Cells in Multidimensional Recurrent Neural Networks

Abstract

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

Returnn: The RWTH extensible training framework for universal recurrent neural networks

“…Additionally, we process multiple images and also the four directions of a multi-directional MDLSTM layer simultaneously by using batched cuBLAS operations and custom CUDA kernels. Optionally, the stable cell described in [31] can be used to improve convergence.…”

Viewing Channel as Sequence Rather Than Image: A 2-D Seq2Seq Approach for Efficient MIMO-OFDM CSI Feedback

“…Thus, we use the RNN with LSTM mechanism extended to the 2-D case named 2-D LSTM [28], [29] Similar with some machine translation tasks [30], [31], the compression and reconstruction of CSI is essentially a feature extraction and representation task from one 2-D sequence to one 2-D sequence. Therefore, we propose a 2-D Sequence to Sequence (Seq2Seq) neural network mainly composed of 2-D LSTMs for the CSI feedback task.…”

“…2-D RNN and 2-D LSTM Mechanism for MIMO-OFDM CSITo deal with sequence-dependent data, RNN shows superior advantage due to its recurrent calculation structure. For data with multi-dimensional (M-D) sequence-dependence, the authors in[28],[29] introduced M-D RNN and M-D LSTM as a generalization of standard RNN and standard LSTM, respectively. Through using a M-D recurrent calculation structure, M-D RNN can overcome the limitation of one-dimensional (1-D) RNN, which can only utilize 1-D sequence dependence.…”