[For beginners] After all, what is written in Deep Learning made from scratch?
It seems that it is selling quite well
"Deep Learning from scratch-The theory and implementation of deep learning learned with Python-" published by O'Reilly Japan
It seems that it is selling quite well.
This time, I will briefly introduce what is written in this book.
For those who are worried about buying, I hope it will be one of the indicators.
This book is a very easy book that explains the knowledge necessary to understand deep learning one by one from the beginning.
What is deep learning after all? What are the characteristics? What is the principle behind it? Etc. are explained in an easy-to-understand manner.
There are eight chapters in this book.
- Introduction to Python
- Perceptron
- Neural network
- Neural network learning
- Error back propagation method
- Learning techniques
- Convolutional neural network
- Deep learning
I will explain each content in a nutshell.
Chapter 1, "Introduction to Python", page 20
This chapter will teach you the basics of Python.
It covers everything from Python installation methods to data types, variables, lists, if statements, for statements, functions, and classes in a comprehensive and easy way.
In addition to these basics of Python, we also briefly touch on how to use the external libraries Numpy and Matplotlib, which are essential for learning machine learning.
** Point! "Deep learning is made with Python" **
Chapter 2, "Perceptron", page 17
This chapter will teach you about the perceptrons that are the basis of neural networks.
After explaining the mechanism and movement of AND gates, NAND gates, and OR gates, there is an explanation of XOR gates made by combining these three.
It is convenient because it also describes how to implement it in Python.
If you try to learn logic circuits in earnest, you will see more gates, but I think the good thing about this book is that it contains the minimum requirements that you should know when learning neural networks. I will.
I can imagine that by accumulating these, it will become deep learning.
** Point! "The basis of deep learning is Perceptron" **
Chapter 3, "Neural Networks", page 43
In this chapter, we finally have a neural network (NN).
You can understand that NN can be created by adding the "activation function" to the perceptron mentioned in Chapter 2.
In addition, you can learn how to calculate and use the "matrix" that is frequently used in NN.
If you read this far, you will be able to create your own NN. In this chapter, we are actually building a 3-layer NN in Python to create an AI that recognizes handwritten characters.
** Point! "Perceptron + activation function-> neural network can be created!" **
Chapter 4, "Learning Neural Networks", page 39
In the previous Chapter 3, we just created the shape of the NN and did not do the essential "learning".
In this Chapter 4, the learning part is finally explained.
It starts with the phrase, "The characteristic of NN is that it can be learned from data," and tells us the difference from conventional machine learning.
About 30 pages of explanations of essential ideas for NN such as "training data" and "test data", details of "loss function", and who and why "differentiation" is necessary are described.
At the end of the chapter, the implementation method (code) in Python is described in detail.
At this point, it's almost time to understand deep learning.
** Point! "NN learns from data!" **
Chapter 5, "Backpropagation Method", page 41
There are 6 kanji, unknown words, and you will want to give up.
However, if you can understand this chapter, it is no exaggeration to say that you have understood all the operations of neural networks or the "artificial intelligence (AI)" that is currently being talked about.
The error back propagation method is simply the method of "adjusting the calculation result", which is the "essential" part of NN learning. Therefore, it is quite difficult.
However, this book uses a lot of easy-to-understand examples and illustrations, so it is very easy to understand.
** Point! "Adjust the calculation result correctly by the error back propagation method!" **
Chapter 6, "Learning Techniques," page 38
This chapter has a lot of very useful information.
Techniques that should be incorporated into NN learning, such as "parameter update", "Batch Normalization", and "normalization", are included.
It's good to make an NN and let it learn, but if you are worried that the accuracy will not improve as you think, this chapter will be very helpful.
** Point! "If you get lost, look here!" **
Chapter 7, "Convolutional Neural Network", page 34
The root of the AI boom that has become a hot topic these days is the announcement of this "convolutional neural network".
Convolutional Neural Network is abbreviated as CNN and is used everywhere in image recognition AI and voice recognition AI.
It seems that almost all deep learning methods are based on CNN at image recognition competitions.
It takes time to understand the contents because it is the hottest and latest technology.
However, in this Chapter 7, CNN is explained in an easy-to-understand manner by using a total of 28 figures on 34 pages.
From the explanation and implementation of the convolution layer and pooling layer to the visualization of CNN and typical CNN, it is written in detail, so you can understand all about CNN.
** Point! "At image recognition competitions, almost all deep learning techniques are based on CNN." **
Chapter 8 "Deep Learning" Page 31
This is the final chapter. It's about "deep learning" that was waiting like a last boss.
Deep learning is a deeper and deeper layer of NN that has been explained in Chapters 1 to 7.
Actually, the shape (model) of deep learning has a certain "type". Large companies such as Google are conducting research and finding high-precision shapes (models, layer structures, numbers, parameters, etc.), so we can just use them as they are for high-precision AI. Can be made.
Thanks to Google's smart and great people.
How did the popular models come to the fore in this book? It also describes what you need to do to speed up learning.
Finally, it touches on how this deep learning is actually used and the future of deep learning.
Summary
I roughly wrote the contents of "Deep Learning from scratch-The theory and implementation of deep learning learned with Python-" published by O'Reilly Japan.
It is very readable because it gives a bird's-eye view of the nature, challenges, and possibilities of deep learning.
I think machine learning programmers are a must-read good book!
** Reference book ** "Deep Learning from scratch-The theory and implementation of deep learning learned in Python-"
Learn more about artificial intelligence / AI / machine learning
** "AI for cat allergies" ** https://t.co/4ltE8gzBVv?amp=1 We publish about machine learning on YouTube. If you have time, please take a look.
created by NekoAllergy
Recommended Posts