[PYTHON] Half a year when the humanities of Gorigori learned AI almost by themselves

Half a year when the humanities of Gorigori (deviation value 40 first half) learned AI

Hi, it's a gorigori ~~ macho ~~ liberal arts. After graduating from a commercial high school, I went to the liberal arts department, so it's different from other liberal arts. Thoroughbred in the humanities world.

It's important math skills, but I just studied math A until high school and got an introduction to math at university. In other words, the combat power is about 0.1.

Graduated from Nada High School, entered Harvard, and graduated at the top. I will write poems for those who have studied what I wanted to do and who will study in the future.

What has come

I wonder if those who want to start from implementation will be helpful to some extent. For those who want to start with theory, it would be better to go to coursera after learning the basics of small products, linear algebra, and probability statistics. (It is an individual impression.) Self-study is motivated and difficult, so please customize it yourself.

0th month Learn the basics of programming through company training.

I learned Java here and now I can build it reasonably well. It took me about two days after the training to learn the basic grammar of python.

After that, I will study while taking time at the company. And after building the environment, I'm all self-taught. If you use google colab, it's okay without building an environment, so it's actually a solo (?)

1st month keras for the time being

For the time being, I started with a keras book.

This is this guy. Is it so easy just to implement it? That's why the motivation exploded. Thanks to this book, I could understand what to put in what and what would come out. In other words, I don't know, but I can make it. Such a state.

After finishing this book (skipping what I don't understand), I did the official tensorflow AI tutorial.

After that, I briefly studied why differentiation and integration are important.

2nd month coursera for the time being

So next is this.

Very popular cousera machine learning. This is a free online course. I ran this in a month.

It was really good. However, I may have received it because I knew the existence of DL4US in the Matsuo laboratory of the University of Tokyo (^^;

The best impression of receiving cousera is that it was good to understand what kind of mathematical formulas play what role. This is where I realized that "** You don't have to solve mathematical formulas, if you understand the meaning **".

3 to 6 months

From here, I will start eating various places. While implementing it, I was watching qiita and slideshare in my spare time. After that, I like studying while reading a book and watching a video. I will carefully select the ones that were particularly good.

Main story

This book will help you understand neural networks from a simple structure! It is a book like. It was easy to concentrate on the theoretical understanding. However, is the English translation a google translate? I'm really sorry that there is something like that. The content is so good that I studied in an atmosphere.

Video version

Study math with videos.

TRY video lesson. It's YOUTUBE, so it's free and easy to understand. It's for high school students, so it's good to go back to the beginning. I learned only differentiation.

Series to learn from preparatory school. It's YOUTUBE, so it's free. Probability statistics and linear algebra were learned here. I like it because the bokeh is not interesting and interesting.

Of course, cousera's machine learning is also recommended.

Site edition

I want to start machine learning! Machine learning algorithm explanation slide summary Is this statistical machine learning? I read it several times because I can learn about (Is it correct?). Of course, there are many things I don't understand. Atmosphere game.

Qiita! Everyone loves qiita. I like it because excellent people explain it in an easy-to-understand manner and say things I don't understand. I was chasing machine learning tags all the time and reading articles that I was interested in. I'm sorry if I give a good article because there is no end.

[Wikipedia Mathematical Symbols Page](https://ja.wikipedia.org/wiki/%E6%95%B0%E5%AD%A6%E8%A8%98%E5%8F%B7%E3%81% AE% E8% A1% A8) If you see a symbol you don't understand, I referred to it here. Is it because of this that I learned that the unknown symbols that appear in the paper are explained in the paper?

chainer tutorial I only studied mathematics here. You can also use it as a roadmap. If you feel that you have studied to some extent with try and yobinori, you can get a decent comprehensive ability by reading through this math chapter.

Current

Assemble properly! It is a level that can be assembled if it is said. However, I was given a dissertation and implemented this! It's impossible if you say that. I feel that I can understand the formula to some extent while looking at it.

Looking back on what has come

If you put it together like this now, it looks like you've walked straight down the road, but in reality, you can hit the wall many times and make detours, groping around and finally improving the road. It feels like it's done.

At first I didn't understand the formulas and codes, but I want to know because something amazing works! I wonder if that was big.

What I want to do from now on

I've been following all the formulas so far, so I want to improve my implementation skills. Specifically, implement the self-made layer, self-made loss function, and self-made optimization function with keras.

For those who are going to study by themselves

First of all, I want you to understand that self-study is a difficult skill, but it is a skill of a living thing. Because if you have the internet, you can study anything by yourself.

Below, I will list some of the things I learned while studying on my own. ~~ There was only one ~~

1. Understand what you don't understand.

The worst thing about self-study is that I'm motivated because I don't understand something. It's not good to be motivated, so let's understand why you don't understand and what you don't understand and take measures.

One also says. "** Clarify the reason you don't understand ", and " Take measures ". I will say it again. " Why " I don't know, where I don't know, and " Countermeasures ". I will say it again. " Cause " and " Countermeasure **".

If you don't understand and want to start screaming, remember the "** cause " and " countermeasure **".

Finding the cause is two steps. Find out where you don't know "** where " or where you don't. Write a sentence about " why **" that you don't understand. only this. Along with programming error resolution.

There are various measures. I think it's okay if you can determine when to stick, when to use another teaching material, when to go back, and when to skip. I feel that if the cause is known, it will be decided which measures to take.

Especially when you don't understand mathematical symbols or mathematical formulas, it's easy to get motivated, so if you don't understand even if you look it up, leave it alone and study another. You can't overdo it.

Finally

I posted it on qiita for the first time, so I think there are various points, but it's easy.

If the theme is AI, it's mathematics! Math! However, if you just want to build something simple, you don't need that much, so feel free to do it. Of course, I want to read the paper and implement the latest NN! It's wonderful, so let's study together.

I hope people of liberal arts have hope.

Postscript I'm glad that the article has reached a lot of people. Thank you for watching.