1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22 """
23 **Crino: a neural-network library based on Theano**
24
25 Crino is an open-source `Python <http://www.python.org/>`_ library aimed at
26 building and training artificial neural-networks. It has been developed on top
27 of `Theano <http://deeplearning.net/software/theano/>`_, by researchers from the
28 `LITIS <http://www.litislab.eu>`_ laboratory. It helps scientists and/or
29 programmers to design neural-network architectures adapted to their needs, using
30 a modular framework inspired by Torch. Our library also provides vanilla
31 implementations, and learning algorithms, for these architectures :
32
33 - auto-encoders (AE)
34 - multi-layer perceptrons (MLP)
35 - deep neural networks (DNN)
36 - input-output deep architectures (IODA)
37
38 IODA is an extension of DNN architectures, which is useful in cases where both
39 input and output spaces are high-dimensional, and where there are strong
40 interdependences between output labels. The input and output layers of a IODA
41 are initialized with an unsupervised pre-training step, based on the stacked
42 auto-encoder strategy, commonly used in DNN training algorithms. Then, the
43 backpropagation algorithm performs the final supervised learning step.
44
45 Crino and IODA are research topics of the `Deep in Normandy
46 <http://deep.normastic.fr/>`_ research program, which is a `NVIDIA GPU Research
47 Center <https://developer.nvidia.com/academia/centers/normastic>`_.
48
49 :see: `GitHub repository <https://github.com/jlerouge/crino>`_,
50 `Project homepage <http://julien.lerouge.me/crino/>`_
51 """
52
53
54 __version__ = "0.2.0"
55 import crino.criterion
56 import crino.module
57 import crino.network
58