I still remember that when I first entered the pit a few years ago, the introductory materials given by the teacher were a bunch of papers! I almost gave up. . .
Nowadays, the development cost of deep learning applications is getting lower and lower, and there are more and more learning materials, so for beginners, they have entered another opposite dilemma-too much data, dazzling people! Even the most classic and top open courses. There are also many doors (Stanford's, Coursera's, MIT's, Berkeley's...). Not to mention the dazzling paper books, not to mention the overwhelming subscription numbers
If you are struggling with how to choose, you might as well take a look at my recommendation first~
First of all, the information I didn’t mention is either Xiaoxi has not read it, or it is hard to say whether it is good or bad, so you have to judge the information that is not mentioned, it does not mean that the information is not good ~
Also solemnly declare! Xiao Xi didn't charge a cent for advertising! this! No! yes! wide! tell! arts! Xiao Xi won't push the information that she hasn't read carefully!
Below, Xiaoxi will recommend it to everyone from video materials, book materials, and other materials.
Video material
In terms of video materials , Xiao Xi does not recommend much, because watching videos is really a time-consuming way of learning. Just mention three courses:
1. Machine learning opened by Ng on coursera.
https://www.coursera.org/learn/machine-learning/
If you have never been exposed to machine learning before getting started with deep learning, then this course is recommended by Xiaoxi. It is very helpful to find out the basic engineering framework of the industry~ It can be said to be an introductory course for both theory and engineering. Of course, those who have already started ignore it.
2. Stanford University's open class CS231n (deep learning and computer vision).
http://cs231n.stanford.edu/
Xiaoxi is doing natural language processing (NLP), so she has heard about this course a long time ago but has never watched it. Recently, I have read a few chapters, and I feel that it is helpful for those who are not in the visual direction to listen to it~ But I personally think that if you are not in the visual direction, and you don’t have a lot of time, you don’t have to worry too much about this course . There are quite a lot of more efficient alternatives to class. It is very suitable for those who are just getting started in the visual direction to brush.
3. Stanford University's open class CS224d (deep learning and natural language processing).
http://cs224d.stanford.edu/syllabus.html
This course is a combination of deep learning and natural language processing. The teacher who started the course is Richard Socher, a genius in the field of NLP, and the creator of almost all kinds of recurrent networks. If you want to get started with deep learning in NLP, this course is very suitable for you, and Xiao Xi has also finished it. After all, if you can’t understand the characteristics of word vectors and other concepts, CNN and RNN can easily become bottlenecks no matter how well they understand them. . However, this course does not talk much about the application of CNN in the NLP field. It is recommended for beginners who are just getting started to brush it up.
4. Neural networks and deep learning opened by Ng on coursera.
https://www.coursera.org/specializations/deep-learning
This course is really popular, and Xiao Xi took the time to read some of it a few days ago, mainly focusing on the second and third chapters. I found that the content of the second chapter of this course is very similar to Ng's "machine learning yearning" , even the illustrations have not changed much =.= For beginners, the content of this course is indispensable for future engineering If it is missing, it is highly recommended to choose at least one of this course and "machine learning yearning"! Especially the bias-variant trade-off theory, no one speaks more clearly than Ng.
To sum up the video materials , if you are a complete novice, then Ng's machine learning is very necessary to watch carefully (it means to watch the code of the big homework carefully! Not just the video!). If you have the basics of machine learning, but deep learning is a noob, and you don’t understand neural networks very well, and mathematics is not very good, then Ng’s Neural networks and deep learning will definitely be rewarding~ If you learn machine learning and deep learning If you have the basics, but lack the application routines in a certain application field, then you can brush CS231n or cs224d if you have time. It doesn’t matter if you don’t have time, just keep reading~
paper books
In terms of paper books, since our goal is the application of deep learning! And energy is limited! Therefore, the effort on books must be less but more refined. Xiao Xi strongly recommends two books, these two books are also the source of 80% of Xiao Xi's knowledge in addition to the thesis~ (including the papers can also account for 40%)
1. "Deep learning" , which is called "Deep Learning" in Chinese, is already on sale (no one should know about it).
I need not repeat the value and authority of this book. but! I want to complain, I think this book is very good except for the mainstream deep learning model, most of the other chapters are quite good! If you read chapters 9 and 10 of this book and feel messed up when learning CNN and RNN, or you feel too long-winded and lack of core, or you feel "Oh my God, how did you pass it by?", then don't doubt your IQ, because Xiao Xi also doubted her IQ =.= . These chapters are mainly used to index papers. (Of course, if you think the speech is amazing, congratulations, you have a tacit understanding with the author)
Still, chapters 6, 7, and 8 of the book are great! In the first part, the chapters on the basics of mathematics are also concise and to the point. In short, the chapters "Deep Feedforward Network", "Regularization", and "Optimization" are worth reading carefully~
In addition, don't think that after reading this book, you will be omnipotent in deep learning. In fact, it is still far away. The value of this book is that it allows you to make up for the main theoretical achievements in recent years at low cost (high efficiency), so that you don't have to face hundreds of important papers. Incidentally, a paper index is provided. The get of engineering ability depends on the following divine book!
2、《Hands On MachineLearning with Scikit Learn and TensorFlow》。
This book is a book that Xiao Xi has kept for a long time! It was also fate when I encountered it. I searched blindly on Google once and found a magical website called "safari books online". My baby has used Safari for many years, but he didn't find that there is such a place! So click in,
Alas? It really is a magical place! Then I habitually looked for books in categories such as artificial intelligence, machine learning, and deep learning. Then I saw hot searches on related topics at that time, and the top1 at that time was this book! Out of curiosity, I clicked in for a trial reading, and then I was poisoned. . .
But don’t worry about the price, as long as you have the title, you will understand (￣∇￣)
This book is really an artifact of deep learning in engineering applications! There are many engineering tricks, such as how to initialize the project, how to choose the activation function, when to use batch normalization and other practical questions can basically find the answer here! It is different from books like "Machine Learning in Practice" that pile up codes, and its theoretical explanations are also great! The organization of the whole book is also persuasive, code and so on are integrated with theory and experiment, and the most important thing is that it will also give a lot of reference papers! The papers will be given at the bottom of the current page. It is simply more convenient to find the New World!
Oh, by the way, the first half of the book is about statistical machine learning, and the second half is about deep learning. Xiao Xi only read the deep learning part. For students who have not used statistical machine learning models such as SVM to solve engineering problems, the first half is also an artifact! So after super beginners finish Ng's machine learning, it can't be more efficient to come to the first half of this book for in-depth study.
To summarize the books and materials , as long as you want to do deep learning, the best theoretical book is "deep learning". But don't be superstitious about the explanatory power of some chapters, just read the paper when you should read the reference paper! Most of the papers will be much clearer than what is written in the book (except for some big cow's papers whose thinking is really difficult for people to understand). As the main book in engineering, Xiao Xi only recommends "Hands On Machine Learning with Scikit Learn and TensorFlow". For super novice, please take a look at the first half by the way. For general beginners and friends with less engineering experience, please read the second half (especially the chapter "training deep neural nets")
other information
1. Projects in Udacity's deep learning course.
https://github.com/udacity/deep-learning
If the deep learning engineering ability only stays at the demo level, and you have not done a large-scale demo or small project independently, and then feel that no one will take it with you, then these projects in Udacity will move you to tears!
The project has typical application scenarios ranging from computer vision to natural language processing. Moreover, each project has already given a framework, the documentation is very detailed, and the code organization of most projects is also very good. The retreat practice that Xiao Xi mentioned not long ago (a long time ago) refers to the projects here! The harvest is very, very big! And Xiaoxi was lucky enough to listen to a few lessons of this course, so don't be too friendly. Students with financial conditions may learn more from Udacity's "deep learning" course than CS231 and CS224. (Should Udacity give me some advertising fees (｡ ́︿ ̀｡))
In addition to the above information, I also compiled the industry's first AI full-stack manual, which is now available for download! !
Up to 3,000 pages, covering AI directions such as the development of large language model technology, the latest trends and applications of AIGC technology, and deep learning technology.
The WeChat public account follows "Xi Xiaoyao Technology Talk", and replies "789" to download materials. Hope to help you!![[picture]](https://img-blog.csdnimg.cn/dd8812ed0c7b414880fe1a79e6db956e.png)