[PYTHON] The full version of Kikagaku's most popular "De-Black Box Course" is now available and will be released for free! Inside story

Introduction

Merry Christmas! I'm Yoshizaki, CEO of Kikagaku Co., Ltd.

It's been two years since I posted an article to Qiita in this way. The flow of the moon is fast.

Then, from the main subject at once.

Everyone, thank you for your patience! !! !! !! !!

Since it was released on Udemy, which you all know in 2017, the Kikagaku style "De-Black Box Course" has been taken by more than 38,000 people in Udemy alone. We have smoothly released the beginner's edition, which deals with differentiation to simple regression analysis, and the intermediate edition, which deals with linear algebra to multiple regression analysis. There is none!

... Actually, I skipped work.

There is nothing like that, but there was a conflict in the story behind this course introduction, and it was difficult to release it.

Such a painful feeling is goodbye today.

Thank you for your patience. Kikagaku's "De-Black Box Course" is now available as a "complete version" that includes probabilistic statistics, logistic regression, neural network mathematics that is the basis of deep learning, and implementation by PyTorch, which is now a hot topic. !! !! !! !! !! !!

Moreover, moreover! !! !! !! !!

The ** complete 12-hour video **, which includes the beginner's, intermediate, and advanced versions that have been brushed up to the 2020 version, is ...

** All are free and open to the public! !! !! !! !! ** (Don) ▶ ︎ Click here if you want to start learning the black box course for free!

I really recommend this course. I personally included all the items I wanted to learn when I was a beginner. And, being able to learn for free means that Kikagaku, including myself, doesn't cost a single yen, so in that case I would only recommend what you really want to recommend.

I will write the inside story of why I decided to release such killer content for free with ** strong feelings for education ** after the course introduction, so I hope you can keep up with me until the end. It's Christmas Eve today, but thank you very much for your relationship.

▶ Click here for press release articles The full version of Kikagaku's most popular "De-Black Box Course", where you can learn the basics of artificial intelligence and machine learning, is now available for free with a 12-hour commentary video!

Introduction of de-black box course

Let's summarize it with "3 Recommended Reasons" like a popular blogger.

Recommended reasons 1. Is the starting point for everything

I think that there are many people who take Udemy, so I think that there are people who know the atmosphere, but what kind of position is the course for the first time is like this.

It is a starting point for those who are going to do research in this field, a starting point for those who start learning by industry specialization, and a starting point for those who are incorporating it into applications.

In other words, I want to study this field, but what should I start with? To the answer

"** Kikagaku de-black box **"

I want you to say.

In a word, it is "** all basics **".

However, not really everyone.

For engineers who want to learn machine learning from now on, I would like them to learn as a gateway to know specific theory and implementation methods. So it's recommended for all engineers.

• Great people in the machine learning area are relatively scared of comments, and knowledge is completely unnecessary, so please refrain from doing so.

On the contrary, I would like the business side to receive it, but when I look at it for three years after providing it, I got the impression that it can be divided into two patterns: those who resonate and those who do not. After exploring the turning point, business people never implement it in the end, so it seems that they feel uncomfortable about understanding mathematics and implementing it in programming.

• Yoshizaki is a Kansai person from Kyoto who occasionally speaks the Kansai dialect (Ayabe Municipal Hatta Junior High School → Maizuru National College of Technology → Kyoto University).

Certainly, it cannot be used for "** concrete **" (this is the point).

Even so, those who feel that it has led to meaningful results on the business side are intermediate to advanced people in the above figure, and from the point of wanting to facilitate conversations with engineers, draw out when designing the business. It was a person who wanted to increase. Almost four years after the advent of killer words such as "artificial intelligence (AI)" and "deep learning," many people in the company will need more or less AI-related terms in their conversations. If you feel that you can't keep up with the conversation, this may be a signal to start learning.

Also, as a feature that is often found on the business side, there are surprisingly many people who are more familiar with specific cases and business models than engineers. The problem with that person's growth is that he has only knowledge of specific cases, and even if he tries to abstract this and summarize his thoughts, he cannot summarize it because he does not have a basic understanding, and cannot organize it as a body of knowledge. In other words, the business side is the exact opposite of an engineer, and it is better to learn the theory to know "** abstraction **" instead of learning to know "materialization". That is why, to some extent, you cannot benefit from "abstraction" without knowledge of specific cases. Even if you are designing a business idea for this AI-related technology, but you are not sure if you are making the best choice because there are not enough drawers, it is a signal to attend.

Conclusion

――The engineer side is the best as the very first teaching material without worrying ――The business side is ideal as a teaching material to learn to broaden the range of ideas after knowing specific cases

Recommended reason 2. Handwritten math and hands-on programming

This way of learning is especially important in my lectures.

The second book was published recently, but I feel that in order to understand mathematics and programming, if you do not convey what you are thinking in that "** process **", the appeal will be halved. Don't do it.

Formulas that may seem difficult are actually filled with the thoughts and intentions of the designer, and the degree of understanding depends on whether or not you can communicate with them. What I think about studying this field is the mathematics "proof finished!" It's not an inorganic world like this, but it's like thinking about it like this and designing it. Due to its nature, it is most efficient to use "mathematical formulas" as the words for dialogue, but this has given birth to people who are allergic to mathematics, and I feel that it is a great waste. If you are allergic to math, please take a look at my lecture. Each formula has the intention of the designer and is introduced so that you can interact with the formula.

The phrase "knowing human thinking through learning AI" is perfect, and the world you know through mathematical formulas is not completely different from the world you live in. Rather, it verbalizes what you think. It's just something you did.

By the way, as an aside, Kikagaku's instructors are unified only to hire inexperienced people who are a little strange, and originally many were from science colleges, but now 70% from liberal arts colleges, about 30% from sciences, and the ratio of liberal arts Has been reversed. If you have someone or a companion to teach you, you should be able to prove that you can learn in any field.

Conclusion

Mathematics and programming explain all the "** processes **", so don't be afraid to take it. It may open up a new world unexpectedly.

Recommended reason 3. It covers up to the latest information

Although this latest is at the moment, it is no exaggeration to say that the highlight of this de-black box course is explained in ** PyTorch ** of the deep learning framework. PyTorch, along with TensorFlow (including Keras), is now the two biggest deep learning frameworks.

The information up to October of this year, but when I looked at the number of searches on Google Trends, it looked like this.

• Please check Google Trends for the latest version.

Japan

world

I was a Chainer sect, but this Chainer also went into maintenance mode at the end of last year, and with the announcement of the move to PyTorch, I inevitably started using PyTorch. That said, TensorFlow can also be used, and there was also a TensorFlow course in the Kikagaku course lineup.

But! !!

In fact, from the end of 2019 to the spring of 2020, the number of TensorFlow students dropped sharply, and that amount went to PyTorch. So, it is good to see the number of searches on Google Trends, but it is also good to see the ratio adopted in the treatise, but it is said that students overwhelmingly choose PyTorch as unique information that Kikagaku has. That's why. By the way, the students are not researchers, but most of them are corporate developers.

With this in mind, I personally think that those who are just starting to learn should choose PyTorch. And the attraction of PyTorch is the "** ecosystem **".

With PyTorch alone, I have to write all of these when training the network, but ...

With ** PyTorch Lightning **, which is certified in the PyTorch ecosystem, you have a lot less to think about.

I also hear the voices of the sect who should write a training loop with a for sentence, but this is a professional idea, and I think it is important for beginners to concentrate on the essence as much as possible. That is why, after understanding the whole thing properly, we will introduce a simple writing style that allows you to concentrate on the essence as much as possible. This course is introduced in this way.

However, there is a risk of ** version upgrade ** to be exposed to the latest information. In fact, while brushing up this course, PyTorch Lightning was officially released from v0 series to v1 series, and the writing style changed to ** significantly **. It's really big. And now, many tutorials using PyTorch Lightning are still written in v0 series. I've always been watching the latest version diffs on GitHub, so I've been revamping the v1 style of writing and should be able to keep you up to date with the latest information at this point.

Also, I introduced the design of a fairly technical network by optimizing hyperparameters with ** Optuna **, so that is also a point that the creator is particular about.

PyTorch itself has a lot of tutorials, and both PyTorch Lightning and Optuna have tutorials, but each one only shows how to use it alone, so it's fine to find the best way to write in collaboration between ecosystems. I have a lot of trouble. I would like you to take a look at that point as it has been cleared through trial and error.

Conclusion

In addition to being introduced in PyTorch, which is a deep learning framework that is currently attracting attention, the latest version also introduces how to write using that ecosystem such as PyTorch Lightning and Optuna.

[PR] Recently, Impress has published a book on how to implement it with this PyTorch, so I'd be happy if you could have it handy instead of a dictionary. Since the content is a de-black box with the basics of image processing and natural language processing, and deployment added, it is easier for your wallet to consider purchasing after learning the "de-black box course" first. Introduction to Deep Learning Implementation Image / Natural Language Processing with PyTorch (impress top gear series)

bonus

In fact, a sequel has already been prepared for this de-black box course, and shooting has been completed. As mentioned in the press release announced today, it will lead to the "Medical AI Specialized Course". Details of this course will be announced in January next year, but it is "extremely good" for those who are not medical professionals but are interested in image processing. Not only the basics, but also the sense of security that the advanced version is prepared is the reason why we recommend it.

We look forward to your participation.

** ▶ ︎ Click here if you want to start learning the black box course for free! **

Inside story start

So, it should be no exaggeration to say that this is the main subject of this article. I will spell out the back side of that decision, why I decided to publish such a hot-selling course for free this time.

A landscape that has not changed for four years from a growing organization

Kikagaku was founded in January 2017, so it's been almost four years and we're entering the fifth term from January. Over the last four years, the company has been able to grow steadily, despite some organizational twists and turns due to lack of power as a manager. I am very grateful for the growth of Kikagaku with the help of many people. It is because of this support that we still genuinely want to be able to deliver better things to society and support more people.

It's a management story, but basically we are conscious of increasing the ratio of equity capital, and only one company that has a capital and business alliance has a small amount of shares, and it is almost 100% our intention. I am able to make a decision. This was also a good steering to build a good system that can take on challenges for long-term profits against the structure of the education industry, which is difficult to change in the short term. As a result, it took about three years, but we have created a training system within the organization that trains professional instructors by hiring inexperienced people, and by feeding back that knowledge to the students, we will build a better education system. I was able to. Personally, can a company that does not have an internal education system really be able to educate outside students? I was wondering, so the background to this is that I have been working to create a company with pride in this regard. I'm relieved that all the companies' projects have been praised now, but I still remember that it was a tightrope walk at first.

While this Kikagaku organization itself has been able to grow for four years with challenging efforts and the support of the boom, there are things that have never changed. That is the level of lectures we are offering to our students. It does not mean the quality of the lectures, but from the first year to the fourth year of its founding in 2017, all companies take this de-black box course called "Deep Learning Hands-on Seminar" for image processing and natural language processing. This is an applied course. At first, we were able to hold it jointly with ** Microsoft Japan **, which gave us an initial speed, and we were able to get certified as an E-qualification course of the Japan Deep Learning Association **, further accelerating it, and ** economic industry. By being certified as the Ministry's Fourth Industrial Revolution Skill Acquisition Certification Course **, we continued to accelerate with a large backup that allows individuals to receive up to 70% of benefits. It is no exaggeration to say that Kikagaku's growth was completed along with this course.

And when the number of students of this course has increased to some extent, I want to release an advanced course and support practical use, but in April 2018 I will release an applied version, but it is difficult for students to gather. .. At the first time, I managed to attract the minimum number of customers, but the students were repeaters who had great trust in Kikagaku and said, "The course that Kikagaku makes is interesting, so I applied without looking at the contents. !! On the other hand, it was not a demand for the advanced version. Two years have passed since then, and I wonder if it will be ready soon ...? I will revenge the application course offer from the spring of 2020, but I decided that it would be difficult to continue because I could hardly attract customers. I think there is also the influence of COVID-19 and the problem of Kikagaku's ability to attract customers, but even after having a meeting with a sales agent, the demand is still only in the "Introduction" course, and discussions continue.

I don't know how many times I've thought about it in the four years since I was founded, but the trade-off between education and business is difficult. Especially when you have a strong desire for "education" like Kikagaku, no matter how much the sales increase, you will not be happy unless the students can actually use it. This AI area may be just a word of "delicious" as a business, but on the other hand, it continues to be awkward as an "education".

First of all, let's thoroughly solidify our feet so that we can take measures to seriously support actual utilization someday. From here on, it's a muddy management story of a company launched by an engineer, and this conclusion leads to this announcement.

Don't get the timing wrong

If you have an educational toothache, you may want to give it away for free as soon as possible and prioritize social "education".

At the stage when teams who value business but are enthusiastic about education have gathered, I thought that this organization might change the society in the future. That is why, if we can establish a mechanism for the organization to survive, we will contribute to the growth of many people by reducing the price so that many people can take it even if it is a hot seller, or offering it free of charge. , I started to work to create a flow that would allow me to take the next level of course. This is a well-known strategy taken by Amazon Web Services (AWS), where if you can offer it to many people, you can reduce the price, if you reduce the price, you can offer it to more people, and you can further reduce the price. Just as AWS has become an infrastructure for IT, in order to create a social infrastructure for "education", we plan to create a structure that will allow business to be established as a result of being disseminated to many people with this strategy. To do.

It was around the beginning of 2018 that I decided to carry out this plan in my mind, and at that time I actually offered it free of charge, including the advanced version of the de-black box course, but from that time on Since I started hiring and started to build an organization, I first solidified a certain foundation, and at that time I gave priority to building an organization in order to pursue the ideal education in the future, while leaving my teeth sick. If Udemy and others provided free versions around this time, I would have been able to earn some income and live a life, but I would not have been able to create an organization like Kikagaku today.

For this reason, I was thinking about releasing it someday, but I kept hesitating to release the intermediate version of the de-black box course.

Long days of continuous preparation for organizational development

In order to build an organization from there, we defined the organization as complete when we could reach the following four steps.

――Step1. Grow instructors who can teach and give lectures on Kikagaku --Step2. The next generation of instructors will grow under the guidance of the instructor I raised. --Step3. I will make a prototype of my company's product and raise it so that a lecturer can make it (instructor → promoted to instructor and engineer) ――Step4. The instructor / engineer I raised will guide you, and the next generation of instructor / engineer will grow up.

• For Kikagaku members, we have picked up only product development so that readers can easily understand it, but we are also thinking about content development for lectures.

It would have been nice to run Step1-2 and Step3-4 in parallel with the hiring of instructors and engineers as separate frames, but the appeal of our products is "domain knowledge." It will take time to create an organization from scratch, so I wanted to take on the challenge of developing strong human resources who can "domain knowledge x product production", so I will work on it from 2018 with a long-term war preparedness. ... and as a result it took three years. It was longer than I imagined. .. ..

Training instructors was quite difficult due to the hiring of inexperienced people, but the difficulty was still higher when moving from Step 2 to Step 3. Until the influence of COVID-19, it was growing as a training business, so if there was time to train engineers who make in-house products from instructors, there was a trend to have instructors take the stage as much as possible to increase sales. For a venture company that imposes its own growth, it is impossible that sales are not growing year-on-year, let alone falling, so it is disappointing that sales may be overlooked. While persuading myself that if sales increase, new instructors can be hired and new inexperienced people can be trained, the number of instructors will increase and the time to train people who can develop their own products will be I also felt that it wasn't there after all ...?

Let's call it the "dilemma of sales and business model transformation."

I came across a book called ** Cash Engine Management ** to make a profit from an existing business and use the profits to invest in a new business without raising a lot of funds, and I went with this management style. I was thinking of doing this. I expected that as the existing business grows, the opportunities to work on new businesses will increase in proportion to the size, but in reality, the opportunities to work on new businesses are inversely proportional to the size of the existing business. I did.

Fortunately or unfortunately, COVID-19 forces a drastic change

That was February 2020. I heard about COVID-19 in the news, but I was aware that the epidemic had come out overseas, and probably many people did. The impact was quicker than I expected, and the Kikagaku style of gathering in the classroom and giving lectures in a face-to-face manner was postponed one after another for "COVID-19 infection prevention". Most of the training from March to April has been postponed, and you may think, "If you postpone it, it's just a shift." Even if the venture company has no income, it is said that it is good to have cash for at least 3 months, ideally 6 months, and considering the office cancellation and the future of employees, the income will be postponed for nearly 3 months. Without it, the situation was quite difficult. Even if the dissolution was the last resort, it was at the level where the management meeting began to consider M & A proposals to group in to larger companies. In fact, there were many companies that asked me, but as a result, I was helped by a large loan from a bank, and the bank staff and CFO were not upset.

Then, from March to June, the training was moved to online, but the reason became more severe from "COVID-19 infection prevention" to "effect of COVID-19", and although I am motivated, I can do it. There were no days. A pinch is a chance, but a pinch is a pinch. I have a stomachache. Nonetheless, product development by Vue/Nuxt, which I had been working on as a new business for a while, has begun to accumulate some knowledge, and maybe I can move on to Step3.

First of all, why not make a "simulator" that can be used for training on the business side for two executive officers (of course, they are also inexperienced groups before joining Kikagaku) ​​who are powerful partners who always deal with new businesses? I will approach you. The training on the business side could only be done in a workshop-style planning system, but after all I can not understand the good and bad points unless I use AI, so I use AI in a virtual environment We have created a web application that gives you a business experience. It feels like a game. It takes 10 minutes to handle 100 cases manually by using a chatbot. If you visualize the frequently-used words for each class, decide the rules, and deal with them, 100 cases will be completed in 5 minutes. However, if it is rule-based, the limit of if statements is one or two layers, and I know how difficult it is to decide rules. Moreover, if you wear frequent words, you will make a mistake. That's where machine learning comes in! It seems that this machine learning will find and decide the rules by data driving from there if you give only the class that was manually assigned first. What's more, it also gives you a value called confidence, so it's an excellent thing that you can sort automatically only when you are confident and switch to manual mode when you are not confident. I've gotten used to the work, but with machine learning, 100 cases take 1 minute and 30 seconds. Moreover, there are fewer mistakes than ever before. I started developing a simulator in March and started developing a simulator in April, thinking that if you understand the advantages and disadvantages of each in this order instead of machine learning from the beginning, you may be able to broaden the range of ideas on the business side. Released in "AI Business Utilization Course". This is a big hit. It can still be held continuously. After all, it seemed that the business side would be happy if they could understand by experiencing while moving their hands. Now, we are introducing this simulator in other courses as well, creating a unique course.

This is not because it is an online lecture, but it was a moment when I realized the importance of "actively learning by students".

Since the development of this simulator was able to cultivate business and improve development capabilities, we also invited two executive officers who are development partners and proposed, "Now, let's create a platform where we can learn online." I will. Compared to the auxiliary product for B that can answer questions in the lecture, we expected that the product for C that keeps moving out of our eyes would be more difficult, so we made a step-by-step decision.

With the results of developing websites and web applications many times, we have built a system that facilitates wireframe design, design ordering, front-end coding, etc., and the rest is long for C. It took a considerable amount of time to design the database, component design, test design, security requirement design, etc. to continue development. In addition, a period of technical research was also set up, and before the overall design, a simplified version of the design content was implemented once for each component, and it was verified whether there was any ambiguity in the understanding. I improved it.

Then, with full satisfaction, it was released as ikus.ai in September of this year, and since this will be the main product of Kikagaku, there was a strong request to name it "Kikagaku", and in December "Kikagaku" ( The name was changed to https://www.kikagaku.ai) as soon as possible. And, alongside that, in cooperation with the two people who are developing, we will prepare in-house training materials so that we can become instructors and engineers in-house, so that we can further accelerate the development system after the release. I prepared. As of December, we were able to form a development team centered on the CTO who was in charge of this development, and finally reached our goal of Step4.

The scenery that I finally arrived at

It's been about three years since I decided to create an organization in January 2018. I was finally able to stand at the starting point of an organization that can do "domain knowledge x product production". At this point, the wonderful members of Kikagaku will develop various contents, and the products will be improved.

** The final finishing touch to this plan was to release the "De-Black Box Course" for free so that many people could take it. ** **

Up to this point, Kikagaku has a trade-off between business and education to grow the organization, and we have continued to provide basic courses that are in demand more than practical use. Over the last four years, other educational companies have also entered the market, and you may have felt that Kikagaku's course content has become ** commoditized **, which should be taken as a fact.

However, for the past four years, I haven't been playing leisurely without renewing the course content. After a long training period of about 3 years, we have put in place a system to pursue better "education" that we believe in.

For today's grown teams, not only basic courses, but also industry-specific courses, courses specialized in actual operations such as cloud computing, courses specialized in theory for research and development, and DX context. The development of a wide range of products, such as courses for business people, has begun.

** I want to become an education company that can support the practical use of more people. ** **

With that in mind, we are welcoming the release of this "De-Black Box Course" as a dawn when we have achieved the plan and as a start for a new Kikagaku.

Right? The inside story is pretty deep, isn't it?

in conclusion

I may have to write more programs at Qiita, but this time I decided that it would be of interest to programmers in the field of "** Technology and Management **", so I love it. I decided to write on Qiita for the first time in two years.

Mysteriously, I have had a lot of trouble since I was founded, but while interest in the field of "** education **" is increasing, I will continue to be involved in the field of education, including Kikagaku.

And, although this article will be published as a Christmas project on December 24, 2020, there is one more finish left as a strategy to continue to grow Kikagaku's organization. This is scheduled to be reported on 2020/01/01, which is the turning point of the 5th term, so please look forward to it.

It's been a long time, but it was a fun sentence to write. I am really grateful to all of you for staying with us until the end. It was my poor writing and real experience, but I hope it is something that I would like to share with many people.

We will continue to do our best for a better society, and we look forward to your continued support.

Merry Christmas!

Kikagaku Co., Ltd. President and CEO Ryosuke Yoshizaki twitter: @yoshizaki_91

Kikagaku is looking for friends who are passionate about education

If you think it would be fun to work for a company that is passionate about education like this, now is your chance.

The day before this article was published was a year-end party, and when I cut out a scene, I found out that it was a team ** that was very difficult to work without laughing at all. If you don't have a smile in your workplace and want to work earnestly, please come to Kikagaku! !! !!

... of course, it's a joke (I'm grinning to make a serious face, haha), ** I'm serious about my work, but I enjoy it when I enjoy it, and it's a very sharp team. **is. This is more natural. Lol

Of course I will do my best for work. However, when you are in your twenties and thirties, it's good to have friends who can share emotions and emotions, including worries.

** ▶ ︎ Check Wantedly for job listings about Kikagaku! **

We are waiting for your application!