非科班投稿NIPS小经验分享

谢谢楼主的精彩分享！你可以继续分享一下读paper一些经验吗？比如，如果确定阅读的顺利和读什么？一个领域的paper有时候很多，不知道怎么取舍。另外，你是直接刷conference paper list吗？

greyman 发表于 2019-9-5 09:04
我的经验可能只适合机器学习里面的算法派：半理论+半实验。事后想来，幸运的是选了贝叶斯这门很花时间但收 ...

感谢楼主分享 已加米

如果是未来打算去业界的PHD，想刷kaggle攒名次，哪些课比较重要呢？

gongchen 发表于 2019-9-11 13:01
感谢楼主分享 已加米

如果是未来打算去业界的PHD，想刷kaggle攒名次，哪些课比较重要呢？

data mining/machine learning + 算法应该足够了。

Kaggle技巧大概是：如何调参，如何控制learning rate，L2, noise等，如何数据预处理，以及ensemble。

jeremyddd 发表于 2019-9-11 09:33
谢谢楼主的精彩分享！你可以继续分享一下读paper一些经验吗？比如，如果确定阅读的顺利和读什么？一个领域 ...

最好有一个有经验的老师帮你指出来适合你的大方向。然后你就刷那个领域的paper就好了，需要做什么contribution以及如何解决就需要自己下功夫了。我来中部某佛学校，accepted paper里面有Bayesian那个就是我的啦，现在还没放arxiv。

Hi, thanks for your share, could you please list math courses from priority orders?

greyman 发表于 2019-9-11 22:05
最好有一个有经验的老师帮你指出来适合你的大方向。然后你就刷那个领域的paper就好了，需要做什么contrib ...

谢谢楼主！再次恭喜啊！有啥其他疑问再来想你请教哈 ：）

jeremyddd 发表于 2019-9-12 23:55
谢谢楼主！再次恭喜啊！有啥其他疑问再来想你请教哈 ：）

忘了说了。一定要各种忽悠老板把计算资源买了。别以为拿一台机器就能把实验做好。买买买。我自己就软磨硬泡让老板先后买了8个1080Ti级别的GPU和相关计算资源。

hyz 发表于 2019-9-12 23:05
Hi, thanks for your share, could you please list math courses from priority orders?

If you are a first-year graduate student with interest in ML, no matter what major you are studying

I would highly recommend:
Real analysis + stochastic processes + mathematical statistics + Probability theory (LLW/ CLT) + Probability theory (Martingale/ Brownian motion) + Functional Analysis.
With respect to algebra, only numerical linear algebra + convex optimization are good enough. No need to take advanced courses. And you can then find a problem and read the related documents.

But if you are almost to graduate, probably avoid any measure related courses, which will be too time-consuming for you with marginal improvement.

Just take the non-measure courses: data mining/ML/Bayesian data analysis/ Monte Carlo, etc.


补充内容 (2021-2-8 03:39):
Dynamical systems and functional inequalities are also very important for studies in SGD.

greyman 发表于 2019-9-13 00:06
If you are a first-year graduate student with interest in ML, no matter what major you are studyin ...

Thanks for your sharing, and what programming skills you would recommend? Algorithms course

Programming is not very hard in ML research. Python should be the most important, then shell scripts in Linux + regular expression skills + a little OS knowledge (thread process etc)