@Techreport{C12tech,
  AUTHOR =   {Iveta Mrázová and Marek Kukačka},
  TITLE =    {Can Deep Convolutional Neural Networks Discover Meaningful Pattern Features?},
  INSTITUTION =  {Charles University, Faculty of Mathematics and Physics, KTIML},
  YEAR =     {2012},
  NUMBER =   {2012/1/KTIML},
  ADDRESS =  {Prague},
  ABSTRACT = {Recent advances in the area of deep neural networks brought a lot of attention
              to some of the key issues important for their design. In particular for 2D-shapes,
              their accuracy has been shown to outperform all other classifiers - e.g., in the German
              Traffic Sign competition run by IJCNN 2011. While the majority of classical image
              processing techniques is based on carefully pre-selected image features, convolutional
              neural networks are, namely, designed to learn local features autonomously.
              On the other hand, the entire training process may be quite cumbersome and the
              structure of the network has to be chosen beforehand. Our paper introduces a new
              sensitivity-based approach capable of picking the right image features from a pretrained
              SOM-like feature detector. Experimental results obtained so far for two case
              studies on face and hand-written digit recognition show that pruned network architectures
              impact a transparent representation of the features actually present in the data
              while improving network robustness.},
}