[ml] Theano python library

[Philippe Hamel]

Hi machine learners,

My name is Philippe. I met some of you at last week's meeting (my first
Noise Bridge experience). I study machine learning applied to music at
Université de Montréal (UdM). I am in San francisco for a few months, and
I'll try to attend the ML meetings. I work a lot with neural networks, in
particular Deep Belief Networks ( some kind of deep network using an
unsupervised pretraining algorithm ). Some people asked me about the
libraries I use. Well, for python users, here is a library that might
interest some of you:

http://deeplearning.net/software/theano/

It is not a Machine Learning library, but rather a tool for symbolic
expression manipulation. It is developped at UdM by students that do
research in deep neural networks.

I find this library useful because :
- You can build neural networks in just a few simple lines
- It automatically calculates gradients (so no need to do those messy
backpropagation calculations)
- It optimizes the computation of your expressions and prevents numerical
errors
- It is compatible with numpy
- It magically generates c++ code and compiles it, for increased performance
- If you have a recent nvidia video card, you can use the parallelization
power of your GPU
- The mailing list is very active, and reported bugs are usually fixed
pretty quickly.

There is a nice tutorial on how to use Theano at :

http://deeplearning.net/tutorial/

I hope this will be usefull to some of you.

See you at the Noise Bridge,

Phil




2010/10/19 Mike Schachter <mike at mindmech.com>

> damnit! obviously the title was screwed up, the GLM
> talk is 10/27/2010 @ 7:30.
>
>
>
>
> On Tue, Oct 19, 2010 at 4:10 PM, Mike Schachter <mike at mindmech.com> wrote:
>
>> Hey everyone!
>>
>> Anything going on this week? I won't be around but
>> strongly suggest some decentralized get-together.
>>
>> Next Wednesday, 10/27, Tom will be giving a talk
>> on GLMs which you should surely attend.
>>
>>   mike
>>
>>
>
> _______________________________________________
> ml mailing list
> ml at lists.noisebridge.net
> https://www.noisebridge.net/mailman/listinfo/ml
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.noisebridge.net/pipermail/ml/attachments/20101019/80092f1f/attachment.html>