Cells in Multidimensional Recurrent Neural Networks

• DOI: 10.48550/arxiv.1412.2620
• Publication Date: 2014-01-01
• Authors: Gundram Leifert; Tobias Strauß; Tobias Grüning; Welf Wustlich; Roger Labahn
• Keywords: Connectionism

The transcription of handwritten text on images is one task in machine learning and solution to solve it using multi-dimensional recurrent neural networks (MDRNN) with connectionist temporal classification (CTC). RNNs can contain special units, the long short-term memory (LSTM) cells. They are able learn term dependencies but they get unstable when dimension chosen greater than one. We defined some useful necessary properties for one-dimensional LSTM cell extend them case. Thereby we introduce several new cells better stability. present a method design theory linear shift invariant systems. compared IFN/ENIT Rimes database, where improve recognition rate cell. So each application MDRNNs used could be improved by substituting developed