Deep Gate Recurrent Neural Network

Yuan Gao, Dorota Glowacka
; Proceedings of The 8th Asian Conference on Machine Learning, PMLR 63:350-365, 2016.

Abstract

This paper explores the possibility of using multiplicative gates to build two recurrent neural network structures. These two structures are called Deep Simple Gated Unit (DSGU) and Simple Gated Unit (SGU), which are structures for learning long-term dependencies. Compared to traditional Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), both structures require fewer parameters and less computation time in sequence classification tasks. Unlike GRU and LSTM, which require more than one gate to control information flow in the network, SGU and DSGU only use one multiplicative gate to control the flow of information. We show that this difference can accelerate the learning speed in tasks that require long dependency information. We also show that DSGU is more numerically stable than SGU. In addition, we also propose a standard way of representing the inner structure of RNN called RNN Conventional Graph (RCG), which helps to analyze the relationship between input units and hidden units of RNN.

Cite this Paper


BibTeX
@InProceedings{pmlr-v63-gao30, title = {Deep Gate Recurrent Neural Network}, author = {Yuan Gao and Dorota Glowacka}, pages = {350--365}, year = {2016}, editor = {Robert J. Durrant and Kee-Eung Kim}, volume = {63}, series = {Proceedings of Machine Learning Research}, address = {The University of Waikato, Hamilton, New Zealand}, month = {16--18 Nov}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v63/gao30.pdf}, url = {http://proceedings.mlr.press/v63/gao30.html}, abstract = {This paper explores the possibility of using multiplicative gates to build two recurrent neural network structures. These two structures are called Deep Simple Gated Unit (DSGU) and Simple Gated Unit (SGU), which are structures for learning long-term dependencies. Compared to traditional Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), both structures require fewer parameters and less computation time in sequence classification tasks. Unlike GRU and LSTM, which require more than one gate to control information flow in the network, SGU and DSGU only use one multiplicative gate to control the flow of information. We show that this difference can accelerate the learning speed in tasks that require long dependency information. We also show that DSGU is more numerically stable than SGU. In addition, we also propose a standard way of representing the inner structure of RNN called RNN Conventional Graph (RCG), which helps to analyze the relationship between input units and hidden units of RNN.} }
Endnote
%0 Conference Paper %T Deep Gate Recurrent Neural Network %A Yuan Gao %A Dorota Glowacka %B Proceedings of The 8th Asian Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2016 %E Robert J. Durrant %E Kee-Eung Kim %F pmlr-v63-gao30 %I PMLR %J Proceedings of Machine Learning Research %P 350--365 %U http://proceedings.mlr.press %V 63 %W PMLR %X This paper explores the possibility of using multiplicative gates to build two recurrent neural network structures. These two structures are called Deep Simple Gated Unit (DSGU) and Simple Gated Unit (SGU), which are structures for learning long-term dependencies. Compared to traditional Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), both structures require fewer parameters and less computation time in sequence classification tasks. Unlike GRU and LSTM, which require more than one gate to control information flow in the network, SGU and DSGU only use one multiplicative gate to control the flow of information. We show that this difference can accelerate the learning speed in tasks that require long dependency information. We also show that DSGU is more numerically stable than SGU. In addition, we also propose a standard way of representing the inner structure of RNN called RNN Conventional Graph (RCG), which helps to analyze the relationship between input units and hidden units of RNN.
RIS
TY - CPAPER TI - Deep Gate Recurrent Neural Network AU - Yuan Gao AU - Dorota Glowacka BT - Proceedings of The 8th Asian Conference on Machine Learning PY - 2016/11/20 DA - 2016/11/20 ED - Robert J. Durrant ED - Kee-Eung Kim ID - pmlr-v63-gao30 PB - PMLR SP - 350 DP - PMLR EP - 365 L1 - http://proceedings.mlr.press/v63/gao30.pdf UR - http://proceedings.mlr.press/v63/gao30.html AB - This paper explores the possibility of using multiplicative gates to build two recurrent neural network structures. These two structures are called Deep Simple Gated Unit (DSGU) and Simple Gated Unit (SGU), which are structures for learning long-term dependencies. Compared to traditional Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), both structures require fewer parameters and less computation time in sequence classification tasks. Unlike GRU and LSTM, which require more than one gate to control information flow in the network, SGU and DSGU only use one multiplicative gate to control the flow of information. We show that this difference can accelerate the learning speed in tasks that require long dependency information. We also show that DSGU is more numerically stable than SGU. In addition, we also propose a standard way of representing the inner structure of RNN called RNN Conventional Graph (RCG), which helps to analyze the relationship between input units and hidden units of RNN. ER -
APA
Gao, Y. & Glowacka, D.. (2016). Deep Gate Recurrent Neural Network. Proceedings of The 8th Asian Conference on Machine Learning, in PMLR 63:350-365

Related Material