沈向洋：为何读论文这么难？

本文来自5月20日沈向洋博士演讲公布视频，视频链接在文末。小站对演讲内容进行听写与翻译，如有错误，欢迎指正。欢迎关注文末公众号进行交流。

Thanks again.

I kind feel like that, you know, I have been associated with GIX for a long time and everything was established. And now I got more time to get know many of you, I just have this great pleasure for meeting students. Before this lecture, I am looking forward to see more students.

我已经和GIX合作很长时间了，GIX现在建设得非常好。现在我有更多的时间来了解大家，和学生相处总让我觉得很开心。我期待之后认识更多的学生。

Let me start by saying I am very envious of you studying at UW, Tsinghua, GIX, i am very envious of you who are still in graduate school. I must have you that from my own experience I was in graduate school for a long long time.

我首先想说，我非常羡慕华盛顿大学，清华大学和GIX的你们，羡慕你们仍然在研究生阶段。从我自身来说，我在研究生院待了很长一段时间。

You know, graduate school is the best time in life. Because you actually have time. You probably donot have enough money, but you actually have enough time in Graduate school. So, that’s a good news. It’s also a great time to goof around or procrastinate, but donot tell your parents. So because you actually have the time, so I encourage you to take the advantage of that, and to read a lot.

在研究生院读书是人生中最美好的时光，在这段时间你可能没有足够的钱，但是却有足够的时间学习。这非常棒。当然你也可以在这段时期消磨时光，但最好还是别告诉你的父母。正因为你们拥有足够的时间，我鼓励你们充分利用好这段时间来进行大量的阅读。

In graduate school, I would argue that you should really learn the important life skills like reading, writing, and presenting. But This talk is really most about the reading. There are a lot of good materials, good stuff on the web, that you can actually find on writing, on presenting, and the general tips for graduate students’ lives.

在研究生院，我非常建议大家学习一些重要的技能，比如阅读、写作和演讲。但我们今天的话题主要关于阅读。现在网上有很多好的阅读材料，你可以发现很多关于写作，演讲甚至关于研究生生活的建议。

One of my favorites is actually Simon Peyton Jones, my former colleague at Microsoft research in Cambridge UK. I actually give you this point here about the two presentation s Simon gave, one is on writing, the other one on presenting. Really really nice stuff. You know one interesting thing I actually notice when I was preparing for this lecture is actually how little content on there, on the web talking about reading papers. Perhaps the best material I have found which I use heavily in this talk is from Prof.Michael of Harvard.

我最爱的一部分是关于Simon Peyton Jones, 他是我在Cambridge UK前微软同事。我非常推荐大家去看看他关于写作和演讲的演讲。有趣的是，我在准备今天的演讲时发现并没有很多讨论阅读论文的材料，我找到的最好的材料应该是来自哈佛大学Michael教授的。

So let me get started by why did you need to read the research papers? There are many good reasons why you do that. You need to read some papers, from the required list, from your class, you want to read more papers which acually you need to write a survey. Or you have to do something with the paper, for example, you are asked to review a conference paper, or doing a presentation, or you want code them out for some algorithms. So effectively, reading is the significant parts of your graduate student life. It is positive for your learning.

那我们就首先说一下为什么你们需要读研究论文？这里可能有不少的原因。你可能读一些论文是因为课程要求，或者是你需要写一个报告，又或者是你必须要用一些论文，比如，你需要根据一篇论文进行演讲，又或者是你想要使用论文中的一些算法。总的来说，阅读对于你们研究生来说应该是非常重要的一部分，它对于你们的学习有很大帮助。

So the question here is why it’s so hard to read research papers, I ask a lot of students,I personally trained thirty+ PhD in the last twenty years. some students just get it, many students donot and say it 's really really hard. And many reasons why reading research paper is hard. Well I started by blaming these authors. You know most of the papers are badly written. You have to realize that English is the official language for most of research papers yet. more than half of the authors really have English as the second language in a people like me. Those papers could be bad written. When I look back some my early papers, I really wish I never wrote those papers out there. But it’s too late.

所以现在我们要思考为什么文章这么难读？在过于20年里，我培养了30多个博士生，我也咨询了他们的想法。有一些学生能够出色地完成阅读论文的任务，但也有些学生却表示很困难。这里有不少原因。我首先可能要归罪于一些作者，他们的论文本身可能写的就不好。目前英语是大多数研究论文的官方语言。但是大多数作者的母语并不是英语，英语往往是这些学者的第二语言，比如我。这可能是文章写得很差的原因之一。我现在回过头看看我早期写的论文，真希望当初自己不要写那些论文。但已经太迟了！

So be careful what paper you write. Of course, the papers are difficult to read, you know, because scientific research papers particularly require significant background for those topics, or in the papers. And probably even more difficult is when you get stuck. Where did you get the help, what kind of help can you ask.

所以大家要认真对待写作。当然，一些论文可能由于其本身要求的科研背景而导致其晦涩难懂。又或者，当你读论文时，你遇到了困难，却得不到帮助。你能向谁寻求帮助？你能获得怎样的帮助？

Many of these thing you really donot know, you have to learn over a long period of time in your the all career in the scientific career. Eventually you get good at that. Well that is a very interesting thing about why reading is so hard.

对于这些你不熟悉的科研背景知识，你可能需要在你之后的科研生涯中，花很长一段时间去学习。但最终，你可以掌握它们。所以说，这可能是阅读十分困难的一个原因。

It’s just get more and more difficult to focus. It’s not very difficult now, because of the web, it’s very easy to find something related to something you are reading. It’s much hard to sit down over a long period of time to a lot of form, articles, journals, books, it’s just new way of life. There is a fascinating article in magazine in Atlantic is called the Google making us stupid. So that’s the new life we are in.

另外，现在在网上可以很容易找到与你阅读相关的东西，但是同时，专注阅读变得越来越难。当你需要坐下来花很长一段时间读各种文章，书籍，杂志时，你可能很难做到。在Atlantic的杂志上有一篇有意思的文章说道，Google让我们变得愚蠢。不过，这就是我们现在所处的时代。

So this is actually a quote that coming from Professor Jonathan. When he was a PhD student at CMU, probably around the same time I was there. It’s just really amazing that I really sa this fascinating. Extracting meaning from most of the papers was like sucking a camel through the eye of the proverbial needle upon which a thousand angles were dancing on my head. I wish I could write something right likeJonathanbut you get the idea.

这里引用了Jonathan教授在PhD阶段写过的一句话。Jonathan教授是在CMU攻读博士学位的，同时期，我也在CMU读书。他曾经写下过一句话：从所阅读的论文中提取中心思想，就像一句谚语所说“从针眼里吸出一头骆驼”（sucking a camel through the eye of the proverbial needle）。这个比喻非常恰当，以至于我读到这句话的时候也是眼前一亮。

It’s just very difficult. In his article, he point out three things. how bad people riding CS and math, but they have so many more. It’s not my point in this lecture to tell you more and more model problems sent you get my point here.

Jonathan指出了CS和数学领域写作的三个常见错误，分别是“祖母式”的引言（意思是引言絮絮叨叨，没有直入主题）、段落式的目录结构、不切题的结论。当然，这三个观点对于写作非常重要，但我们今天的主题不是它，我们今天的主题是：如何阅读。

Paper are bad written, but it’s your responsibility to even read this badly written paper. I actually think I know why reading papers so difficult. I think main reason is the disconnect between reading and writing. Because the writer just want to get something out. So the writer is all about what.

有些论文确实写的很烂，但写得不好的论文你也要读，因为有时候你没有选择。读论文之所以很难，最主要的原因是阅读和写作的脱节。作者一心想把东西“拿”出来，而读者只想要获取一点东西。

So many writers are just so exciting about get something out. The reader, however, want to get something out of the paper. So I read these things, so what?Why these things are so important? Why did i get it. This is just the intent between the author’s intent and reader’s learning.

作者沉迷于书写，而读者只想要从文章中获取点什么。然后呢？为什么这些东西很重要，为什么我要知道这个东西？可见，这里面存在着作者想表达的意图和读者想学习的内容之间的偏差。

So understanding can be very different between the writer and the reader. And the understanding can be very different even among the readers. So it has been like this forever, you know, over thousands of years. Everything human being invented, language, and start to write. So there was no technology in the form of feedback loop. Until Web happened, there was never a feedback loop.

因此，读者和作者之间对文章的理解可能是不同的，甚至读者与读者之间也会有很大的不同。事实上，这种现象已经存在了几千年，任何人类创造的东西，书写的东西，都存在这个问题。而在网络出现之前，并没有方法来形成一种反馈循环。

So there was really one way street. Write something and put it out there. and opposite, the readers see that. For us, the Chinese guys, even 2000 years later, we are still debating what Confucious really meant. It’s funny that. Sometimes, we never know what Confucious really meant when he wrote those stupid words stuff. But he never take any feedback I assume.

在过去，我们拥有的一直都是单行道，作者写，读者读。对于我们中国人来说，即使在2000年后的今天，我们仍然在争论孔夫子究竟想表达什么意思。当我们看到他写的那些文字，我们并知道孔夫子究竟想表达什么。但是我认为他并不能获得我们的反馈。

It’s really interesting way to think about similar, I learned that from my friend. When we think about reading and writing is that Shannon’s information theory. It’s really mostly about one way transmission. However, in the from the source to desination from the writer to the reader.

我从我的朋友那里听到了类似的思考，非常有趣。当我们讨论写作和阅读时，会想到Shannon的信息理论。实际上，写作和阅读在大多数时候都是单向信息传输。具体一点就是，从发源地到目的地，从作者到读者。

So writing is like encoding, you write something, you encode the messsage there. And therefore, reading is the decoding, you need a codebook. And this codebook is about agreeable knowledge base between the reader and the writer. So the reader need have the knowledge base, need have the codebook to know what’s going on. If you donot, you have to acquire. It’s that simple.

因此，写作就像编码，你写了一些东西，然后对信息进行编码。之后，读者进行解码，这就需要一本编码书。这本编码书应该是基于读者和作者的共识。读者需要编码书来获取了解信息的所有知识基础。如果没有，读者只能去学习那些东西。就是这么简单。

But to me, it 's really beyond the Shannon’s information theory, as I thinking about reading and writing, because reading is really beyond transmision and compression. It’s more about understanding that user’s intent. It’s the readers want to interpret the intent into some explainable piece, which can be built into the readers cognitive model. And the reader, sometimes, does even know when they need this piece of knowledge. So it’s really important that you think about that reading is practically equivalent to understanding in my opinion.

但是在我看来，我思考阅读和写作的内容远超过Shannon的信息论，因为阅读往往超越了传统的“传输-压缩”框架，它更多的是一个对作者想法理解的过程。这是一个读者想要把作者想法转化为一个可理解的片段的过程。各种片段组合起来将构建读者的认知体系。而阅读等同于理解作者的想法。

Deep learning means deep understanding, shallow learning means shallow understanding. And reading, is not an instinctive skill that we actually born to. reading need to be applied. reading is very important skill to acquire in life.

深度学习意味着深度理解，浅显学习意味着浅显理解。不幸的是，阅读并不是天生的技能，它是一项需要习得的生活技能。

总结：为什么阅读论文如此难？

论文作者本身写的不好。
一些论文需要一些知识基础。
现在这个时代很难长时间专注于读文章。
读者和作者脱节。没有好的反馈机制。

下期文章将分享沈博士关于具体阅读方法的介绍，欢迎持续关注！

欢迎关注公众号“小站精读”。

沈向洋：为何读论文这么难？相关推荐

沈向洋：读论文的三个层次
Datawhale干货来源:AI科技评论,沈向洋博士作者 | 蒋宝尚编辑 | 丛末 5月14日,沈向洋博士在全球创新学院(GIX)课程上曾做了一场线上公开课<You are how ...
【深度好文】沈向洋：读论文的三个层次
5月14日,沈向洋博士在全球创新学院(GIX)课程上曾做了一场线上公开课<You are how you read>,分享他对于科研论文阅读.撰写的宝贵经验,引起一时轰动.由于围观网友太多 ...
沈向洋、华刚：读科研论文的三个层次、四个阶段与十个问题
来源:微软学术合作本文约6000字,建议阅读8分钟. 阅读文章不仅是大家在科研道路上进步的必由之路,也能使我们的心智不断成长,认知模型和思维方式不断完善. 沈向洋博士:如何以正确方式打开一篇科研论文 ...
【转】沈向洋、华刚：读科研论文的三个层次、四个阶段与十个问题
转自知乎,微软亚洲研究院,文章<沈向洋.华刚:读科研论文的三个层次.四个阶段与十个问题> 作者:微软亚洲研究院链接:https://zhuanlan.zhihu.com/p/163227 ...
观沈向洋博士论文阅读技巧有感
观沈向洋博士论文阅读技巧有感前述论文阅读技巧快速浏览标题摘要引言批判性阅读创造性阅读总结参考前述 5月14日,沈向洋博士在全球创新学院做了一堂公开课<You are how ...
覆盖近2亿篇论文还免费！沈向洋旗下团队「读论文神器」登B站热搜
视学算法报道编辑:小咸鱼好困 [新智元导读]无意中发现B站上有个叫ReadPaper的在线论文阅读笔记神器冲上了热榜!ReadPaper由沈向洋博士创办的IDEA旗下团队研发,其收录了近2 ...
搞科研，从好好读论文开始：沈向洋带你读论文了
「或许你永远不知道你以前读过的书能在什么时候派上用场,但请保持阅读,因为阅读的过程也是在你大脑中建立认知的过程.」对于科研人员来说,读论文是一种必修技能.去年,沈向洋博士曾在线上公开课<You ...
readpaper使用+沈向洋讲如何读论文__简单好用
读论文的三个层次 0.readpaper使用 1.直面阅读的困难 2.方法论 2.1 快速阅读--图文浏览 2.2 仔细阅读--批判创造 2.3 读者作者的思考 3.论文阅读实操 3.1 引言 3.2 ...
马毅沈向洋曹颖最新AI综述火了！耗时3月打造，网友：必读论文
白交发自凹非寺量子位 | 公众号 QbitAI 千呼万唤始出来,马毅教授的AI综述论文终于出炉! 耗时三个多月,联合神经科学家曹颖.计算机大牛沈向洋,协作完成. 据本人描述,这篇论文是将他&qu ...
智源社区周刊：LeCun等撰文回应Marcus；朱松纯团队价值对齐工作登Science官网头条；马毅沈向洋等公开AI智能综述论文...
汇聚每周AI观点.研究和各类资源,不错过真知灼见和重要资讯!欢迎扫码,关注并订阅智源社区AI周刊. 观点 Yann LeCun等撰文回应Marcus:当前对符号推理的争论都是边缘问题 [摘编]深度学习 ...

沈向洋：为何读论文这么难？

沈向洋：为何读论文这么难？相关推荐

最新文章

热门文章