- Internal research project -

Up to recently humans were by far superior in recognizing and classifying objects. It seemed to be a hard problem to capture the variances and invariances humans can deal with easily in the real world. After many years of research this is slowly changing. Besides highly increased training data sets (big data), two new insights into the information processing principles of the brain now make a difference: i) sparse coding of information, with the "dual" concept of compressed sensing, and ii) deep convolutional networks, which use several layers of artificial simple and complex cells. We develop and apply sparse coding methods combined with principles of deep networks to pattern and object recognition tasks. 

For many years our approaches were "world champion" (without artificially augmented training data) on the MNIST data set of hand written digits.