[2020 version for beginners] Recommended study method for those who want to become an AI engineer by themselves

Service introduction (AI Academy and AI Academy Bootcamp)

It is an AI learning service for individuals that we operate.

①AI Academy Online AI programming learning service where you can learn Python, machine learning, and deep learning almost for free

• Practical content is available for about 300 types for 980 yen per month.

②AI Academy Bootcamp Short-term intensive AI programming boot camp for individuals (AI programming school) Python course (50,000 yen per month) Data Scientist Course (120,000 yen for 2 months) Machine learning engineer course (120,000 yen for 2 months)

Who is the target of this article

・ Those who want to analyze data with Python in the future but want to know what to start with ・ Beginners who are considering engaging in work related to artificial intelligence in the future ・ Those who want to become an AI engineer from an inexperienced person and want to know what kind of knowledge is required for that ・ Those who are thinking about going to an AI programming school or vocational school but want to know how to study on their own

Those who are not eligible

・ Those who have already analyzed data in business ・ Those who want to make web applications using deep learning

• Because we will not discuss how to study technologies such as HTML / CSS / JavaScript, skills required for server construction and Web service development, and a series of stories) ・ How to study for the deep learning qualification test (G test, E qualification, etc.) Etc. are not targeted.

Minimum knowledge required for AI engineers

First, I divided the minimum knowledge required for AI engineers into six parts. Here, when you work as an AI engineer in the future, please roughly understand the overall picture of the basic knowledge of the six contents. ① Programming skills --Python (R, C ++, Julia, etc., if needed) --Indispensable library for data analysis such as numpy, pandas, matplotlib --Scikit-learn, TensorFlow, keras (and other PyTorch, etc.) machine learning libraries and frameworks

Data preprocessing is indispensable for machine learning, but it is a convenient library for data preprocessing. ② Mathematics --Differentiation, linear algebra, vectors, matrices, probabilities, etc. ③ Knowledge of statistics --Standard deviation, variance, probability distribution, estimation, test, etc. ④ Basic knowledge of machine learning --Supervised learning and unsupervised learning --Pretreatment, feature design, learning and evaluation --Simple regression, multiple regression analysis, Lasso, Ridge, least squares, perceptron, logistic regression --Decision tree, random forest, support vector machine, K-means, ensemble learning --Deep learning implementation skills and knowledge --scikit-learn and XGBoost --Knowledge of frameworks such as TensorFlow and Keras --The flow of creating a trained model with scikit-learn.

1. Data collection and data preprocessing Completion of missing values and deletion of outliers)

2. Feature design (feature selection)

3. Model development (model selection and learning)

4. Model evaluation: Cross-validation, evaluation with mixed matrix, etc.

⑤ Knowledge of operating a database using SQL --select, insert, update, delete, where, like, limit, sum, avg, max, group by, having, order by, table join, view, subquery, case, etc. ⓺ Cloud knowledge --Knowledge around cloud infrastructure such as AWS, GCP and Azure There are six big ones, and you may think that there are many, but it is recommended that you first focus on two, ① and ④, instead of doing all at once. The reason is that by actually writing the program and making it visible, it will be easier to continue learning. First of all, if you start from theory, you will be frustrated if you are self-taught.

Recommended way to study artificial intelligence on your own

The necessary knowledge was introduced in the previous section, but how do you learn them? In learning from ① to ⑥, we recommend that you acquire the knowledge in the following order. Phase 1 Learn machine learning programming with python and an introduction to artificial intelligence Phase 2 machine learning programming Challenge Phase 3 Kaggle You will also acquire technologies such as Phase 4 SQL, scraping, and cloud computing. Phase 5 teach I think that learning in each phase is the least burdensome to myself and is fun to learn. Phase 1 refers to programming beginners. If you are new to programming, please take a look from Phase 1. In Phase 2, we will explain how to study machine learning programming based on what you actually learned in Phase 1. If you are already doing machine learning programming using scikit-learn, you can skip it. Phase 3 describes how to learn practical programming through competitions such as Kaggle. Phase 4 Since data is frequently retrieved from the database for machine learning, knowledge of SQL is essential. Here, we will introduce how to acquire technologies such as scraping (for data collection) and cloud in addition to SQL. Phase 5 Teaching people can reveal things that you didn't know. Therefore, it is a good idea to teach machine learning to friends and others to deepen their understanding.

After that, I will explain how to tackle these in each of the five phases, introducing recommended books and so on.

Phase 1 Learn machine learning programming with python and an introduction to artificial intelligence

Self-taught, using books, The minimum knowledge required for artificial intelligence development is "knowledge of machine learning", "knowledge of Python", and "mathematics and probability / statistics for machine learning". In this phase, you can focus on "knowledge of machine learning" and "knowledge of Python", and once move "math and probability / statistics for machine learning" to different phases. Even if you don't understand mathematics or probability, you can experience machine learning by using a library for machine learning, which is more motivated and cheaper than learning theory alone. (Some people may be able to continue from theory, so I will not affirm.) After that, after implementing it by relying on the library, after wondering what is going on inside, it is recommended to go back and study mathematics and statistics. Then, I will introduce __ recommended books. __ First of all, if you are new to programming or are not confident in the basic grammar of Python, the following books are recommended. ・ Python3 introductory note http://amzn.asia/8ccnGQH 2894 yen In this book, you can learn not only the basic grammar that is the basis of python, but also the basics of machine learning programming. After reading through this book, you will be able to experience the basics of python programming and machine learning programming. Also, when you start python programming, you can use it for free with a browser that does not require a development environment Google Colaboratory with a CPU / GPU machine learning environment. Let's do Python programming using notebooks / welcome.ipynb # recent = true) __. Next, if you want to learn more about the outline of artificial intelligence, I recommend the following books. ・ Is artificial intelligence surpassing humans? Beyond deep learning (Kakugawa EPUB selection) [Book] http://amzn.asia/iW1Ms5S 1512 yen For the above two books, read the Python book and proceed with the programs that appear while moving your hands. By the way, if you have solidified the knowledge about artificial intelligence and the basics of Python programming in the above two books, Let's start programming machine learning and statistical machine learning.

Phase 2 machine learning programming

・ Machine learning started with Python Write the code while moving your hand, and then read the code (copying). Rather than just reading a book, it is better to move your hands and proceed. http://amzn.asia/0GskuMa 3672 yen +α http://amzn.asia/9EPUIpg 4320 yen If you've come this far and are curious about what's going on in the library, try implementing a neural network with a minimal library such as numpy. If you are uncertain about your knowledge of mathematics, please refer to the mathematics books introduced below. ・ __ Neural network, deep learning __ Deep Learning from scratch-the theory and implementation of deep learning learned in Python http://amzn.asia/caXkYL3 You will also learn while making a neural network while actually moving your hands. 3672 yen ·Math Easy learning Mathematics to understand machine learning http://amzn.asia/b2XXln9 2786 yen If you are allergic to math, we recommend that you use the two videos in addition to the above. [Kikagaku style] Artificial intelligence / machine learning de-black box course --Beginner-- | Udemy [Kikagaku style] Artificial intelligence / machine learning de-black box course --intermediate level- | Udemy

After watching the above two videos,

Mathematics course for AI: Linear algebra / probability / statistics / differentiation for artificial intelligence learned little by little

When

[Deep learning mathematics in the shortest course](https://www.amazon.co.jp/%E6%9C%80%E7%9F%AD%E3%82%B3%E3%83%BC%E3% 82% B9% E3% 81% A7% E3% 82% 8F% E3% 81% 8B% E3% 82% 8B-% E3% 83% 87% E3% 82% A3% E3% 83% BC% E3% 83 % 97% E3% 83% A9% E3% 83% BC% E3% 83% 8B% E3% 83% B3% E3% 82% B0% E3% 81% AE% E6% 95% B0% E5% AD% A6 -% E8% B5% A4% E7% 9F% B3-% E9% 9B% 85% E5% 85% B8 / dp / 4296102508 /)

Please continue reading.

·statistics

If you are studying statistics for the first time, we recommend the first book, "The first step in statistics that can be used for work even in the humanities". The good thing about this book is that you can read it as if you were reading a manga.

The first step in statistics that can be used for work in the humanities

It's as easy to read as the previous book and as difficult as it was.

[Understanding Statistics](https://www.amazon.co.jp/%E7%B5%B1%E8%A8%88%E5%AD%A6%E3%81%8C%E3%82%8F% E3% 81% 8B% E3% 82% 8B-% E3% 83% 95% E3% 82% A1% E3% 83% BC% E3% 82% B9% E3% 83% 88% E3% 83% 96% E3 % 83% 83% E3% 82% AF-% E5% 90% 91% E5% BE% 8C-% E5% 8D% 83% E6% 98% A5 / dp / 4774131903 /)

・ __ Data preprocessing __ Complete pre-processing [SQL / R / Python practice techniques for data analysis] http://amzn.asia/aa5dHB0 3240 yen Next, let's learn Deep Learning using the library. We recommend TensorFlow or Keras for the library. Keras is recommended for beginners. ・ Keras Deep learning with Python and Keras http://amzn.asia/8Ya7Xu5 4190 yen

2808 yen In addition to books, we also recommend Udemy. Please use the following video as well. Machine learning with Python: Introduction to identification learned with scikit-learn

Challenge Phase 3 Kaggle

By the way, in Phases 1 and 2, we introduced books related to machine learning sold at bookstores. If you move through the above books in order for about 1 to 3 months, you should be at a level where you can read and write the machine learning sequence and the machine learning model yourself. Here, we will practice using Kaggle and SIGNATE (formerly DeepAnalytics) as the next phase. Kaggle is a competition site that brings together data scientists from around the world. The data that companies want to analyze is posted here, and users analyze it and compete for the accuracy of the analysis. The top three companies will also receive prize money from the companies that provided the data. The recommended way to use Kaggle is to use a feature called Kernels. In the kernel, the code description of the model built by other users is written in an easy-to-understand manner for each dataset. As you can see from the actual data analysis with __Kaggle, data preprocessing is very important. __ __ The key here is to be able to use pandas well. __ If you have any questions about pandas data processing, Please review again from pandas Official Tutorial. As a book, Introduction to Data Analysis with Python is recommended.

Phase 4 Acquire technologies such as SQL, scraping, and cloud.

As for actual data, if it is data on the Web, it is often stored in a database, so knowledge of SQL is also required. Here are some books to help you analyze your data using SQL. ・ Progate SQL (for SQL beginners) Progate SQL

・ Analysis using SQL (for SQL beginners)

[Introduction to SQL data analysis and utilization Technology for opening the door of data science MySQL / PostgreSQL compatible](https://www.amazon.co.jp/SQL%E3%83%87%E3%83%BC%E3 % 82% BF% E5% 88% 86% E6% 9E% 90% E3% 83% BB% E6% B4% BB% E7% 94% A8% E5% 85% A5% E9% 96% 80-% E3% 83% 87% E3% 83% BC% E3% 82% BF% E3% 82% B5% E3% 82% A4% E3% 82% A8% E3% 83% B3% E3% 82% B9% E3% 81% AE% E6% 89% 89% E3% 82% 92% E9% 96% 8B% E3% 81% 8F% E3% 81% 9F% E3% 82% 81% E3% 81% AE% E6% 8A% 80% E8% A1% 93-MySQL-PostgreSQL-% E4% B8% A1% E5% AF% BE% E5% BF% 9C / dp / 4802612265 /)

・ Analysis using SQL (for intermediate SQL users) SQL recipe for big data analysis and utilization Data collection may be done from a database or scraped. Scraping is a technology for collecting information posted on websites. Here, we will introduce two recommended books for learning scraping techniques. ・ Recommended for the first book Scraping & Machine Learning with Python Development Techniques Let's Use BeautifulSoup, scikit-learn, TensorFlow ・ Recommended for the second book Python Crawling & Scraping-Practical Development Guide for Data Collection and Analysis ·Cloud Here is a brief description of Google Cloud Platform (GCP). Google Cloud Platform is a general term for cloud services that can be used by paying only for the amount of services such as huge infrastructure and machine learning owned by Google, and when developing services by using GCP. You can use the infrastructure and high-performance data analysis / machine learning services through various APIs. Services that are often compared with GCP include AWS of Amazon and Azure of Micro soft, but please refer to the following article for the differences. Differences between Google Cloud Platform, AWS, and Azure In addition, GCP API can be called with Python, but APIs such as image recognition / video analysis, voice recognition, and natural language processing are provided as shown in the image below. When you make what you want to make, it is easier and more accurate to use the API than to make it yourself using scikit-learn or Keras from scratch. __ You can implement it by searching for the API that suits you, searching the document, Google, or searching with qiita, and referring to the articles that came out. __ PYTHON on GOOGLE CLOUD PLATFORM

Phase 5 teach

Once you've learned programming and machine learning, it's a good idea to teach someone, even friends. As a method of teaching, it is good to hold an event or seminar, use Skype etc. to give one-on-one lessons to friends, or teach directly at a cafe. Besides teaching, I think it would be good to write articles on Qiita and blogs, and write books. AI Academy also holds student exchange meetings and LT meetings, and offers opportunities to teach in public, so please take advantage of it!

Learning roadmap

Please read from the book on the left (STEP 1) to the book on the right (STEP 3) in order. If you already know the basics of python programming, you can skip STEP1. If you already know the basics of machine learning programming and machine learning mathematics with scikit-learn, you can skip STEP2. Start with a STEP that suits your current level. If you have no experience in __python programming, what is artificial intelligence? For those who say, please start from STEP1. __ As I have written many times, the point is that basically the code that appears in the book is actually written and proceeded. For example, when studying English, it is more effective to write, read, and listen than just looking at the teaching materials. The same is true for programming languages, and you can deepen your understanding by actually copying sutras rather than just looking at a book. However, it is not necessary to understand 100% of any teaching materials. It is a criterion for judging when to proceed to the next STEP, but it is enough to understand 70% without understanding everything written in the book. When you get there, let's move on. For details, please refer to the above __Recommended way to study artificial intelligence by yourself __. Those who can start from STEP1 want to be able to analyze XX data! If so, you can skip the part of deep learning by neural network or Keras. Once, STEP1 → STEP2 → Challenge Phase 3 Kaggle → Then, if necessary, proceed to neural network implementation & deep learning library (TensorFlow, keras, etc.). Because you can also use neural networks for scikit-learn, if you haven't decided to use deep learning in your business, or if your goal is data analysis, you don't need to learn deep learning. That's why. However, if you want to get a job or change jobs as an AI engineer, make sure you understand the contents of "Deep Learning from scratch".

For those who want to acquire skills in specialized fields

If you have read all the above books, there are some books (good books) that you should read in order to become an AI engineer (machine learning engineer), so I will introduce them as well.

• There are several types of artificial intelligence, such as image recognition, natural language processing, voice recognition, and generation. Here are some recommended books that will help you acquire the skills of those specialized fields. (* Scheduled to be updated)

・ General deep learning

[Learn while making! Deep learning by PyTorch](https://www.amazon.co.jp/%E3%81%A4%E3%81%8F%E3%82%8A%E3%81%AA%E3 % 81% 8C% E3% 82% 89% E5% AD% A6% E3% 81% B6-PyTorch% E3% 81% AB% E3% 82% 88% E3% 82% 8B% E7% 99% BA% E5 % B1% 95% E3% 83% 87% E3% 82% A3% E3% 83% BC% E3% 83% 97% E3% 83% A9% E3% 83% BC% E3% 83% 8B% E3% 83 % B3% E3% 82% B0-% E5% B0% 8F% E5% B7% 9D% E9% 9B% 84% E5% A4% AA% E9% 83% 8E / dp / 4839970254 /)

・ Natural language processing

[Deep Learning from scratch ❷ ― Natural language processing](https://www.amazon.co.jp/%E3%82%BC%E3%83%AD%E3%81%8B%E3%82%89 % E4% BD% 9C% E3% 82% 8BDeep-Learning-% E2% 80% 95% E8% 87% AA% E7% 84% B6% E8% A8% 80% E8% AA% 9E% E5% 87% A6% E7% 90% 86% E7% B7% A8-% E6% 96% 8E% E8% 97% A4-% E5% BA% B7% E6% AF% 85 / dp / 4873118360 /)

[Introduction to Natural Language Processing Application Development in 15Step](https://www.amazon.co.jp/15Step%E3%81%A7%E8%B8%8F%E7%A0%B4-%E8%87%AA % E7% 84% B6% E8% A8% 80% E8% AA% 9E% E5% 87% A6% E7% 90% 86% E3% 82% A2% E3% 83% 97% E3% 83% AA% E3 % 82% B1% E3% 83% BC% E3% 82% B7% E3% 83% A7% E3% 83% B3% E9% 96% 8B% E7% 99% BA% E5% 85% A5% E9% 96 % 80-StepUp-% E9% 81% B8% E6% 9B% B8-% E7% A5% 90% E4% B8% 80% E9% 83% 8E / dp / 4869541320 /)

[Introduction to natural language processing by machine learning / deep learning ~ Practical programming using scikit-learn and TensorFlow ~](https://www.amazon.co.jp/%E6%A9%9F%E6%A2%B0% E5% AD% A6% E7% BF% 92% E3% 83% BB% E6% B7% B1% E5% B1% A4% E5% AD% A6% E7% BF% 92% E3% 81% AB% E3% 82% 88% E3% 82% 8B% E8% 87% AA% E7% 84% B6% E8% A8% 80% E8% AA% 9E% E5% 87% A6% E7% 90% 86% E5% 85% A5% E9% 96% 80-scikit-learn% E3% 81% A8TensorFlow% E3% 82% 92% E4% BD% BF% E3% 81% A3% E3% 81% 9F% E5% AE% 9F% E8% B7% B5% E3% 83% 97% E3% 83% AD% E3% 82% B0% E3% 83% A9% E3% 83% 9F% E3% 83% B3% E3% 82% B0-Compass-Data- Science / dp / 4839966605 /)

・ Image generation

[Practical GAN ~ Deep Learning with Hostile Generation Network ~](https://www.amazon.co.jp/%E5%AE%9F%E8%B7%B5GAN-%E6%95%B5%E5%AF% BE% E7% 9A% 84% E7% 94% 9F% E6% 88% 90% E3% 83% 8D% E3% 83% 83% E3% 83% 88% E3% 83% AF% E3% 83% BC% E3% 82% AF% E3% 81% AB% E3% 82% 88% E3% 82% 8B% E6% B7% B1% E5% B1% A4% E5% AD% A6% E7% BF% 92-Compass- Books% E3% 82% B7% E3% 83% AA% E3% 83% BC% E3% 82% BA-Jakub / dp / 4839967717 /)

・ Image recognition

[I want to try it now! Machine learning / deep learning image recognition programming recipe](https://www.amazon.co.jp/%E4%BB%8A%E3%81%99%E3%81% 90% E8% A9% A6% E3% 81% 97% E3% 81% 9F% E3% 81% 84-% E6% A9% 9F% E6% A2% B0% E5% AD% A6% E7% BF% 92 % E3% 83% BB% E6% B7% B1% E5% B1% A4% E5% AD% A6% E7% BF% 92-% E3% 83% 87% E3% 82% A3% E3% 83% BC% E3% 83% 97% E3% 83% A9% E3% 83% BC% E3% 83% 8B% E3% 83% B3% E3% 82% B0-% E7% 94% BB% E5% 83% 8F% E8 % AA% 8D% E8% AD% 98% E3% 83% 97% E3% 83% AD% E3% 82% B0% E3% 83% A9% E3% 83% 9F% E3% 83% B3% E3% 82 % B0% E3% 83% AC% E3% 82% B7% E3% 83% 94-% E5% B7% 9D% E5% B3% B6 / dp / 4798056839 /)

・ Data analysis [Data analysis technology that wins with Kaggle](https://www.amazon.co.jp/Kaggle%E3%81%A7%E5%8B%9D%E3%81%A4%E3%83%87%E3% 83% BC% E3% 82% BF% E5% 88% 86% E6% 9E% 90% E3% 81% AE% E6% 8A% 80% E8% A1% 93-% E9% 96% 80% E8% 84 % 87-% E5% A4% A7% E8% BC% 94 / dp / 4297108437 /)

Summary

I will summarize the flow up to this point. Phase 1 Learn machine learning programming with python and an introduction to artificial intelligence Phase 2 machine learning programming Challenge Phase 3 Kaggle You will also acquire technologies such as Phase 4 SQL, scraping, and cloud computing. Phase 5 teach

For those who are confident in English

So far, we have introduced learning methods and recommended books, but for those who are confident in English, we will introduce recommended services. __ * Even if you are not good at English, we recommend that you take Coursera's machine learning. __ ① Coursera Coursera Machine Laerning Coursera is an online education service whose philosophy is to provide the best education to people all over the world for free. In particular, the instructor of the machine learning course is Andrew Ng of Stanford University. Andrew is a researcher in artificial intelligence and is also a founder of Google Brain and a fan of Coursera. A person who works at Baidu's laboratory in Silicon Valley, it is the most suitable teaching material for learning machine learning, so that you can intuitively understand and actually program the main algorithms of machine learning. teach. Andrew Ng gives me a lot of compliments, which makes me very happy. (* The programming language is Octave / MATLAB, not Python) ②Learn with Google AI (ai.google education) Learn with Google AI It's available free of charge in the Google in-house education program. The content is a very easy-to-understand video.

This is also recommended

Udacity「Intro to Machine Learning Machine Learning with Python (YouTube)

For those who are seriously aiming to become an AI engineer

By learning the above contents in order, I think that you will acquire certain basic skills and enter the entrance as an AI engineer. However, there is a possibility that the knowledge is still insufficient to play an active role as an AI engineer in practice. In that case, we recommend that you read the following study methods and their books.

1. Be able to write machine learning algorithms in C ++ and Java languages.
2. "Introduction to Statistical Modeling for Data Analysis: Generalized Linear Models, Hierarchical Bayesian Models, MCMC" Knowledge of statistics above And Introduction to Statistics (Basic Statistics I)
3. Basics of Statistical Learning-Data Mining / Inference / Prediction

At the end

Thank you for reading. We have an online AI programming called AI Academy We operate a learning service and support those who want to learn AI to achieve their goals in the shortest possible time. The following three points are provided as the main service. -Text content that extracts only the necessary parts of the books featured this time (can be viewed from both smartphones and PCs) ・ Chat questions and code reviews (__ * You can ask questions other than AI Academy content. __) ・ Learning progress service (presentation of the shortest route to the product you want to make, presentation of what to do next) I've written it for a long time, but it costs a certain amount of learning cost (at least about 6 months) to learn all the above books from scratch. If you can use a short-term boot camp (online and offline two-axis school) like ours, you can significantly reduce the learning cost, so we would appreciate it if you would consider it once.

If you would like to hear more about AI Academy Bootcamp, please email [email protected].

[Limited to the first 50 people] AI / Data Utilization Package Course

Corporate AI Academy can also be used by corporate companies from 5,000 yen per account. Click here for details

The person who wrote this article

Cyber Brain Co., Ltd. CEO Kazunori Tani We look forward to your follow-up! Twitter Facebook We also run an AI community with more than 5,000 participants. We provide information about AI every day, so we look forward to your participation! Artificial Intelligence Research Community