Anything you would like to discuss?

There are several things. Let’s say the first topic could be just a little bit of introduction of myself, as well as our company, and then an intention. We are originally from Japan, but we are now expecting to have a AI researcher branch here in Taiwan.

Shares a little bit of background. Number one is I’m a Japanese. I used to be very domestic person until 30 years old. I got a PhD of neural network. Since then, I enjoy a lot in the start-up world, exiting the company to the big corporation, like Yahoo Japan and Mixi, until when I was 30 years old. Then my passion is upgraded.

Since then, my dream was simply I wanted to do something technical in Japan and I just wanted to do the business in Japan. I found that the IT area, actually more in technical area, is much bigger in a global market, including the US and China. I really wanted to do something in a global market, rather than only limiting myself in Japan.

Currently, we do the company called Cinnamon, which is a AI provider for enterprises. One of the product of ourself is the name, Fax Scanner. For example, in the insurance company or let’s say in a bank, 20,000 of the sales people are working on the sales of the mortgage.

What they are doing, of course, 70 percent of their time is still sales, but 30 percent of their time is just inputting a lot of information. They have to receive, let’s say, the pay slip, the salary proof, or a bank statement, or clinical record, medical records, the blood check or a urine check. Those documents are sent by the customer. Then they are checking all through those data.

Actually, how to check, because they’re not medical people, they have to input all of those information manually. It takes 30 percent of the time of those sales people. That means 30 percent of the time of 20,000 sales people. Those processes are, to me, ridiculous, because it can be done by a computer.

Let’s say hospital A, hospital B, hospital C, they have a different format. That is why the simple IT solution cannot solve this problem. That is where the AI plays a big role.

We are working on this kind of an AI to automate some of the business processes. We are currently very successful. For example, last month’s sales is a half million seller only for selling just document scanning solutions in Japan. This is what we are now working on. We came back to Japan since the market is very huge in terms of this kind of business automation.

The reason is very simple. Japan has the biggest amount of the wasteful work. That means, for example, the US are pretty well digitalized. China is actually not bad in terms of the digitalization. But tons of the operation in Japan is still analog, the paper-based. That is why the market is huge.

Of course, even in US, there are several businesses processes which require automation, as well, but actually it’s a little bit behind, maybe one year behind compared to Japan. Now Japan is very excited with the AI solutions. Now I observe that the US or the UK has several big insurance companies, or the bank started to pay attention to those kind of AI automation.

That is why we came back to Japan once. Then, of course, we have aspiration to go to the US or the UK market. This is what we are working on. We also found that the talent side, especially the AI -- this is related to Taiwan -- but the talent in AI is very precious. It’s almost not possible to find, let’s say, hundreds of AI talents in Silicon Valley, because they are hired already by a Google or a Facebook.

I found that, for example, in Vietnam, we already hired 50-plus AI engineers who can code the deep learning from scratch. The reason is because Vietnam is very strong at the computer science. Their education system encourages everyone to be a computer scientist, because it’s a very great industry for them.

Hence, in terms of the top tier, very smart people. In terms of science and technology, they want to be the computer scientists. That is why we enjoy hiring those top-tier people. I found that Taiwan is exactly the same. Very top tier, very smart people are willing to be the computer scientists here.

There are a shift from very hardware focusing here to recently very AI, or the software industry, at least. That is why I really want to get some help from Taiwanese people, as well. This is why we decided to come to Taiwan. More fundamentally, what I really want to do is to build up a ecosystem here as a band of talents, including Japan, Taiwan, Vietnam, and maybe India.

Those have tons of suppliers. I mean that technically very good people exist. Actually, except Japan, the market is not so huge. Even in India, a AI solution can’t be sold to a big corporation there, simply because of the impact of AI is still very huge only in the developed countries.

For example, like the US, they have a very huge revenue. That means, just one percent of optimization, it is already a big impact for them. For example, Taiwan, even though they still want to do the investment...

Let’s say in Japan, it’s very easy for the big corporations to invest one million US dollars for the AI solution, very easy for them, but it’s not easy even for a top-tier insurance companies here in Taiwan. They are very conservative in investing their money for those solutions.

We are capable of bringing those foreign demands to the Asian countries. Then, if we can connect the local Taiwan talents or local Vietnamese talents, local Indian talents to the demand, then it could be possible for an APAC region to be the global leader of AI invention.

That is what we really want to do. What essentially our company, Cinnamon, is doing is connecting the talents in Japan and maybe Korea, in Taiwan, Vietnam, and then maybe India, connecting all those AI talents together to build up a AI invention cycle.

We also foster several business center in Japan, in the US, and the UK to deliver the demand side into this area to finally connect. This is what I really want to do.

This is what Cinnamon is working on, but as a personal level. I also have some aspect as an investor. I have two activities. One is an engineer and a scientist. The other is a fund. In total, I invested in more than 50 of the companies across Asia. Let’s say that 90 percent are related to the AI.

That means I am not only interested in hiring people, but I want to be connected with a local AI inventor. Then I want to deliver some opportunities for them to be great big guys or a big start-ups, let’s say. This is my personal level interest, as well. Sorry, [laughs] I just made a bit of a random explanation, but this is, overall, what I am doing.

Now, in terms of the development phase of Taiwan, we are just starting to talk with several AI engineers. Of course, we want to do something for the Taiwan community, as well. How to say? It’s not only for just hiring people. I just want to identify what is the real piece that the Taiwanese community really need.

Let’s say, maybe it is a demand. Maybe, if I can transfer my know-how of how to commercialize the AI into the production level, then I’m very happy to share my know-how. Then I could be helpful for, maybe, the start-up people, or maybe the students, or maybe some business people here.

Then I’m very happy to have conversations or speaking with them, so that I can just give something to the community first.

My approach is, basically, I don’t want to be just a give-and-take person. I just want to be the giver first -- how to say? -- pay-forward kind of things. Finally, it will be a happy scenario if somebody is very interested or attracted by themself and working together with some of the people. This is what I’m now starting.

You’re based in Taiwan now?

I’m on the visiting basis, because I still have to manage the company side, the data R&D side. I’m going to spend, at least, one week a month. Maybe for the regulation, I can’t be more than half here, but I’m staying as long as I can.

Is there a local team or a partnership that you’re working with?

I am now working with several non-tech people who are setting up everything. For the technical people, we are very conservative, because the first, and the second, third tech-worker people is the core of the team. I’m interviewing a lot of people, actually more than 20, but I only made a offer to one.

It’s mostly on the visual domain of AI, optical recognition?

OCR, Optical Character Recognition.

Handwriting and also printed?

Handwriting, the printed, and also understanding the structures of the documents. It’s a layout analysis. For example, the bank statement is a stable structure. Then we have to understand this is a credit, this is a debit, or this is the remaining, those kind of semantic understanding.

I read about a start-up started by a Japanese teenager, I think high school student, where you can use an app to take photo of your invoices, and he just gives you money.

That’s kind of fun, but it’s shut down within a day.

I know, but it’s kind of a provocation, right? It’s like performance art.

Your technology will be able to make sense, actually, of the invoices taken this way? Or does it have to be a professional scanner? Is it hardware dependent?

For the actual operation, even though, of course, we have the capability of the photos, the usual invoices are exchanged via email, so the invoice is exchanged as a fax or email. The scan quality’s not too bad, usually. How to say?

Like 150 dpi or something?

The resolution is bad, because it’s a fax, but it’s not distorted. Even if it’s a photo and it’s a really light...

Lighting and everything...

Distortion.

...like beauty filters.

You’re mostly working with scanning. Even at fax resolution, it’s usually pretty regular.

Yeah.

It sounds very lucrative, actually, [laughs] because it’s a well-defined domain.

Actually, it’s technically very difficult. Of course, handwriting is sometimes very dirty, especially for the medical records. [laughs] Doctors are writing so fast.

It needs a lot of domain knowledge, I’m sure.

Another thing is, for example, Chinese letters are maybe 5,000 or something, and they’re 5,000 multiplied by...for the deep learning, usually we need at least the number above 1,000 of samples required for the best accuracy. That means it’s almost impossible to acquire those data. That means we have to think about something else.

The recent trends on the deep learning is to computationally generate the contents for the training. We are now letting computer to write [laughs] instead of using a real human to write the letters.

You still need supervisors, right, people who do the typing? You use, what, Mechanical Turk? [laughs]

For example, the theory is I use 100 people to write only 72 letters. Then we pick up some of the parts, and then we’re synthesizing the other characters, so we can inflate the data size.

That’s pretty good technology.

Behind the OCR, actually we did a lot of this kind of thing. That is why we require a lot of AI talents.

I’m just checking my understanding. You are saying that you’re willing to share this knowledge of how you build this pipeline, the general idea of how to apply machine learning for large-scale problems, to the local AI community.

If people are interested, you’re very willing to engage with the meet-ups and AI learning circles here.

Correct. This is part of the content that I usually do. Maybe, the first one is the basic stuff of the deep learning, more architecture, and things. The second is...

GAN, yeah.

...is GAN and the current trends on the data synthesis. The capsule network is the latest one. Less data is a trend how the start-ups can win over Google, this kind of things. It’s one part that’s technical. The other part is more of a business application, like, "What is the real demands?"

Usually, technical people are thinking about something not so great in terms of, for example, the size of the market. I’m teaching where is actually the really hard part. I don’t know. This is a very big industry kind of things. Also, this is more of thinking about how to reduce the workload.

The blue color side is the real value. The white color side is more of how to reduce the cost. Of course, the AI make errors, so how to handle the error by collaborating with the human. Those theories are gradually installed to the students. In Vietnam we did it already, and I am willing to have similar things in Taiwan, as well.

That’s awesome. Thank you. It’s very well-balanced.

How to say? I don’t believe much that only a business person who are not familiar with the technology or who are not interested in the technology can understand the mechanism of how the AI business is working. I strongly believe in some excellent technical people who are willing to invent something. Then they can understand more about the business context, as well.

Usually and finally, some of them really want to make money. For example, half of our engineers’ motivation is to learn something from myself or our R&D. Three years later or four years later, they really want to have another company founded by themselves. Then they could do some of the innovation.

We are more than happy with it rather than sticking to us for 10 years. Our ultimate intention is more to build up a ecosystem. We want to strengthen the community itself, rather than just sticking to our own interests.

I’m hoping to have the similar talents here in Taiwan, as well. If immediately people can pick up with all the concept in AI, you can start your start-up right now. I want to invest, maybe. In case that they want to learn from us in more detail. Then once digesting, then they want to spin out, then we are happy to do it.

This is one of the differentiator from the AI Academy or some other community builders in this space.

You just mentioned AI Academy. I think their collaboration with the private sector is pretty solid, in the sense that they both have mentors and have real-world business problems that they use to solve. Of course, it’s not just AI.

There’s a larger data scientist community here, as well. They have lots of meet-ups. I’m sure you already know about it. They have a lot of sharing experiences. If you sign up to be a instructor, or a lecturer, or a mentor, I’m sure that people will love [laughs] this resource, because it’s really useful.

What I’m also hearing is that it’s not just about knowledge sharing. You also want to identify the particular inventions that is being produced here, and to align it with business problems that you discovered elsewhere, right?

Yeah.

That’s the next level of collaboration.

Yes, that’s correct.

Do you have a local partner for that? In Taiwan, for example, Microsoft is committed to have their own AI lab, maybe 100 to 200 researchers. They will, of course, build their own ecosystem.

Just yesterday, we were talking about the voice domain with Mozilla. Mozilla is going to work with the Taiwan Mozilla team to do machine learning, but around the language education domain.

Also, Google with its Intelligent Taiwan program, and NVIDIA. There’s lots of people. Mostly they come and they find one specific domain. Then they start a incubator program or they join an existing incubator program.

I see.

Maybe, they work with Taiwanese Startup Stadium, AppWorks, or whatever. That saves you time to uncover talents, because you, basically, just come to the demo day and see which talents fit your use cases. You don’t even have to invest in them. You can just have a supply chain relationship or a business development relationship with them.

That actually saves everybody a lot of time. That’s the usual way we’re looking at. If you have one specific SDK or one specific technology, of course, you can establish your own training courses, but the capital investment is a lot more then.

I understand what you meant. To be honest, actually we just started to investigate who can be the partner. I felt like several candidates, demand is more of the university side. Some of the professors are...

There’s the four AI centers, and they’re all located within universities, that’s right.

Some of the professors really want to have commercialization aspect of the AI, because they used to just stick to the very logical, theoretical things.

I’m aware of that.

Some of the professors are willing, but I’m not sure it is the best resource for us.

Then the second could be, let’s say, the AI Academy people. I found that, of course, a number of the students are quite big. Maybe, some of the part of them could be our target, but in addition, actually the mentor side or the lecturer side is very, very solid. This is what I found. I just met with several teachers. Everyone is very professional in terms of AI.

They’re very realistic. Nowadays, there’s a lot of hype. They’re very realistic about the current limitation, and also inspiring the student to explore the frontier. I think that’s the most important part, because nobody knows where this field will go, right? [laughs] It’s important to keep a relationship with the research community, as well.

The one thing is I have to come up with some idea of what aspect I can give to them. They have a very good resource. I have to give something, otherwise I cannot get anything.

Your technology as a SDK or as a API, that’s something super useful.

Maybe, one part could be the technology. We essentially want to hire people, right?

I know, but if people already have experience and know the limitation, as well as the application of your technology, that means that they get hired with better understanding.

Instead of you having to train them, you essentially have the customers train themselves, [laughs] and then become your employees. It’s always a win-win. There’s a Taiwan Technology Arena, or TTA. I don’t know whether you know about this. It is a effort by the Ministry of Science and Technology.

I went to the opening session just last month.

That’s right. They’re mostly working with also international investors and international accelerators to give start-up teams here access to global markets and global capital. Just having the TTA knowing your existence and your preferred topics for start-ups, maybe, you can just come to their demo day. It’s also very useful.

I see.

It’s not like a MOU. It’s just keeping a relationship every couple months with just that. Also, with maybe AppWorks or Taiwan Startup Stadium, basically, all the AI-related accelerators. You can be a consultant or just a friend and come to their demo day.

I see. AppWorks and the TSS, I just know that TSS, I don’t have any contacts.

With your both technology and your offering to help, this paying for it idea, I think it’s best if you just kept direct contact instead of through a proxy.

OK.

I’ll have to take the video conference call now.

OK.

Thank you.