该不该quit cs博士？决定quit了

科研狗 发表于 2019-2-9 02:36
所以topic需要和老板讨论，什么是好的研究基础，不能太水也不能太难，不然会掉进坑里去。现在老板明确提 ...

The most important thing for PhD study is to find a non-trivial but solvable problem. It is called the thesis. Quite a few people combine lots of trivial sub-problems into a "new" topic and then claim that topic is non-trivial. It is acceptable. Professors still got fundings and students still graduated with PhD. However, if you only want to solve the real hard problem, it's very likely that you won't graduate with a degree. Why? Very likely those hard problems are widely known and even prestige researchers won't risk their academic career just to try to solve them.

jscw 发表于 2019-2-9 13:05
The most important thing for PhD study is to find a non-trivial but solvable problem. It is called ...

lz好像从始自终没有说过要解决"real hard problem"...............
我自己也是做ml的，lz在做的似乎都谈不上trivial啊朋友，要认真读题

jscw 发表于 2019-2-9 13:05
The most important thing for PhD study is to find a non-trivial but solvable problem. It is called ...

？？？？呃你没看帖子？？

发点自己的拙见，楼主想认真做点事情，也并非非要搞啥突破，奈何老板急于灌水。如果老板不支持，其实可以quit。理由就是lz已经可以找到不错的工作，估计本硕的学校均不错无需镀金。读博士无外乎找工作和找教职。如果是划水读完博士，找教职，讲白了自己不舒服还害人，估计lz不会这么想。现在各个学科领域灌水的都是大多数，多数人接受现实，装扮成学者的样子，混饭吃，因为毕竟还是有极少数真学者，所以也没人非要把滥竽充数的赶走。cs薪水不错，lz又可以找到工作，没必要自欺欺人。

科研狗 发表于 2019-2-9 15:25
？？？？呃你没看帖子？？

let me elaborate a little bit more.

most popular ml methods, such as SVM, various decision trees and YOLO, are based on some solid math assumptions or intuitions. Those methods are easy to modify (or to improve), but hard to get results published just because of the modification. It is possible, but just hard (high risk).

Assistant professors are under the pressure of tenure review constantly, and PhD students want to get degree with 3-5 years. It's human nature to avoid those high risk directions.



补充内容 (2019-2-9 17:30):
One popular approach chosen by many researchers is empirical study. For new data set, try method 1,2,...N, and if some method performs way better/worse than all the others, propose an explanation.

想搞科研就看看能不能换导师，想进工业界就麻利点退学

楼主的导师在评tenure阶段，想多灌水是正确策略。楼主不可能要求导师改变自己的策略迎合学生的追求的，只能改变自己这里

EroicaCMCS 发表于 2019-2-7 14:52
LZ太矫情了，身在福中不知福。

首先，“老板是对学生很好”光是这点就超过了70%的老板，更别说他还是大 ...

灌水组对申请绿卡有很大好处，对导师评tenure也是，感觉大方向没问题，不可能为了一个学生调整方向。只能说楼主调整自己这里比较有效

作为非CS PhD真心给楼主点赞！感觉身边很多同学或为生活所迫，或为了别的目的，读博士的确是为了混身份，水文章，等等。个人有个人的追求和难处，外人都没什么好评判的。但是楼主所追求的，也是我认为的PhD真正该有的追求，就是无论大小，在自己的领域做一点小小的创新的突破。说句矫情的话，站在巨人的肩膀上，试着戳破现有天花板，在我心里，才是一个博士生理想的追求。哪怕少有人能做到，但是至少态度上，不应该觉得灌水就够了。调库，评价现有算法等等，我这个普通工科外行phd做起来都觉得容易得很。楼主敢于坚持自己的初心，当机立断，为楼主点赞。
我个人感觉，机器学习在工业界做的也很棒，楼主有心的话，在工业界也可以做想做的事情。祝福～

jscw 发表于 2019/02/09 13:05:26


The most important thing for PhD study is to find a non-trivial but solvable problem. It is called...

想补充2点自己的观点:
1. trivial还是non-trivial这个是很主观的问题 文章被拒最常见的原因就是作者感觉这个问题non-trivial但是reviewer感觉很trivial。。
2. 先找到问题然后以thesis为主是很经典的PhD模式 现在很多非CS的PhD其实也是这样的。。但是CS PhD已经量产 质量控制最有效的方法就是要求paper数量和发表的会议等级 所以最后大家的thesis基本都是四五篇paper拼接在一起。。it’s not optimal but who to blame?