Lucas Mauch

Deep Learning / Pattern Recognition

Deep Learning is a new area in machine learning and deals with the structural design and training of deep architectures. One question, for example, is how deep architectures can be used to learn abstract representations of data (representation learning) or to identify patterns that are in demand (pattern recognition).

Although the idea of ​​deep architectures is old, for a long time they have not been applicable due to the lack of suitable training algorithms. It was only in 2006 that the breakthrough came with the introduction of greedy layer-wise training. Since then, deep architects have been successfully used in speech recognition, music transcription, and OCR (Optical Character Recognition) applications, setting new standards in recognition accuracy due to their ability to abstract data.

Emphases are:

•     Theoretical description and analysis of deep architectures
•     Development and analysis of training algorithms
•     Efficient implementation
•     Application of the algorithms for the solution of pattern recognition problems

L. Mauch , G. Steidle, J. Machann, B. Yang, F. Schick A fully automatic reference deconvolution strategy to increase the accuracy of in vivo lipid Magnetic Resonance in Medicine, 2015 (ahead of print)