VORTRAG Dipl.-Phys. Welf Wustlich:

"State of the art visual object detection and text recognition with deep multidimensional recurrent neural networks"



By nature Recurrent Neural Networks are the deepest and most complex artificial biologically inspired systems recently used for machine learning. The system we will introduce in the announced talk combines several modern aspects like convolutional architecture, recurrent neural network, deep learning, multi-dimensional context sensitivity within a single compact architecture. Also we will demonstrate the outstanding performance in several real world scenarios at different tasks, like visual object detection and handwriting recognition.

The demonstrated system won the last international handwriting recognition contest, see: www.citlab.uni-rostock.de

Keywords: Recurrent Neural Networks, Multi Dimensional RNNs, MDLSTM, handwriting recognition, visual object detection

Beginn: 12:15 Uhr - INB Seminarraum

siehe auch: Lunch-Seminar