KERNEL, which is a co-working space for AI engineers supported by Deepcore Inc. ,provides computational resources, food and consulting supports for its members. It is expected to become an incubator for prospective start-ups based on the deep learning technology. In the previous article, Niki Katsumasa, the CEO of Deepcore and Masataka Osaki, the representative of NVIDIA Inc. shared the concept of KERNEL and their thoughts.
For this article, we have visited the KERNEL members to ask about the life in KERNEL and their goals in the future. They have completely different backgrounds and field of studies. Why do they join KERNEL? What is their goal? Let’s imagine the future designed by deep learning technology!
He is a student of The University of Tokyo majoring in mathematical engineering. He is interested in the learning process of human-robot interactions.
He is a master-course student of University of Tokyo. He majors in civil engineering and has developed the system of the automatic map generation by using machine learning.
He is an ophthalmologist and researcher in Jichi Medical University. He has developed an AI based diagnosis for the diabetic retinopathy. He also endeavors to apply AI technology to treatment process of various medical fields.
She majors in precision engineering at The University of Tokyo. She studies signal processing in her lab. Recently, She endeavors to apply deep learning technology to environmental sound recognition. ＊She joined this interview online.
A master-course student of The University of Tokyo. He has majored in law and educational administration in his bachelor’s degree. Now he studies educational evaluation and creativity. In this interview, he took over a role as the facilitator of the discussion.
Members having various themes and different backgrounds
Hiroshi Thank you for coming today, everyone. Firstly, please tell us about your research, and after that, the reason why you joined in KERNEL.
Riku Ok. My research focuses on human-robot interactions by using reinforcement learning, which is represented by Alpha-Go. My goal is developing the system where robots can naturally acquire knowledges and skills like humans. To put it into practice, I joined in KERNEL.
Hiroshi What do you mean by “human-robot interactions”? I guess that they are concerned with the learning process of machine learning.
Riku In the status quo, it is necessary to prepare some setups(ex. rewards) for learning. I do not think it is a natural way to learn as humans. So, I want to develop a new algorithm for machine learning where robots can learn by themselves. But it’s quite difficult.
Emiri I also feel the difficulty of recognition process. Now, I study environmental sound recognition. Especially, sounds other than human voice are my targets. Deep learning has strong and weak area for applications. We have significant hurdles to overcome for human-like recognition.
Hiroshi What strong and weak points of machine learning are reflected specifically?
Emiri The process of environmental sound recognition can be separated into 2 parts. They are how to extract individual sounds, and how to understand the meaning of extracted sounds. Deep learning can be applied to the latter process. For example, if you want to distinguish “good morning”, ”good afternoon” and “good evening”, deep learning works well by iterative learning.
However, when you deal with environmental sounds, it is necessary to consider “why it sounds like this?” It is extremely difficult in applying to environmental sounds because there are almost infinite patterns of sound in surroundings. That is my big challenge.
Hiroshi I understand that deep learning is not a panacea for now. Batran is from Egypt right? And why did you choose to come to Japan for the present position?
Batran Yes. I graduated from Benha University in Cairo and got a bachelor’s degree in civil engineering. After graduation, I fortunately watched a presentation of my current supervisor in TEDxTalk. In the talk, he talked about the idea of modeling human mobility from mobile phone data to obtain the distributions of people in cities. I liked that, and was really got interested in it. Now, I finished my masters thesis in that work.
Hiroshi Using images or videos are typical approaches of civil engineering, isn’t it? What is the impact of AI technology on civil engineering?
Batran For example, using mobile data, such as GPS, can be a new approach in civil engineering. In our study, we focus on the human mobility in the cities. Such data can be used for the design of public transportation and controls. Now, it is my big goal.
Hiroshi Thank you. I heard Mr.Takahashi is a doctor. So I am curious about your story to join KERNEL.
Hidenori I work as an ophthalmologist and also have an academic position as an associate professor. I came to know deep learning technology 3 years ago, then I developed the algorithm for diagnosis of Diabetic keratopathy* by using deep learning. Now, I am trying to apply deep learning not only to diagnoses but also to treatment processes. My entry to KERNEL is for finding the method to realize it.
* a complication of diabetes, which causes blind by bleeding.
Hiroshi Specifically, which treatment parts do you try to solve by using deep learning?
Hidenori The treatment of Age-related Macular Degeneration (AMD)** is my target. AMD is a disease of retina, which can lead to blindness. In status quo, there is a specific medicine for AMD. However, it is difficult to make a decision over when to inject the medicine. If AI can judge whether and when patients should have injections, such problem can be solved.
** a disease which may result in blurred or no vision in the center of the visual field.
Hiroshi I see. Then, what is the difference between the typical method of judgement by using images and your proposal?
Hidenori For example, “If we find bleeding, we inject.”; previous algorithms are like this. It is possible to teach AI to a certain level by such algorithms. However, actual judgement is not so simple. There are different diseases having similar symptoms. In addition, the progress of symptoms should be considered to decide on whether to inject. In other words, it is necessary to teach AIs of tacit knowledge of doctors. On this point, it is different from the typical AI judgement systems.
Supreme environment provided by KERNEL
Hiroshi How do you use the supports by KERNEL? What is the good point?
Hidenori Basically, strong computational resources and hi-speed communication infra are very good. It is essential for deep learning to have GPU or something. Especially, if your laboratory does not have enough computational resources, I recommend for you to join.
Batran This location is very useful and the rooms are quiet. In addition, there are a lot of smart guys which have different themes and backgrounds. We can teach methods and theories to each other. I made the best choice to come over to Japan and join KERNEL.
Emiri KERNEL provides chances to talk with learners of deep learning and entrepreneurs. It is hard to get ideas in other fields only in my laboratory. Connections in KERNEL must be my treasures and assets.
Riku I really appreciate great lineup of books. Of course, supports of foods are also very good. I sometimes go to KERNEL office just to have dinner, haha.
Hiroshi Food! I think that the support of dinner can be a good incentive. Dinner time can accelerate discussion among members. It should be a good occasion to create new ideas.
Riku I agree with you. Moreover, reading societies of papers are held to deepen the understanding of several approaches and algorithms. It helps to improve my skills.
Hiroshi You are a teacher and also a student. That’s great! You have various backgrounds and skills. If we have some matching system to connect each other, we can easily share technical interests and communications will become more active.
The future accelerated by the deeplearning
Hiroshi If you have any plan in 5 to 10 years, please share with us.
Emiri I will be a sales person in an IT company in the next year. However, I would not continue it during my whole life. By seeing various people and thoughts, I want to absorb new ideas and inspirations. When I will get a chance, I would start a business and/or get back to academia world.
Batran I have applied to some companies to be an AI engineer this year. However, I do not think that a life as a salaried worker is promising and one should always seek what’s beyond.
Riku After graduation, I will apply to graduate school. I want to apply deep learning to primary industries such as agriculture or fishing industry. For example, as for fishing, I consider that a support tool which can predict the information of fish under the sea would be a good idea.
Batran That sounds nice. His study can be used to choose fishing points. It can strongly support the immature fishermen by saving the time to find fishes.
Riku The primary industry is not optimized well. I want to tackle a development of new applications targeting next 10 years.
Hidenori My goal is also saving time. In the status quo, many doctors need to wait for several hours to do 3-5-minute treatments because the management and judgement process take time so much. To solve this problem, I have launched an AI company for ophthalmological development. If the new AI technology of judgement for treatments is realized, the whole process of medical treatment will be accelerated.
KERNEL is seeking new members. Let’s join the perfect environment for learning AI technology. KERNEL and its members will provide the utmost support and stimulus for you to advance your research!
Conditions for application
・student or engineer whose age is over 18
・frequently come to the office of KERNEL
・have interest of AI technologies and knowledge about machine learning (especially deep learning)
・graduate/undergraduate student in science and technology
・have interests in application of computer science in society
※KERNEL welcomes your application regardless of whether you have any achievements
Please check the detail in the following link.