Crino is an open-source Python library aimed at building and training artificial neural-networks. It has been developed on top of Theano, by researchers from the LITIS laboratory. It helps scientists and/or programmers to design neural-network architectures adapted to their needs, using a modular framework inspired by Torch. Our library also provides vanilla implementations, and learning algorithms, for these architectures :
- auto-encoders (AE)
- multi-layer perceptrons (MLP)
- deep neural networks (DNN)
- input-output deep architectures (IODA)
IODA is a novel DNN architecture, which is useful in cases where both input and output spaces are high-dimensional, and where there are strong interdependences between output labels. The input and output layers of a IODA are initialized with an unsupervised pre-training step, based on the stacked auto-encoder strategy, commonly used in DNN training algorithms. Then, the backpropagation algorithm performs the final supervised learning step.
Crino and IODA are research topics of the Deep in Normandy research program of the NormaSTIC federation, which has been awarded as a NVIDIA GPU Research Center in 2015.
- Install Crino :
cd to/some/path
git clone https://github.com/jlerouge/crino.git
cd crino
sudo python setup.py install- Run the given example :
cd example
chmod +x example.py
./example.py- Adapt it to your needs! Crino is natively compatible with Matlab-like data or any format handled by SciPy/NumPy.
- Check the project documentation.
- What does "device gpu is not available" mean ? Your GPU card may not be compatible with CUDA technology (check http://www.geforce.com/hardware/technology/cuda/supported-gpus). If so, there is nothing to do. Otherwise, your theano installation may have a problem (see http://deeplearning.net/software/theano/install.html#using-the-gpu).
- Where does the name "Crino" come from ? We developed this library as an extension of Theano. In Greek mythology, Crino is the daughter of Theano.
If you use Crino and/or our IODA framework for academic research, you are highly encouraged (though not required) to cite the following paper:
- J. Lerouge, R. Herault, C. Chatelain, F. Jardin and R. Modzelewski, "IODA: an Input/Output Deep Architecture for image labeling", Pattern Recognition (2015), DOI: 10.1016/j.patcog.2015.03.017 [Epub ahead of print]
We would like to thank the authors of Theano :
- J. Bergstra, O. Breuleux, F. Bastien, P. Lamblin, R. Pascanu, G. Desjardins, J. Turian, D. Warde-Farley and Y. Bengio. “Theano: A CPU and GPU Math Expression Compiler”. Proceedings of the Python for Scientific Computing Conference (SciPy) 2010. June 30 - July 3, Austin, TX
IODA is partly based on the original work of B. Labbé et al. :
- B. Labbé, R. Hérault and C. Chatelain . “Learning Deep Neural Networks for High Dimensional Output Problems”. In IEEE International Conference on Machine Learning and Applications (ICMLA'09), December 2009.
You can contact us with the following e-mail address : [email protected]. Feel free to open a new issue in case you have found a bug in Crino.