I am quite proficient with Machine Learning libraries and now want to get into Deep Learning. I am even quite comfortable with neural networks as far as understanding back propagation algorithm is concerned.

Three languages I am most comfortable with are R, Java & Python.

To start with deep learning, I see that there are various libraries such as Caffe, Torch, DL4J and so on.

I want to start implementing deep learning as soon as possible with a library that has the least learning curve w.r.t the language, its interfaces and so on.

Which among the existing libraries is the easiest to start with?


1 Answer 1


If you are comfortable with python then I recommend using Theano


In Theano you can set the calculations to be ran on the GPU using a CUDA backend and on multiple cores. This is useful since you time so you do not have to code in CUDA.

There are also autoencoders and Boltzmann machines in the examples. Another advantage is that it is friendly working with scikit learn another python machine learning library and the code is pretty well documented.

R is ok but can be a bit slow, unless you send things to the GPU. I also know that Torch7 has one advantage is there is support for LeNet and whatever facebook has but I have never use those.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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