We are in an era of "technical explosion" and "shared, open source". The rate of update and iteration of advanced technology has exceeded any period in history, and these technologies are no longer occluded, and everyone can access and learn. Lifelong learning is a problem that every one of us has to face. This is especially true in the field of big data/artificial intelligence: the endless stream of new technologies has brought us convenience, but it also makes us difficult to be efficient. The dilemma of learning and choice. Therefore, to learn big data knowledge in this era, we need to have appropriate logic and methods.
This article attempts to help readers to use all kinds of "shared, open source" learning tools and learning channels, to avoid the "deep pit" that all kinds of novices are easy to enter, and to complete the target technology learning with minimum time cost and economic cost. And mastery. This paper first analyzes the background of the times, and then divides the current talent echelon in the field of big data. Finally, it gives an advanced guide for big data/artificial intelligence talents from rookie to master.
First, the background paving"Technology explosion" and "shared open source" are the most distinctive labels of this era. The author believes that the two are mutually causal and closely related. First of all, in the era of "technical explosion", for the research team at the forefront of technological development. In other words, the best means of "technical realization" is "shared open source." In contrast, before the Internet and mobile Internet matured, the information was very closed. Once a certain technological innovation appeared, it needed to register the patent for the first time. The technology needs to be protected by the government. The only way to realize the technology is to sell the patent or organize the production to form a product. .
Nowadays, the Internet and mobile internet have been very mature, and new information will be transmitted to every corner of the world in a very short time at a very low cost, so the research team at the forefront of technology only needs to be in the first place. Uploading your own work results to a neutral shared and open source website such as “arxiv†or “github†will immediately be protected by global public opinion, which is far stronger than the patent protection of a certain country.
Then, as long as the new technology has application value or academic value, all kinds of capital giants, technology predators and related organizations will line up to send out a generous offer. For the frontier team, the time for technical realization It is far ahead of the point of time when technology is being productized.
Secondly, because the “technical explosion†always has new technologies waiting for the front-end team to research and discover, the best way for the front-line team to stay ahead is not to keep the existing results, but to “share open source†as soon as possible, and then invest To new research work.
Finally, “shared open source†has also greatly promoted the “technical explosionâ€. No matter the long-term development of any technology and technology, it needs a huge talent system to support it. In contrast to the various periods in history, the channels for sharing knowledge and cultivating talents It is mainly a "school". This channel is not only in a single form but often has a considerable threshold. It will block a considerable number of "involuntary youths".
In today's era, the fastest channel for knowledge dissemination is the Internet. Because of the “shared open sourceâ€, the world's best quality educational resources and the most advanced academic and technical ideas suddenly have no threshold, and are open to all individuals without distinction. The result is that as long as a certain technology, technology field has made a great breakthrough and has broad application prospects (such as big data, artificial intelligence), then the corresponding talent team will automatically fill in in a short time.
The research team standing at the forefront of big data academics only needs to push forward the frontiers. The talent team will then automatically carry out "guarantee" work such as "new technology demonstration" and "technical productization" to protect this technology field and related industries. The healthy development will further promote the convergence of resources to the front-end team of the pyramid and support its pioneering work.
We divided the talent echelon mentioned above into four levels: rookie building foundation, initial entry into the rivers and lakes, boarding the room, and Huashan theory sword:
Rookie building base: At this stage, the talents are mainly based on the study of big data basic theory, and they are still not qualified for real projects or work;
Initially entering the rivers and lakes: At this stage, the talents have already possessed the ability of preliminary big data practice, and it is recommended to improve the learning through practice (doing projects, playing games, etc.);
Entering the room: The talents at this stage need to have the research, reading and understanding ability of big data research papers, and can successfully reproduce the algorithms in the paper;
Huashan's sword: At this stage, talents can independently carry out research on big data and new technologies, and have the ability to publish original papers.
The following sections will give different suggestions for cultivation and upgrading for big data talents at different stages.
Second, the rookie base1. The best resources are often open
After reading the background, I believe that I don't need to explain why the best resources are often open, and directly give some channels for obtaining high-quality resources. First recommend three foreign websites, namely "Coursera", "Arxiv" and "Github".
Coursera is the world's leading online learning site, founded by the industry's highly academic and shared spirit. Courses on Coursera are relatively basic and should be the best platform for "Little White" to take off. Here, Andrew Ng's "machine learning" and "deep learning" are recommended. For domestic students, the biggest problem may be English. It is necessary to be clear here. If you want to be a true master, then English is the eternal entanglement. The latest and best information in the industry is without exception. In English, even the top experts from China will not choose to use Chinese when they publish their papers.
In fact, for the vast majority of people, English should not be used as a "discipline" to learn, but should be used as a "tool." There is no shortcut to the specific approach. Just see the words that you don't understand and check them immediately. The words don't need to be memorized. You won't check them again next time. Everything aims to quickly understand the meaning of the sentence.
Arxiv and Github are two websites/tools that readers will use in the future. Arxiv has the latest and most comprehensive shared papers. The papers will explain the various algorithms in detail, and the latest and best open source code on Github. The code is often the implementation of an algorithm. There are many tutorials on the specific usage method, and we will not expand it here.
Readers can easily understand that Arxiv is the place to practice internal strength, and Github is the place to practice external work. It is impossible to solve practical problems only by practicing internal strength and not practicing external skills. However, it is often impossible to practice external skills without practicing internal strength, and must be both internal and external. Finally, I will introduce you to a magical website called "gitxiv", which will help you find the correspondence between the paper and the code.
2. Don't read books, don't read books, don't read books
How to get started with a subject? In the face of this problem, the rookie is the easiest step to step into the "deep pit" is to find an authoritative book to learn from the beginning, once you step into this pit, you will ruin your own weeks, and then focus on a certain subject Completely disappointed for life. First of all, there are not many books, and often they are not available. Secondly, even in the case of a good book, in order to ensure academicity, the language in the book is often "rigorous" but difficult to understand, and will lay a "solid foundation" for readers from the early history of the discipline, and will stop abruptly when it comes to the latest technical means. Finally, even if the reader has spent a few months of skill and insisted on reading it, the author can tell you through the practice of Bloodlinlin that the contents of the first half of the book will definitely be forgotten.
Of course, there are also special circumstances. If you have already determined your research direction and have a high-level person/mentor to give you a list of books that must be read in the corresponding fields, this kind of book is worth seeing. However, you should also pay attention when you look at it. Don't get caught up in some details. If you can't understand it, you can write it down first. Such details will often come to light under the specific scenes in the practice process.
The correct approach can be summed up in one sentence. Good books are used for investigation rather than for embarrassment. When do you check it? The following will be answered step by step.
3. Find a good friend, even roll to go forward
Now it’s not a single fight. After falling off the cliff, you can find a time when the cheats can be retired for a few years, and it’s like the Hinton (the father of the BP algorithm that overthrew the BP algorithm), or like He Kaiming. (Best paper is as easy as a normal person to send paper.) These rookies are all explored with their friends in their very reliable team. Good friends don't need much, one or two really reliable is enough, as the importance of teammates will be slowly explained.
The final recommendation of this part of the rookie building is that you should not stay at this stage for too long. Don’t wait for “ready†and start practicing, because the “ready†here often includes the lack of confidence of the rookie. Not to further improve yourself is always prepared not to be "good." Under normal circumstances, students who want to do "computer vision" or "natural language processing" and other AI directions after completing Wu Enda's "Deep Learning" course, students who want to do "data mining" are completing Wu Enda's "Machine Learning" course. After that, you can choose the corresponding practice project to prepare for the next stage.
So what practical methods should we choose? The best situation is that there is a big god to lead the team to do real projects, but such opportunities are often met and not available, and will not be discussed here. The popular method is to participate in a big data competition project. Now the domestic “Ali Tianchi†and the foreign “Kaggle†are open big data competition platforms. There are various real projects released by various organizations on the platform. Everyone practices and competes. There may be a lot of doubts in your mind here: "Even if you have learned the basic course, can you get started without anyone taking it?", the following will continue to answer the question of how to "continue to climb".
Third, the first entry into the rivers and lakes1. Find the highest baseline
The "baseline" here can be understood as the reference that the predecessors have already made when they happen to need to do the same work. For the above mentioned situation, if there is a big god to lead the team to practice, then the previous experience of the team god has become the "baseline" of all the team members. Does it have a more general solution for the majority of readers who do not have the resources of "Great God"? The answer is yes. If the reader is currently unable to start with a type of problem, such as a course that has just completed the "deep learning" course, but does not know how to do the "natural language processing" project, the best way is to make good use of the domestic "Wanfang" and The paper inquiry platform such as “Knowledge Network†is used to query the dissertations of domestic universities in related fields. Most of these papers are in Chinese and will introduce a large amount of basic background knowledge in the paper, which just meets our needs.
If the specific knowledge of a certain technical side is unknown, for example, in the direction of "natural language processing", but do not know much about "LSTM", you can make good use of domestic such as "know", "book" Knowledge sharing websites such as "CSDN" can find corresponding blog posts or answers as long as they are not too new. The common skill of using the above two types of channels is to compare several articles and compare them. The same concept or technology, an article is difficult to fully describe clearly, and because the author of the article is different, the starting point for explaining the problem is not the same, so if you encounter a situation that does not understand an article, don't be impatient, then look The next article is just fine. In addition, the “good book†mentioned above can be used here to check. The readers will find that the knowledge points that you want to remember can’t remember, as long as you “check†and “useâ€, then you generally forget to forget. Can't fall.
The definition of the so-called "high" in the baseline is that the closer to the academic front, the better the practical effect, the more "high". In general, the more "high" the results can be referenced, the less Chinese literature.
I don’t know if I answered the questions raised in the previous chapter. The “rolling and crawling†mentioned in the previous chapter refers to the fact that after we select a certain practice direction, we will return to the final result according to the practice. To carry out the process of “checking and filling the gap†for our relevant knowledge. This kind of learning process is more targeted, and the participants are completely targeted to learn. The things they learn can be practiced immediately, thus avoiding the embarrassment of “learning and forgettingâ€.
2. Reasonable pursuit of quick win
The author has carefully studied why girls go shopping "indefatigable", the answer is that girls go to a shop, look at the shoes / clothes / bags in the store to get a certain excitement, get a point of excitement Then I thought about going straight to the next excitement. Analogy To the process of our project/play competition, we need to set up such “exciting points†for our team, so that team members can enjoy the “quick win†pleasure to support everyone to continue.
The key to achieving "quick win" is to divide the work/task in the hand into several sub-tasks that can be reached with a little effort. The details are too complicated and will not be discussed here. The most important thing a teamleader needs to do is to help the team to rationally divide the task and continue to get "quick win". As long as one has such ability, no matter how high or low the technology can unite a group of like-minded partners.
3. Your biggest motivation is often from DDL (Deadline)
There is such a successful sentence that is "there is not an alarm clock but a dream every day." This sentence sounds very inspirational, but for 90% of people it is nonsense, we look back and find that we wake up every day. It is often "the wages that are deducted after work is late" or "the murder of the boss after the lab is late". This is reality, it sounds cruel but we can make good use of it. Specific to our upgrades and project advancement, the biggest driving force for us to continue to move forward is often "the contempt of the small partners after the DDL can not complete the task" and "the sense of accomplishment after the completion of the quick win."
Doing this well, in addition to the rational division of tasks mentioned in the previous section, the most important thing is that there is a reliable team leader that constantly pushes, and every time after the established node, it can't move. Finally, one sentence, according to Maslow's hierarchy of needs, dreams should belong to the "self-realization needs" at the top of the model. If a person can be awakened by "dreams," then the other needs of this person should have been well met. So I sincerely wish that everyone will be able to wake up in the morning by their "dreams".
Fourth, boarding the room and Huashan swordIf one day you find yourself working in a workplace, you need to keep paying attention to the most cutting-edge papers, and you need to constantly try to reproduce the algorithms in the paper for practice. Congratulations to everyone who has entered the field of big data/artificial intelligence. The ranks of the masters. The distinction between the two stages of entering the hall and the Huashan sword is not particularly obvious. Because the papers are read a lot, there will always be some new ideas of their own. These ideas can be published after being verified by experiments. Conversely, even if you have published cutting-edge papers, you still need to follow up on other papers.
1. The circle of friends determines the height of your life
At the beginning of this section, the author first has to put out a bowl of poisoned chicken soup. Even in this era of “open source, sharingâ€, the distribution of academic/technical resources is extremely uneven, and such unevenness will become more and more obvious. There are two reasons for this. The first reason can be quoted in the words of a school leader at the Tsinghua 17 Graduate School Opening Ceremony----"The most effective research method is to communicate face-to-face with a considerable level of peers." The translation is that the more the masters, the easier it is to produce masters, which leads to the uneven distribution of high-end talents.
In addition, the economic cost of doing academic frontier research is very high. The cost of electricity for a global AI company's global R&D work can reach 10 million. Even for ordinary AI projects, the cost of servers and GPUs can cause ordinary researchers to find sufficient funds to support their research.
After drinking the poisoned chicken soup, some positive energy is also coming. Although the resources are not evenly distributed, the talent channel is still open, but the threshold is getting higher and higher. I will be able to work while I am four or five years after graduation. After reviewing the graduate students of Tsinghua University, I finally came into contact with examples of cutting-edge scientific research.
2. Choice is always more important than hard work
This title sounds like a bowl of "poisonous chicken soup", but this is the experience of the bloody forest life brought to the author. I have seen an algorithm team who has been boring for a few months and has not progressed in research. After the completion of the big god on demand, it will be completed within one month.
Let's take a more dramatic example. Natural language processing has been a fierce battle between two schools with distinct boundaries around the 1970s. It is a "rule school" that wants to use speech rules to make speech recognition. One is a "statistical" based on statistical methods. These two scholars who have been engaged in research in the same field have actually held their own academic conferences. Even if they attend the same conference, they have to open a small meeting.
By the 1990s, the recognition rate of the “statisticalists†had reached more than 90%, and the “rule school†was less than 70%. The winners and losers had already been divided (Mr. Wu Jun’s book The Beauty of Mathematics) Detailed and interesting explanations have been carried out). But if a Ph.D. student defined his academic direction as a “rule school†in the 1970s, what should he think of in the 1990s?
After the stage of “going into the roomâ€, it is especially important to make a choice. This choice is not limited to the academic direction, but also covers a broader scope such as “doing academics†or “making industryâ€. One experience that can be referenced is that if a major decision-making mistake in life is made, it will take five years to recover. You must consider that you have several such five years.
3. The only restriction is often your own compromise
Looking at the title of this section, readers may feel that this article is going to end with "Poison Chicken Soup." But in fact, "compromise" here is not a derogatory term. The author believes that it is at least a neutral word. In a sense, everyone will eventually reach a certain kind of "compromise". Without compromise, it means that there is ambition or desire behind the status quo. When the ambition and desire match the reality, it will definitely " compromise". This is the secret of Huashan's sword. Everyone who can stand on the peak must hold some kind of ambition or desire beyond ordinary people. Of course, the ambition or desire here is broad, and it also refers to the pursuit of scholarship.
Finally, the "Poison Chicken Soup" does not live up to expectations. According to the author's observation, each person's "compromise point" is not set by himself. Under normal circumstances, he can't influence himself, so the height that everyone ultimately has to go is often definite.
However, from the author's point of view, I don't think it is good to stand at the top of Huashan. The real "good" is to be able to accept my own "compromise point" and be able to settle down in my own "compromise point". It is the most intelligent choice to work and live happily.
Bldc Ceiling Fan Pcb Price,Worcester Bosch Fan Pcb,Bldc Fan Controller Pcb,Pcb Mount Fan
Full Industrial CO.,ltd. , https://www.iotaindustrial.com