《暗时间》–阅读方法(Dark time — Reading Methods)
1. 趁着对一件事情有热情的时候，一股脑儿把万事开头那个最难的阶段熬过去。万事开头难，因为从不了解到了解基本的一些事实，是一个新知识暴涨的阶段，这个时候的困难是最大的。有人熬不过去，觉得困难太大就放弃了。不过，狂热的兴趣可以抵消对困难的感觉，所以趁着对一件事情有热情的时候，开一个好头是很重要的。（当然，这并不是说持之以恒就不重要了）。当然，也许这个是因人而异的，对我来说我会在对一件事情有浓厚兴趣的时候非常专注地学习，把很多 groundworks 做掉。后面就会顺利一些了。
2. 根据主题来查阅资料，而不是根据资料来查阅主题。以前读书的时候是一本一本的读，眼里看到的是一本一本的书，现在则是一章、甚至一节一节的读，眼中看到的不是一本一本的书，而是一堆一堆的章节，一个一个的知识主题，按照主题来阅读，你会发现读的时候不再是老老实实地一本书看完看另一本，而是非常频繁地从一本书跳到另一本书，从一处资料跳到另一处资料，从而来获得多个不同的人对同一个主题是如何讲解的。比如最近我发现在看蒙特卡罗算法时就查了十来处资料，其中有三四篇 paper 和六七本书；这是因为即便是经典的书，你也不能指望它对其中每一个主题的介绍都是尽善尽美的，有些书对某个主题（知识点）的介绍比较到位，有些书则对另一些知识点介绍得比较到位。而有时候一篇紧凑的 paper 比一本书上讲得还要好。我硬盘里面的书按主题分类，每个主题下面都有一堆书，当我需要学习某个主题的知识时（譬如贝叶斯学习或者神经网络），我会把里面涉及这个主题的书都翻开来，索引到相关章节，然后挑讲得好的看。那么，如何判断一个资料是好资料还是坏资料呢？
3. 好资料，坏资料。好资料的特点：从问题出发；重点介绍方法背后的理念（ rationale ），注重直观解释，而不是方法的技术细节；按照方法被发明的时间流程来介绍（先是遇到了什么什么问题，然后怎样分析，推理，最后发现目前所使用的方法）。坏资料的特点是好资料的反面：上来就讲方法细节，仿佛某方法是从天上掉下来的，他们往往这样写“我们定义… 我们称… 我们进行以下几个步骤… ”。根本不讲为什么要用这个方法，人们最初是因为面对什么问题才想到这个方法的，其间又是怎样才想出了这么个方法的，方法背后的直观思想又是什么。实际上一个方法如果将其最终最简洁的形式直接表达出来往往丢失掉了绝大多数信息，这个丢掉的信息就是问题解决背后的思维过程。至于为什么大多数书做不到这一点，我在这里试着分析过。
5. 有选择地阅读。很多人觉得我读书速度很快，其实我只是有选择地阅读。这里的选择体现在两个地方，一是选择一本书中感兴趣的章节优先阅读。二是对一本书中技术性较弱或信息密度较低的部分快速地略读。一般来说，除了技术性非常强的书之外，大多数书的信息密度很低，有很多废话。一般来说在阅读的时候应该这样来切分内容：1. 问题是什么？2. 方案是什么？3. 例子是什么？如果是需要解释一个现象的（譬如《黑天鹅》），那么1. 现象是什么？2. 解释是什么？3. 支撑这个解释的理由是什么？4. 例子是什么？一般来说，这一二三四用不了多少字就可以写完了（如果假设只举一到两个精到的例子的话），这样的无废话著作的典型是《合作的进化》；那为什么有些书，明明核心观点就那点东西（顶多加上几个精要的例子罢了）却写得长得要命呢？因为人的思维都有一个“联想”的特点，写着写着就容易旁逸斜出，而且作者自己也往往觉得引申出去挺牛逼，有时候很多与主题无关的废话就掺和进来了；那么，阅读的时候就应该有选择性地滤掉这些不相干的废话；此外还有一种可能性就是大量冗余的例子。一般来说组织得比较好的书会有详细且一目了然的目录和索引，根据目录首先就可以滤掉一部分（比如某个子章节的内容你以前是看过的），然后有时候作者还会举很多冗余的例子，如果你已经觉得印象够深刻了这些例子完全可以不看（一些书就非常厚道地对每个观点只辅以一两个最最经典的例子，譬如《与众不同的心理学——如何正视心理学》，这样的书我最是喜欢）。
6. 为什么看不懂？如果看不懂一个知识，一般有如下几个可能的原因：1. 你看得不够使劲。对此古人总结过——书读百遍其义自现。虽然这个规律不是任何时候都成立的，但是从认知科学的角度看是完全可以解释的，我们在阅读的时候，注意力往往会有选择性地关注其中的某一些“点”，而忽略了另一些“点”，于是一遍看下来可能因为某一些忽略导致无法理解整体。或者干脆看的时候就没注意其中一些细节但重要的东西。此外，大脑理解一个东西需要一定的处理时间，人脑的处理速度很慢，神经冲动每秒传输速度不过百米，所以不能指望看到哪懂到哪。最后，我们可能因为思维定势的原因会从某个特定的角度去看一句话而忽略了从不同角度去理解的可能性。对于这类情况，仔仔细细地再多读两遍，多试着去理解两遍，往往会“哦！原来这样。”地恍然大悟。2. 其中涉及到了你不懂的概念。这是技术性的不理解。这种情况就需要 Cross Reference 。如果一句话中用到了你不懂的概念，那就去查，现在很多书都是电子书，直接搜索一下，或者，对于纸书，看一下书后面的索引就行了。奇怪的是很多人看不懂也不分析一下为什么不懂，就直接放弃了。正如解决问题一样，问题卡住解决不了，第一时间要做的就是分析到底为什么解决不了，而不是直接求救。3. 作者讲述的顺序不对，你接着往下看，也许看到后面就明白了前面的了。
（3） Reading method
This article mainly writes some basic methods of learning (especially reading).
1. When you are passionate about something, you can go through the most difficult stage at the beginning of everything. Everything is difficult at the beginning, because never knowing some basic facts is a stage of soaring new knowledge, and the difficulties at this time are the greatest. Some people can’t get through it and give up when they think it’s too difficult. However, passionate interest can offset the feeling of difficulty, so it is important to make a good start when you are enthusiastic about something. (of course, this is not to say that perseverance is not important). Of course, this may vary from person to person. For me, I will study very attentively when I am very interested in one thing and do a lot of groundworks. It will go better later.
2. Access to information according to the subject, not according to the subject. In the past, when you read books one by one, you see books one by one in your eyes. Now you read chapters or even sections one by one. Instead of books one by one, you see Chapters and knowledge topics one by one. Read according to the theme, you will find that when you read, you no longer honestly read one book after another, but jump from one book to another very frequently, Jump from one place to another to get how many different people explain the same topic. For example, recently I found that when I read Monte Carlo algorithm, I checked more than ten materials, including three or four papers and six or seven books; This is because even if it is a classic book, you can’t expect its introduction to every subject to be perfect. Some books introduce a certain subject (knowledge points) in a better way, while others introduce other knowledge points in a better way. And sometimes a compact paper is better than a book. The books in my hard disk are classified by subject. There are a pile of books under each subject. When I need to learn the knowledge of a subject (such as Bayesian learning or neural network), I will open all the books related to this subject, index them to the relevant chapters, and then pick the ones that are good. So, how to judge whether a data is good data or bad data?
3. Good data, bad data. Characteristics of good materials: starting from problems; Focus on the concept behind the method (rational E), pay attention to intuitive explanation, rather than the technical details of the method; According to the time flow of the invention of the method (first what problems are encountered, then how to analyze and reason, and finally find the current method). The characteristic of bad data is the opposite of good data: the details of the method are discussed, as if a method fell from the sky. They often write “we define… We call… We carry out the following steps…”. It doesn’t say why we should use this method at all. People first thought of this method because of what problems they faced, how they came up with such a method, and what is the intuitive thought behind the method. In fact, if a method expresses its final and most concise form directly, it often loses most of the information, which is the thinking process behind problem solving. As for why most books don’t try to analyze it here.
4. Before learning something, first accumulate sufficient “doubt” in the brain. That is, find out what the problem is. Before browsing the method itself, it’s best to ask yourself what method you can think of. It is a recognized fact that the greater your doubts about the problem, the more you think about yourself before, and the more impressed you will be when you see the answer. I remember that the textbooks in the university always have a waterfall view of the whole knowledge structure in front of me. The reading process is very cool. I nodded repeatedly and forgot it soon after reading. Why? Because there is no doubt to learn.
5. Read selectively. Many people think I read very fast. In fact, I just read selectively. The choice here is reflected in two places. One is to choose the chapters of a book that are interested in and give priority to reading. The second is to skim the parts of a book with weak technology or low information density quickly. Generally speaking, except for highly technical books, most books have low information density and a lot of nonsense. Generally speaking, when reading, we should segment the content in this way: 1 What’s the problem? What is the plan 2? 3. What is the example? If it is necessary to explain a phenomenon (such as black swan), then 1 What is the phenomenon? 2. What is the explanation? 3. What are the reasons supporting this explanation? 4. What is the example? Generally speaking, this one, two, three, four can be written in a few words (if only one or two excellent examples are given). The typical example of such nonsense free works is the evolution of cooperation; Then why do some books, which clearly have the core idea of that point (just a few essential examples at most), write so long? Because people’s thinking has the characteristic of “association”, it’s easy to escape when writing, and the author himself often feels that it’s awesome to extend it, and sometimes a lot of nonsense unrelated to the theme is mixed in; Then, when reading, we should selectively filter out these irrelevant nonsense; In addition, there is a possibility of a large number of redundant examples. Generally speaking, a well-organized book will have a detailed and clear catalogue and index. According to the catalogue, you can first filter out some parts (such as the content of a sub chapter you have seen before), and then sometimes the author will give many redundant examples, If you’ve been impressed enough, you can ignore these examples (some books are very kind to supplement each point with only one or two of the most classic examples, such as “different psychology – how to face psychology”, which I like best).
6. Why don’t you understand? If you can’t understand a knowledge, there are generally the following possible reasons: 1 You don’t look hard enough. In this regard, the ancients summed up – reading a book a hundred times reveals its meaning. Although this law is not always true, it can be explained from the perspective of cognitive science. When we read, our attention often selectively pays attention to some “points” and ignores others. Therefore, we may not understand the whole because of some neglect. Or I didn’t pay attention to some details but important things when I just read them. In addition, the brain needs a certain processing time to understand a thing. The processing speed of the brain is very slow, and the transmission speed of nerve impulses is only 100 meters per second, so we can’t expect to know where we see. Finally, we may look at a sentence from a specific point of view because of our thinking set, ignoring the possibility of understanding from different angles. For this kind of situation, if you carefully read it twice and try to understand it twice, you will often “Oh, so it is.” Suddenly realized. 2. It involves concepts you don’t understand. This is a technical misunderstanding. In this case, a cross reference is required. If a concept you don’t understand is used in a sentence, check it. Now many books are e-books. Search directly, or for paper books, just look at the index at the back of the book. The strange thing is that many people give up without understanding or analyzing why they don’t understand. Just like solving problems, problems can’t be solved if they are stuck. The first thing to do is to analyze why they can’t be solved, rather than asking for help directly. 3. The order of the author’s narration is wrong. If you continue to look down, you may understand the previous one when you see the back.