2
$\begingroup$

I am very interested in deep learning, and I was recommended to read The Elements of Statistical Learning, but it is very hard to get into. So my question is :

How important is learning statistics and reading books like elements of statistical learning, if my end goal is to learn deep learning?

I should have mentioned that I have already done Andrew Ng's Machine Learning course.

$\endgroup$
  • $\begingroup$ Sorry, but specifically asking for opinions is not a good question to the standards of this site. Consider searching and asking on other venues, such as internet forums about deep learning or online course sites. $\endgroup$ – E_net4 the ferris ball Aug 10 '17 at 14:13
  • $\begingroup$ Yes, if you want to be able to do more than implement other people's code and actually understand what you're doing. It's like asking Do I need to understand algorithms and data structures if I want to become a software developer? Not if you want to do basic Web development, but is the lowest level of understanding what you aspire to? Visit your local library to find a simpler book. If it's attached to a university, look at what book they have on the curriculum. $\endgroup$ – Emre Aug 10 '17 at 17:36
  • $\begingroup$ @Emre Okay, I get your point. But I find it overwhelming since I do not know how deep I have to go on a particular topic, say Statistics. What happens is I go on a DFS-type-Study-routine and exhaust myself. How deep should I pursue statistics in the context of ML? Also, what book would you recommend? $\endgroup$ – Gaurav Kumar Aug 10 '17 at 19:12
  • $\begingroup$ Use BFS instead, then; look something up when you get stuck. A good book for you might be All of Statistics by Larry Wasserman. How much you need to know depends on what your application is. For example, if you need to do any sort of experimentation you will want to know about experimental design and hypothesis testing. Then again, you might never have to run a real experiment. Going back to the computer science analogy, it's like asking Do I need to know the whole Algorithms book? or Will I ever encounter graph theory? It depends on you $\endgroup$ – Emre Aug 14 '17 at 16:05
5
$\begingroup$

I would not call it a pre-requisite, but I would say reading books the that is necessary for knowing it all very well.

I would recommend you to start with Andrew Ng's Machine Learning course on coursera.

Once you are familiar with that, move on to deep learning videos:

  1. Nando de Freitas

  2. Hugo Larochelle

While you are going through topics one after the other, go back to the books and read what they have to offer. Before you know, you would have done it all.

Good Luck!

$\endgroup$
  • $\begingroup$ @oW_: Thanks for the edit. I was not allowed to post 3 links as this was my first time answering on datascience.stackexchange and did not have enough points for that. $\endgroup$ – VijayR Aug 10 '17 at 21:06
4
$\begingroup$

If you don't like to start with statistics and bias variance trade off.I suggest you to go through

with Andrew Ng's class after that you can proceed to Neural Networks and deep learning book

It is a good starter for Deep Learning

$\endgroup$
2
$\begingroup$

It is not a pre-requisite, and you can learn it easily once you encounter something you are not familiar with. Statistics is pretty old and there are many learning resources on the Web, which you can get to whenever you hit the wall while learning about deep learning.

As to which field is pre-requisite, I think it is enough to first learn about Gaussian (normal) distribution, linear regression, and logistic regression. Then when you encounter something you don't understand, it is time to invest your time on statistics.

The more requisite fields I believe are calculus and linear algebra. If you haven't learned about them (such as partial derivative, matrix transpose, etc), it is very difficult to start to learn deep learning.

Also, the prior exposure to some of machine learning algorithms would make your learning faster. But I'm sure you already got it given that you finished Andrew Ng's course on Coursera. Andrew Ng starts deep learning course on Coursera from August 15th, so you can join it and get a grasp of what is required.

The book you linked sounds more like focused on machine learning than on statistics, BTW.

$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.