PhD defense of Sönke Michalik on automated, FPGA accelerated deep learning of object classes

Congratulations to Sönke Michalik, who has sucessfully defended his thesis entitled „ Automatisiertes Lernen von Neuronalen Netzen zur Objektklassifikation auf heterogenen Systemen“ (in German, Englisch translation: Automated Learning von Neural Networks for Object Classification on heterogeneous systems). The thesis proposes a full automated and hardware accelerated system for learning object classes based on image sets that are retrieved automatically from the internet. The architecture includes depth based image segmentation and execution of binary deep networks on the FPGA with framerates up to 100 images per second at very low power consumption. Applications in real-time mobile visual processing are targeted. The thesis was evaluated by the examination commitee including Dr.-Ing Ulf Kulau at the helm and reviewers and supervisors Prof. M. Berekovic and Prof. J. Steil.