital, and there is a feeling of that now.
In Japan, LINE has announced that it will enter the development of a large-scale language model, and development is continuing.
How to digitize real data is important
DX is generally classified into “digitization”, which digitizes paper documents using technologies such as OCR, and “digitalization,” which improves efficiency and value by utilizing digitized items.
Mr. Matsuo explained the differences between Japan and other countries in the way DX is approached, saying, “One of the characteristics of DX in Japan is that there is a tendency to work step-by-step after digitizing.”
For example, in the case of taxi dispatch, communication between customers, operators, and drivers was all done manually. In order to realize the automation of vehicle dispatch by AI, a method of creating a new model after converting it into data and improving efficiency is adopted.
However, overseas, they suddenly create a new model like Uber.
Even in the case of retail, in Japan, the general trend is to digitize paper and personalize customer data. However, overseas, like Amazon GO, we have suddenly realized a new model store by utilizing deep learning technology such as apps, products, and face recognition.
What DX realizes is a new model, not an improvement in existing operations. In the future, all operations will be simplified and almost all automated.
Mr. Matsuo went on to explain what has become possible through the development of deep learning by breaking down the flow of DX initiatives into elements.
What is important is how to digitize the data in the real world. It has been possible to track POS data and movement history from the past. What is new now is this part in red (in the figure below).
For example, it is possible to use real situations such as faces, characters, and images, improve the accuracy of predictions using data, generate natural language, and automate machine control. From now on, it will be important to consider what kind of business these technologies will be used for and how they will be combined.
With DX, you can realize what you couldn’t do before and discover new added value. This change will occur in every industry and industry.
It is necessary to strengthen the literacy of management and general employees
I think that there are many people in your company who do not have high digital and AI literacy. Many people find it difficult to communicate with such people. In particular, it is important to reach out to management and general employees who are not directly involved in development or IT. In order to advance AI and DX projects and utilize them in technology to create solutions, it is necessary to study technology and improve the understanding of the entire company.
New AI Basic Course “AI For Everyone” Starts May 6, 2021
JDLA provides G test (Deep Learning for GENERAL) for business people and E qualification (Deep Learning for ENGINEER) for engineers, aiming to develop AI human resources. The total number of people who have taken the G Test has exceeded 50,000, and in recent years it has been ranked high in the ranking of qualifications that people want to acquire.
“However, these qualifications are highly difficult and difficult for beginners to AI,” said Mr. Matsuo, explaining the new free online course “AI For Everyone” that will be launched at JDLA . .
JDLA will start a new AI basic course “AI For Everyone” on May 6, 2021. This course is a gateway for general business people and managers to understand how AI and deep learning can be used in business.
Deep learning will continue to develop significantly in the future. We believe that it is necessary to create an environment that makes it easy for people with a strong desire to utilize deep learning to play an active role in the company, and JDLA will continue to support that.
