1、简介

来源于 

Reddit上有一个帖子，一个小哥读了博，主要研究领域为计算机视觉和自然语言处理，但他每天却都很苦恼，感觉自己不是在做科研，而是在做工程！

Hi, throwaway because everyone in my lab uses reddit.

I am doing a PhD in machine learning but my field is heavily based in computer vision and also some techniques from natural language processing, so I'm mostly doing deep learning.

I have some conference contributions, but none of them in major conferences. Reviewers are always fairly critical but I have not gotten a rejection yet (though last time was pretty close).

I get why they are critical too. I'm not a top student, our lab is not a top lab, and what I do is mostly repurpose existing methods for different domains. Think taking a ResNet and applying it to medical imaging, or transformers for music classification (not actually my domains).

I feel like compared to many others, I heavily lack in mathematical background even though I try to read up, I often immediately forget concepts that I don't actually apply. I couldn't tell you what the rank of a matrix is, let alone how to use it.

This is partly why I don't really come up with new methods. I'm better at combining existing stuff, but it doesn't feel like research but more like engineering at times.

Because my contributions are fairly underwhelming, I don't think I will be able to achieve a career in academia. So I will likely look for a job in the industry.

But there I would like to be able to show something more than "I applied method X to data Y and got a slightly better result so I published it".

Do you have any tips for (1) growing beyond the niche of your PhD, and (2) making actual contributions that are not purely incremental and applied during your PhD?

Perhaps side projects that I should do if I have some left over energy in the weekend?

文中他说自己正在攻读机器学习的博士学位，主要从事深度学习，并且还基于此有了一些会议上的论文，但是这些会议并不是公认的顶级会议。

尽管不是顶会，但是审稿人对他的要求和回复还都是相当苛刻的。

他也明白他们为什么这么挑剔，坦言自己不是一个优秀的学生，实验室也不是一个优秀的实验室，所做的主要工作就是重新使用现有的方法为不同的领域，例如使用一个 ResNet 用来做医疗成像，或是基于Transformer的音乐分类。

和其他人相比，他自认为严重缺乏数学背景知识，即使努力阅读，也经常立即忘记那些实际上并没有用到的概念，甚至都不会计算矩阵的秩是多少，更不用说如何使用它了。

这就是为什么没有真正想出新方法的部分原因，更擅长组合现有的东西，但有时感觉不像是研究，而更像是工程学。

因为贡献比较平庸，所以小哥不认为他能够在学术界有所作为，所以未来可能会在这个行业内找一份工作，而不是去大学谋一份教职。

最后这位小哥希望网友们对他的博士生涯给他两个建议1）博士之外兼职2）非科研成长

2、精选回答

可以参考学习 https://github.com/shengyp/doing_the_PhD

详细笔记+严格编程+保持良好习惯

 每天进步一点点

 耐得住寂寞