pylearn2

While not open journals per se, many (new) papers will typically be available at academic homepages (preprints), via "All [X] versions"-links in a google scholar [0] search, or on arXiv [1], depending on the subfield. For instance, the machine learning group (Yoshua Bengio, Ian Goodfellow, et al.) at the University of Montreal (the people behind open source software like Theano [2], Pylearn 2 [3], etc.) regularly makes papers available on arXiv, and is currently working on a book about Deep Learning, drafts of which are freely available [4]. [0]: http://scholar.google.com [1]: http://arxiv.org [2]: https://github.com/Theano/Theano [3]: https://github.com/lisa-lab/pylearn2 [4: http://www.deeplearningbook.org

I know of 4 projects for deep learning based on Theano. Keras, Blocks and Lasagne all seem to share the same goal of being more libraries than framework. You can use only one part (e.g. a Layer implementation, training algo) without having to pull in everything : https://github.com/bartvm/blocks https://github.com/benanne/Lasagne Then there is pylearn2, which look more like a framework and seems to be a good candidate for becoming the GPU-accelerated scikit-learn : https://github.com/lisa-lab/pylearn2 I have started using blocks and did some tests with pylearn2. Anybody with more experience want to share the strength/weaknesses of each of those projects ?

Short answer: Python and R for academic. Java or C++ for production. Julia has some followers, LUA has a Deep Learning Implementation. Python+CUDA+C++ is very common in deep learning. --- Chiming in as a machine learning researcher. My experience is primarily focused on the 'deep learning' buzzword at the moment, so people still researching SVM/NN/clustering may have different experiences. Academic machine learning is starting to shift towards Python. Most (all?) of the main deep learning packages have either Python as the core (like PyLearn2 + Theano) or a strong Python component (like Caffe or DeepLearn). There is some Java presence (like DeepLearning4j) and some C++ mixed in for the very high performance code. Usually people will write the glue code in Python and then do the heavy matrix-intensive operations using PyCuda or cuBL

Hm. And here I was thinking I'd seen most of the accelerated deep learning implementations. Although this one seems a touch incomplete as far as features go. There are some other for those not aware: [1] [Python] pylearn2 by the Lisa Lab @ uMontreal. Probably one of the beefiest of the bunch. Has a pretty extensive feature set, but IMO isn't the simplest to use. [2] [Python] MORB by Sander Dielman. Just restricted boltzman machines, but very nice and intuitive to use. Also built on Theano. [3] [C++] CUV by the AIS lab at Bonn University in Germany. Not strictly deep learning, but it's a fast CUDA-backed library written in templated C++ with bindings to other languages, including python. It's been used to implement deep learning models. [4] [Python] DeepNet by Nitish Srivastava and co. at U of Toronto. I don't have as much experience with

Theano [1] is developed at the LISA lab led by Yoshua Bengio at l'Université de Montréal. Pylearn2 [2] is developed by the same group, and is built on top of Theano. [1] http://deeplearning.net/software/theano/ [2] http://github.com/lisa-lab/pylearn2

If you're interested in theano primarily because of deep learning, I highly recommend you check out pylearn2 (not too much documentation, but docs here: http://deeplearning.net/software/pylearn2/ and source on github here: github.com/lisa-lab/pylearn2 ). Pylearn2 is a set of deep learning algorithms implemented with theano. The LISA (deep learning) group at the University of Montreal (same group that created theano) maintains this library and puts a lot of the code they use for their papers in pylearn2. pylearn2 thus makes it quite easy to use a lot of state of the art algorithms, such as maxout.