如何让机器学习的初学者,已经学习了机器学习和深度学习的慕课,达到一个新的高度,可以阅读论文,可以在行业中高效的产出?

tinyfool 发布于 2017年09月19日 | 更新于 2017年09月29日
Chuns 等1人欣赏。

作者:吴恩达

译自:Quora

网络课程是非常非常高效的学习方式,所以从此开始非常合理。完成机器学习慕课(ml-class.org) 和深度学习专业(http://deeplearning.ai)以后,你还可以做的有:

Courses are a very very efficient way to learn, so starting there definitely makes sense. After finishing the ML MOOC (ml-class.org) and Deep learning specialization (http://deeplearning.ai), here’re some additional steps you can take:

  1. 在Twitter关注机器学习领袖,去阅读他们感兴趣的论文,blog等等。
    Follow leaders in ML on twitter to see what research papers/blog posts/etc. they’re excited about, and go read them too.
  2. 复现其他人发表的结果。这是精通机器学习,非常高效但是严重被低估的方法。通过长期观察大量斯坦福博士新生变成厉害的研究者,我敢说,复现别人的结果(而不仅仅是读论文),是帮助你理解最新算法的细节最有效的方法。很多人过早的进入了发明新东西的阶段,当然也值得做,但是实际上这是学习和构建你的基础知识比较低效的方法。
    Replicate others’ published results. This is a very effective but highly under-rated way to get good at ML. Having seen a lot of new Stanford PhD students grow to become great researchers, I can say confidently that replicating others’ results (not just reading the papers) is one of the most effective ways to see and make sure you understand the details of the latest algorithms. Many people jump too quickly into trying to invent something new, which is also worth doing, but is actually a slower way to learn and build up your foundation of knowledge.
  3. 当你阅读了足够的论文,blog等等。也复现了足够多的结果,几乎是魔法一样你就会开始有自己的观点和自己的主意。当你构建新东西的时候,把它写成论文或者blog,并考虑开源你的代码,把它分享回社区!希望这可以帮你获得更多来自社区的反馈,进一步加速你的学习。
    When you’ve read enough papers/blogs/etc. and replicated enough results, almost magically you’ll start to have your own opinions and your own ideas. When you do build something new, publish it in a paper or blog post and consider open-sourcing your code, and share it back out with the community! Hopefully this will help you get more feedback from the community, and further accelerate your learning.
  4. 参加任何可以帮助你学习的活动,在线竞赛,线下聚会,参加(或者在线看)好的人工智能/机器学习/视觉/自然语言处理/语音,等等会议,例如ICML,NIPS,ICLR等等。
    Participate in any other enrichment activities that help you learn, such online competitions, going to meetups, attending (or watching online videos of) good AI/ML/vision/NLP/speech/etc. conferences like ICML, NIPS and ICLR.
  5. 找到一些同行做朋友。你自己就可以取得很多进展,但是有朋友可以一起讨论想法,会帮助你学习,而且有很多乐趣。如果你可以接触到人工智能专家,如教授,PhD学生,或者好的研究者,跟他们聊聊。有时候,我跟Geoff Hinton,Yoshua Bengio, Yann LeCun这样的人只要聊5分钟就可以学习到很多;但是我从斯坦福的PhD学生,deeplearning.ai的团队成员,以及我拜访的不同公司的工程师那里也学到了很多。
    Find friends to do this with. You can make a lot of progress by yourself, but having friends to bounce ideas off will help your learning and also make it more fun. If you have access to AI experts like professors, PhD students, or good researchers, talk to them too. Sometimes I’ve learned a ton from a 5 minute conversation with people like Geoff Hinton, Yoshua Bengio, Yann LeCun; but also from my PhD students at Stanford, team members at deeplearning.ai, or engineers at the various companies I sometimes visit.
  6. 虽然有工作一起工作很重要,但是你的朋友不同意你的主意,有时候,你还是要去实现它,做出来给自己看。Geoff Hinton在我做的“深度学习英雄”的采访里也说过类似的话。
    Despite the importance of having friends to work with, if your friends disagree with your ideas, sometimes you should still implement it and try it out to see for yourself. Geoff Hinton also said something similar in his “Heroes of Deep Learning” interview.

我认识的每个世界级的机器学习研究者,都花了大量的独占时间去实现算法,调节超参数,阅读论文,找出对于他们自己什么有用,什么没用。我仍旧觉得这样的工作很有趣,希望你也这么看。

Every world class ML researcher I know has spent a lot of solitary hours implementing algorithms, tuning hyperparameters, reading papers, and figuring out for themselves what does and doesn’t work. I still find this type of work fun, and hope you will too.

共4条回复
saneryee 回复于 2017年09月20日

根据 “Heroes of Deep Learning” 里的 Hero 总结了一个 twitter list.有些没找到欢迎补充。

https://twitter.com/tsaneryeer/lists/ai-ml-leaders

tinyfool 回复于 2017年09月20日

1楼 @saneryee

uuspider 回复于 2017年09月29日

现在发现搞机器学习,搞到新鲜的数据很重要,数据既是机器学习的原材料,也是动力燃料,简直就是机器学习世界里的矿产资源。有了数据就会有各种想法。

tinyfool 回复于 2017年09月29日

3楼 @uuspider 有想法的话,其实数据好弄

登录 或者 注册
相关帖子