Deep learning from the beginning

Deep learning has been fire for a long time, some people have been here for many years, and some people have just begun, such as myself.

How to get into this field quickly in a short period of time to master deep learning the latest technology is a question worth thinking about.

In the present situation, it is the best way to study this area through courses on the web and various tutorials and various papers.

After a period of groping, I thought it was the best concrete way to start learning about the four-bit Daniel in the deep learning field. These four Daniel are Andrew Ng, Geoffrey Hinton, Yann Lecun,yoshua Bengio.

Andrew Ng has a machine learning course on his Coursera and Stanford's UFLDL tutorial, Geoffrey Hinton has a Coursera Network for machine Learning course, Yoshua Bengio out the deep learning of the textbook and his lab built a python-based machine learning Library Theano and deep learning.net this site. And Yann LeCun has a lot of relevant papers, and he's very active on Facebook and Google +, often sharing some very useful information.

Personally, in the case of having sufficient language foundation and linear algebra and probability mathematical basis and MATLAB programming basis, My Learning route is:

1) Coursera on Andrew Ng's machine learning course

2) Stanford CS229 essentially (1) and (2) courses can be combined

3) Stanford UFLDL Tutorial in conjunction with related important papers

4) Coursera on Geoffrey Hinton Neural Network for machine learning

5) According to the specific direction of reading NIPS,ICML,ICLR,ICCV and other top conference papers, as well as various talk,mlss and so on, combined with the latest information to master the latest technical progress.

Accomplishing the above tasks can basically achieve the following goals:

1) Formula derivation to complete deep learning concrete algorithm

2) Ability to read the latest papers

3) can use Matlab to write deep learning algorithm and realize. Considering that the MATLAB code can be converted into C code, the implemented deep learning can be applied to the concrete practice.

After reaching this level, the next step is to break through innovation.

And of course there's another route. If you have a python base, it's tutorials to go Yoshua bengio.

So how long does it take to complete the previous learning task? How to achieve a better combination of theory and practice. This needs to be adjusted repeatedly in learning.

As the first article of Deep Learning series, this article is just about how to learn to say their own views, these articles will also be My Learning side summary of the product.