KeyMemoryRNN: A Flexible Prediction Framework for Spatiotemporal Prediction Networks

Most previous recurrent neural networks for spatiotemporal prediction have difficulty in learning the long-term spatiotemporal correlations and capturing skip-frame correlations. The reason is that the recurrent neural networks update the memory states only using information from the previous time s...

Descripción completa

Guardado en:

Detalles Bibliográficos
Autores principales:	Shengchun Wang, Xiang Lin, Huijie Zhu
Formato:	article
Lenguaje:	EN
Publicado:	IEEE 2021
Materias:	Recurrent neural networks spatiotemporal prediction prediction network Electrical engineering. Electronics. Nuclear engineering TK1-9971
Acceso en línea:	https://doaj.org/article/2b32dae151a54f3baa0971aa0135b6a9
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

id	oai:doaj.org-article:2b32dae151a54f3baa0971aa0135b6a9
record_format	dspace
spelling	oai:doaj.org-article:2b32dae151a54f3baa0971aa0135b6a92021-11-18T00:11:22ZKeyMemoryRNN: A Flexible Prediction Framework for Spatiotemporal Prediction Networks2169-353610.1109/ACCESS.2021.3114215https://doaj.org/article/2b32dae151a54f3baa0971aa0135b6a92021-01-01T00:00:00Zhttps://ieeexplore.ieee.org/document/9541390/https://doaj.org/toc/2169-3536Most previous recurrent neural networks for spatiotemporal prediction have difficulty in learning the long-term spatiotemporal correlations and capturing skip-frame correlations. The reason is that the recurrent neural networks update the memory states only using information from the previous time step node and the networks tend to suffer from gradient propagation difficulties. We propose a new framework, KeyMemoryRNN, which has two contributions. The first is that we propose the KeyTranslate Module to extract the most effective historical memory state named keyword state, and we propose the KeyMemory-LSTM which uses the keyword state to update the hidden state to capture the skip-frame correlation. In particular, KeyMemoryLSTM has two training stages. In the second stage, KeyMemoryLSTM adaptively skips the update of sometime step nodes to build a shorter memory information flow to alleviate the difficulty of gradient propagation to learn the long-term spatiotemporal correlations. The second is that both KeyTranslate Module and KeyMemoryLSTM are flexible additional modules, so we can apply them to most RNN-based prediction networks to build KeyMemoryRNN with different base network. The KeyMemoryRNN achieves the state-of-the-art on three spatiotemporal prediction tasks, and we provide ablation studies and memory analysis to verify the effectiveness of KeyMemoryRNN.Shengchun WangXiang LinHuijie ZhuIEEEarticleRecurrent neural networksspatiotemporal predictionprediction networkElectrical engineering. Electronics. Nuclear engineeringTK1-9971ENIEEE Access, Vol 9, Pp 147678-147691 (2021)
institution	DOAJ
collection	DOAJ
language	EN
topic	Recurrent neural networks spatiotemporal prediction prediction network Electrical engineering. Electronics. Nuclear engineering TK1-9971
spellingShingle	Recurrent neural networks spatiotemporal prediction prediction network Electrical engineering. Electronics. Nuclear engineering TK1-9971 Shengchun Wang Xiang Lin Huijie Zhu KeyMemoryRNN: A Flexible Prediction Framework for Spatiotemporal Prediction Networks
description	Most previous recurrent neural networks for spatiotemporal prediction have difficulty in learning the long-term spatiotemporal correlations and capturing skip-frame correlations. The reason is that the recurrent neural networks update the memory states only using information from the previous time step node and the networks tend to suffer from gradient propagation difficulties. We propose a new framework, KeyMemoryRNN, which has two contributions. The first is that we propose the KeyTranslate Module to extract the most effective historical memory state named keyword state, and we propose the KeyMemory-LSTM which uses the keyword state to update the hidden state to capture the skip-frame correlation. In particular, KeyMemoryLSTM has two training stages. In the second stage, KeyMemoryLSTM adaptively skips the update of sometime step nodes to build a shorter memory information flow to alleviate the difficulty of gradient propagation to learn the long-term spatiotemporal correlations. The second is that both KeyTranslate Module and KeyMemoryLSTM are flexible additional modules, so we can apply them to most RNN-based prediction networks to build KeyMemoryRNN with different base network. The KeyMemoryRNN achieves the state-of-the-art on three spatiotemporal prediction tasks, and we provide ablation studies and memory analysis to verify the effectiveness of KeyMemoryRNN.
format	article
author	Shengchun Wang Xiang Lin Huijie Zhu
author_facet	Shengchun Wang Xiang Lin Huijie Zhu
author_sort	Shengchun Wang
title	KeyMemoryRNN: A Flexible Prediction Framework for Spatiotemporal Prediction Networks
title_short	KeyMemoryRNN: A Flexible Prediction Framework for Spatiotemporal Prediction Networks
title_full	KeyMemoryRNN: A Flexible Prediction Framework for Spatiotemporal Prediction Networks
title_fullStr	KeyMemoryRNN: A Flexible Prediction Framework for Spatiotemporal Prediction Networks
title_full_unstemmed	KeyMemoryRNN: A Flexible Prediction Framework for Spatiotemporal Prediction Networks
title_sort	keymemoryrnn: a flexible prediction framework for spatiotemporal prediction networks
publisher	IEEE
publishDate	2021
url	https://doaj.org/article/2b32dae151a54f3baa0971aa0135b6a9
work_keys_str_mv	AT shengchunwang keymemoryrnnaflexiblepredictionframeworkforspatiotemporalpredictionnetworks AT xianglin keymemoryrnnaflexiblepredictionframeworkforspatiotemporalpredictionnetworks AT huijiezhu keymemoryrnnaflexiblepredictionframeworkforspatiotemporalpredictionnetworks
_version_	1718425151728517120

KeyMemoryRNN: A Flexible Prediction Framework for Spatiotemporal Prediction Networks

Ejemplares similares