I'm currently working on deep learning for my research and will be releasing a parallelized, educational library for this in the near future--this library is written in Java and contains significant documentation from my work on teaching myself the mechanics/mathematical workings of these deep learning algorithms and how to practically overcome some issues faced when using a language like Java as opposed to more numerical computation languages like MatLab or Python (which has been traditionally/popularly used by Hinton, Bengio, and others, as can be seen on deeplearning.net).

If your research study/project is more long-term, maybe this can help you. My hope is my work will also help students wanting to use deep learning methods easily (like most state-of-the-art machine learning algorithms).

I would propose to use Java for just about anything. Unless you plan to work close with the hardware, then C would be customary. I've never heard of a Deep Belief Networks, care to share?

Thx. But from almost everything I read people tend to use Python, Matlab or R. Still not sure about it and I will check out Theano, a Python framework. I read about it before and it was also recommended in the other question about the project:

http://deeplearning.net/software/theano/introduction.html#introduction

About Deep Belief Networks: It's basically a two step learning method. In the first step you use unsupervised learning algorithms for feature detection and in the second step you use supervised algorithms on the results for classification. The Deep just hints that you have more than one layer in your neural network. Well, but that's just a very short explanation. Anybody correct me, if I am wrong please.

Here is the answer from the inventor:

http://www.scholarpedia.org/article/Deep_belief_networks

The paper that gave me the idea is this: https://www.researchgate.net/publication/221345149_Multimodal_Deep_Learning?ev=srch_pub

If you are interested also check out the other question linked in the top post, it has some very good answers.

Conference Paper Multimodal Deep Learning

I'm currently working on deep learning for my research and will be releasing a parallelized, educational library for this in the near future--this library is written in Java and contains significant documentation from my work on teaching myself the mechanics/mathematical workings of these deep learning algorithms and how to practically overcome some issues faced when using a language like Java as opposed to more numerical computation languages like MatLab or Python (which has been traditionally/popularly used by Hinton, Bengio, and others, as can be seen on deeplearning.net).

If your research study/project is more long-term, maybe this can help you. My hope is my work will also help students wanting to use deep learning methods easily (like most state-of-the-art machine learning algorithms).

Hi Alexander,

that sounds great!

Many are using Java to do this so I would say go for it. Here are some code examples:

http://research.dwi.ufl.edu/ufdw/j4k/examples.php

Thx for that late answer, Bruce. I actually refrained from using Java at all because instead of capturing the data myself I relied on a provided dataset from the Chalearn Contest 2014. So I just had to work with the data and after some playing around I found that much easier using Python with Numpy, Theano and OpenCV.

Dear Glatt, 

You can try deeplearning4J or H2O with GPU support.

http://deeplearning4j.org/

http://0xdata.com/

Hope this would help.

Thank you Mohamad! Especially the second link seems to be very interesting.  They have support for many programming languages and provide a tutorial-like documentation. I will definitely try it on a small dataset in R or Python.

Hi Rubin

I have the same task can you help me 

What is the best library for applying DBN with java

Thanks

HI,

I have the task about applying DBN to predict personality.

Please kindly recommend some library with Java in applying DBN!!

you may have a look at deeplearning4j

https://deeplearning4j.org/deepbeliefnetwork.html

https://deeplearning4j.org/deepbeliefnetwork.html