“…Deep networks can be constructed by stacking convolutions, pooling operators and point-wise nonlinearities such as rectified linear activation functions. CNNs have been intensively applied to vision and image recognition problems ( Farabet et al, 2010 , Matusugu et al, 2013 , Ciresan et al, 2013 , Behnke, 2003 , Yaniv et al, 2015 , Masci et al, 2013 ), document recognition ( LeCun et al, 1998 ) and medical signal processing ( Graupe et al, 1988 , Graupe et al, 1989 ), among numerous other applications. However, the learned CNN transformations have mostly led to very good classification results in cases where a lot of data is available.…”