如何系統(tǒng)地學(xué)習(xí)機(jī)器學(xué)習(xí)？知乎里有很多回答，近日，吳恩達(dá)老師親自在知乎回答了這個(gè)問題：

作者：吳恩達(dá)

鏈接：https://www.zhihu.com/question/266291909/answer/2429781356

想要成為一名人工智能從業(yè)者？系統(tǒng)學(xué)習(xí)機(jī)器學(xué)習(xí)是重點(diǎn)！

機(jī)器學(xué)習(xí)是一門不需要進(jìn)行明確編程就能使計(jì)算機(jī)發(fā)揮作用的科學(xué)。在過去的十年里，機(jī)器學(xué)習(xí)已經(jīng)為我們提供了自動駕駛汽車、實(shí)時(shí)語音識別、高效網(wǎng)絡(luò)搜索等實(shí)用工具，并幫助我們極大地提升了對人類基因組的認(rèn)知。許多研究人員都認(rèn)為發(fā)展機(jī)器學(xué)習(xí)是向人類水平的人工智能邁進(jìn)的最好方式。

這里向大家提供三個(gè)系統(tǒng)學(xué)習(xí)機(jī)器學(xué)習(xí)的步驟：學(xué)習(xí)基礎(chǔ)編碼知識、學(xué)習(xí)機(jī)器學(xué)習(xí)及深度學(xué)習(xí)、專注于一個(gè)角色。

想要成功構(gòu)建機(jī)器學(xué)習(xí)系統(tǒng)，基本的編程技能是先決條件。在開始實(shí)踐簡單的機(jī)器學(xué)習(xí)算法之前，你需要具備編寫一個(gè)簡單的計(jì)算機(jī)程序(函數(shù)調(diào)用，for loops，條件語句，基本的數(shù)學(xué)操作)的能力。雖然掌握更多數(shù)學(xué)知識能讓你更具優(yōu)勢，但也不必將精力過多投入到諸如線性代數(shù)、概率和統(tǒng)計(jì)這樣的數(shù)學(xué)基礎(chǔ)上。

在學(xué)習(xí)了基礎(chǔ)編碼知識后，就可以正式開始你的機(jī)器學(xué)習(xí)之旅了。由斯坦福大學(xué)推出的“機(jī)器學(xué)習(xí)課程”是你不錯(cuò)的選擇。該課程提供了對機(jī)器學(xué)習(xí)、數(shù)據(jù)挖掘和統(tǒng)計(jì)模式識別的廣泛介紹，能幫助大家有效構(gòu)建對機(jī)器學(xué)習(xí)的認(rèn)知和理解。主要內(nèi)容包括：監(jiān)督學(xué)習(xí)、無監(jiān)督學(xué)習(xí)和機(jī)器學(xué)習(xí)的最佳實(shí)踐。

該課程從大量的案例研究和應(yīng)用中汲取經(jīng)驗(yàn)，便于大家學(xué)習(xí)如何將學(xué)習(xí)算法應(yīng)用于構(gòu)建智能機(jī)器人(感知、控制)、文本理解(網(wǎng)絡(luò)搜索、反垃圾郵件)、計(jì)算機(jī)視覺等任務(wù)。

此外，深度學(xué)習(xí)也是你需要涉獵的領(lǐng)域。由DeepLearning.AI開發(fā)的“深度學(xué)習(xí)專業(yè)課程”涵蓋了你在計(jì)算機(jī)視覺、自然語言處理和語音識別等領(lǐng)域構(gòu)建應(yīng)用程序所需的知識。你將從醫(yī)療保健、自動駕駛、手語閱讀、音樂生成和自然語言處理等方面開展案例研究，以便于在掌握理論知識的基礎(chǔ)上了解深度學(xué)習(xí)在各行業(yè)中的實(shí)際應(yīng)用。

當(dāng)你對機(jī)器學(xué)習(xí)和深度學(xué)習(xí)都有了較為深入的學(xué)習(xí)后，下一步行動將取決于你心中想要成為的角色，例如成為數(shù)據(jù)科學(xué)家、機(jī)器學(xué)習(xí)工程師或機(jī)器學(xué)習(xí)研究員等，亦或是將所學(xué)的AI技能與你目前從事的工作相結(jié)合，將人工智能更好地應(yīng)用于現(xiàn)實(shí)世界問題。

確定角色之后就要邁入真正的實(shí)踐環(huán)節(jié)了。對此，項(xiàng)目選擇和團(tuán)隊(duì)合作至關(guān)重要。確定可行和有價(jià)值的項(xiàng)目是一個(gè)重要的步驟，必須在你的職業(yè)生涯中反復(fù)實(shí)踐。在完成項(xiàng)目的過程中，團(tuán)隊(duì)合作比單打獨(dú)斗更容易取得成功。與他人合作、提供及聽取建議的能力至關(guān)重要，這能幫助你在協(xié)作過程中建立廣泛的關(guān)系網(wǎng)。當(dāng)你需要幫助或建議的時(shí)候，擁有一個(gè)強(qiáng)大的職業(yè)關(guān)系網(wǎng)能夠助你前行。

在積累了一定的機(jī)器學(xué)習(xí)理論知識和實(shí)踐經(jīng)驗(yàn)后，找到一份相關(guān)的工作看似是每個(gè)人的最終目標(biāo)，但它只是漫長職業(yè)生涯中的一小步。你需要保持自律，不斷學(xué)習(xí)。身邊的人并不清楚你把周末的時(shí)間是用在學(xué)習(xí)還是刷手機(jī)上了，但隨著時(shí)間的推移，他們終將注意到差異。自律的生活可以幫助你在保持健康的同時(shí)繼續(xù)進(jìn)步。

希望上述建議能為你打開機(jī)器學(xué)習(xí)的大門，從初學(xué)者一路走向從業(yè)者。這條路注定是寬闊卻不平坦的，但這一路上遇到的人和事都將助你走向成功。

我的深度學(xué)習(xí)相關(guān)課程也將在近期登錄知乎，敬請關(guān)注，我們下次再見！

請繼續(xù)學(xué)習(xí)！

吳恩達(dá)

英文原文：

Do you want to become an AI professional? The key to machine learning mastery is to approach your learning systematically!?

Machine learning is the science of making a computer perform work without explicit programming. ?In the past decade, machine learning has enabled utilities such as self-driving cars, real-time speech recognition, efficient web search, and boosting our knowledge of the human genome. Many researchers believe that machine learning promises the greatest possibility in realizing human-level AI.

Here, I‘d like to share three steps to learn machine learning in a systematic way:?
First, you should learn coding basics. Second, you should study machine learning and deep learning. Third, you should focus on the role you would like to have. ?
Fundamental programming skills are a prerequisite for building machine learning systems. You will need to be able to write a simple computer program (function calls, for loops, conditional statements, basic mathematical operations) before you can start implementing preliminary machine learning algorithms. Knowing more math can give you an edge, but it won’t be necessary to spend much time on specific mathematical issues such as linear algebra, probability and statistics.

Having gained some fundamental coding skills, you can officially begin your journey of machine learning. My Machine Learning course from Stanford University is a great choice. It provides a general introduction to machine learning, data mining, and the statistical approach of pattern recognition. The course will also help you to develop your practical understanding of how to use machine learning in the real world. For instance, when to use supervised learning, unsupervised learning, and machine learning. ?The machine learning course draws insights from numerous case studies and applications. It is suitable for learning how to apply algorithms to a wide-variety of tasks, such as intelligent robots building (perception, control), natural language understanding (web search, anti-spam emails), computer vision (identifying diseases in medical imagery, finding defects in manufacturing), and much more.

Deep learning is a subset of machine learning that is growing more important, and is worth your attention as well. It uses neural networks to make powerful predictions, and is the driving force behind many of today’s most exciting technologies. For example, self-driving cars, advanced web search, and face recognition all use deep learning. The Deep Learning Specialization, developed by DeepLearning.AI, covers the knowledge you need to build deep learning applications in fields such as computer vision, natural language processing, and speech recognition. You will conduct case studies in healthcare, autonomous driving, sign language reading, music creation, and natural language processing, so you can familiarize yourself with the practical application of deep learning in various industries while mastering theoretical knowledge at the same time.

Once you have learned the foundations of machine learning and deep learning, the next move depends on the role you have in mind. For example, do you want to be a data scientist, ?engineer, or machine learning researcher? Or, do you consider developing AI skills to complement your existing expertise? If so, you can learn AI as a way to better apply your expertise to real-world problems.

After deciding the role, it's time to move on to real practice. You’ll want to get experience working on projects and as a part of a team. Identifying viable and valuable projects is an important skill, and it’s one that you’ll continue to develop throughout your career. The best way to start is to volunteer to help with other peoples’ projects. Eventually you will develop the confidence and experience to lead your own. For completing a project, teamwork is more likely to succeed than solo effort. It is critical to have the ability to collaborate with others, give and take advice, as this helps you build connections. Teamwork also helps you build out your network of professional connections. You can call on people who you have worked with in the past to provide advice and support as you move through your career.?

The ultimate goal, of course, is to find a job in machine learning. This will come after you have acquired both theoretical knowledge as well as practical experience. When looking for a job, don’t be shy about reaching out to people you have met while taking courses or working on projects. You can also connect directly with professionals who are already working in the field. Many of them are happy to act as your mentor. ?Finding your first job, however, is a small step in a long-term career. It is important to cultivate self-discipline and commit to constant learning. People around you may not be able to tell whether you spend your weekends studying or on your smartphone, but day by day, and year over year, it will make a difference. Discipline ensures that you move forward while staying healthy.

I hope these suggestions could open the door to machine learning and help get you job-ready. The journey ahead will surely be a bumpy one, but rest assured that what you encounter along the way will help you succeed.

By the way, courses from DeepLearning.AI will be available on Zhihu soon. Stay tuned and see you next time!

Keep Learning!

Andrew