・Start of blog greetings and Introduction
I work at a tech startup in Tokyo. When I was a student, I worked with BASIC programming, then moved on to HTML, CSS, and web service development with Ruby on Rails, and then deep learning using Python. Currently I serve as CtoC Web service business manager and SaaS model Web service business manager. I do product development as a hobby.For the background of why I started blogging, I felt that some content is hard to understand when it is just talked about. For example, questions frequently asked from clients or in an interview. It is possible to share the document in the form of a blog so I thought that I could help others by starting this blog now.
The target audience is for those on the business side who are interested in startups and technology. If you have basic knowledge of HTML and CSS and experience in project management, I think that you can understand the contents.
Furthermore, since this is not a developer’s blog, I will describe the technical content simply so that it is easier to understand for those who are not engineers. Regarding the special technical contents of deep learning, it is out of the scope of this blog, so please understand (since I think that it might be unsatisfactory for an engineer).
・About Deep learning
1:Frequently asked questions about AI
http://qiita.com/sakaiakira/items/9da1edda802c4884865cI think that not only the startup industry but also major companies are paying attention or are becoming aware of artificial intelligence so I think it should be easy to understand machine learning and deep learning well. I feel that the definition often differs for each book.
Current AI is in a period called the third AI boom, and there were some circumstances for its demise in the past two booms. However, this time a breakthrough called Deep learning has occurred, coupled with the spread and development of the Internet, I believe that it has driven artificial intelligence.
２：What is AI? Machine learning and deep learning
In the first place, what is Deep learning? For this question, deep learning itself is a part of machine learning. Basically, Deep learning adds functions for a "computer itself to think" such as extraction of features and figures. In addition to machine learning, it has mainly recommendation functions centering on supervised learning. This is a famous turning point, and Google has produced results that can distinguish images of cats.３：American AI Startups
・About Affectiva（Image Recognition, Emotion Recognition)
http://www.affectiva.com/#what-you-can-do
http://www.affectiva.com/
Affectiva, well known as a startup for Deep learning, was founded by Rana El Kaliouby and Rosalind Picard from MIT Media Lab in the United States. This company succeeded in analyzing emotions from facial expressions by analyzing human facial expressions around the world for about 50,000 people in 30 countries and extracting their features. It calculates and utilizes methods for advertising effectiveness as a business. It has already raised 33.72 million dollars (more than 3.5 billion yen), and it is a startup that will be drawing much attention in the future.
I met you when I came to Japan and felt you had a lot of charisma.
４：Japanese AI Startup
・ABEJA
http://tech-blog.abeja.asia/entry/object-detection-summary
Meanwhile, ABEJA, a deep learning service for retailers, is in Japan. As a start-up since 2012, it has developed services that utilize deep learning, and has already raised 850 million yen. In addition to programming, I feel that it is progressing one step ahead of the other startups by developing a GUI awareness service.
・Preferred Networks
http://www.todaishimbun.org/pfn-nishikawa20160724/
https://www.preferred-networks.jp/ja/
In addition, Preferred Networks is listed as a company that is drawing attention in automated operations and IoT by partnering with Toyota Motor Corporation and Fanuc. They have hired excellent engineers to have high technical capability. They are also collaborating not only for cars but also drones, and it is a startup that can be expected to rise in the future, and has already procured more than 1.8 billion yen.
５：Utilization of AI from major American companies
https://www.google.co.jp/
GAFA (Google-Apple-Facebook-Amazon) has an overwhelming presence in the world's leading companies, and Google is leading programming libraries. Deep learning technology is used for Google translation, and it is getting more accurate and has the power to continue to improve accuracy by continually learning.
・IBM
https://www.ibm.com/jp-ja/
On the BtoB side, IBM's Watson is entering various areas including call centers, medical systems and so on. IBM is reinforcing its AI department not only for engineers but also on the business side, I feel that the point is to break into affecting businesses compared to an engineer-oriented company.
６：Utilization of AI at major Japanese companies
・Recruit
http://www.recruit.jp/meet_recruit/2016/01/it02-2.html
http://www.recruit.jp/
For a major Japanese company, Recruit invites talent from the MIT Media Lab in the United States and utilizes a large amount of data to develop new solutions one after another. Recruit is working on building a system that can utilize artificial intelligence even for business-side talents who are not engineers, and more than 4,000 employees have already been able to utilize AI.
Mizuho Bank
https://robotstart.info/2016/09/29/mizuho-watson-yt.html
https://www.mizuhobank.co.jp
Also, Mizuho Bank is proceeding with AI use in its call centers and has replaced call center operations that were supported by outsourcing to AI. By the way, what is being tried here is IBM's Watson.
５：Trying to implement Deep learning
I am using Python when implementing Deep learning myself. It is convenient to use for beginners to experienced people, and it is convenient to be able to handle the development on the server side as it uses the same development language. The library is also substantial and it is possible to do development from a state where packages are prepared to some extent rather than from a zero base.６：Important things for Deep learning
As for deep learning, not only programming but also the original data and its quality are important points. If there is no clean data, the learning effect will be diminished. For example, it is difficult to do deep learning that can improve in stages even if low quality data is loaded many times. Therefore, how do we build a learning cycle and continue learning good quality data? Good data is required.Also, with respect to data, it is necessary not only to have data but it needs to be normalized so that it can be used as a database. For this point, contents belonging to the field of statistics rather than programming are included.
Although there is not much data available in Japan, in the United States it is more available to use data. I am trying to do deep learning by trial and error, mostly by reading in American data.
How can you get good quality data? I think that it is an important point in Deep learning.
At Kabuku, where I work, we are looking for talent who can change the world through democratization of the manufacturing industry.
If you want to hear our story and feel like coming to visit our office, please click the button below.
In addition, this blog and wantedly will cover information not only about Kabuku but also about various information on technology and startups. If you are interested, please by all means follow me.
Writer Masafumi Asakura
