Salvato in:
Dettagli Bibliografici
Autori principali: Sanchis-Agudo, Marcial, Wang, Yuning, Arnau, Roger, Guastoni, Luca, Lim, Jasmin, Duraisamy, Karthik, Vinuesa, Ricardo
Natura: Preprint
Pubblicazione: 2023
Soggetti:
Accesso online:https://arxiv.org/abs/2308.12874
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866910984306688000
author Sanchis-Agudo, Marcial
Wang, Yuning
Arnau, Roger
Guastoni, Luca
Lim, Jasmin
Duraisamy, Karthik
Vinuesa, Ricardo
author_facet Sanchis-Agudo, Marcial
Wang, Yuning
Arnau, Roger
Guastoni, Luca
Lim, Jasmin
Duraisamy, Karthik
Vinuesa, Ricardo
contents To improve the robustness of transformer neural networks used for temporal-dynamics prediction of chaotic systems, we propose a novel attention mechanism called easy attention which we demonstrate in time-series reconstruction and prediction. While the standard self attention only makes use of the inner product of queries and keys, it is demonstrated that the keys, queries and softmax are not necessary for obtaining the attention score required to capture long-term dependencies in temporal sequences. Through the singular-value decomposition (SVD) on the softmax attention score, we further observe that self attention compresses the contributions from both queries and keys in the space spanned by the attention score. Therefore, our proposed easy-attention method directly treats the attention scores as learnable parameters. This approach produces excellent results when reconstructing and predicting the temporal dynamics of chaotic systems exhibiting more robustness and less complexity than self attention or the widely-used long short-term memory (LSTM) network. We show the improved performance of the easy-attention method in the Lorenz system, a turbulence shear flow and a model of a nuclear reactor.
format Preprint
id arxiv_https___arxiv_org_abs_2308_12874
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Easy attention: A simple attention mechanism for temporal predictions with transformers
Sanchis-Agudo, Marcial
Wang, Yuning
Arnau, Roger
Guastoni, Luca
Lim, Jasmin
Duraisamy, Karthik
Vinuesa, Ricardo
Machine Learning
To improve the robustness of transformer neural networks used for temporal-dynamics prediction of chaotic systems, we propose a novel attention mechanism called easy attention which we demonstrate in time-series reconstruction and prediction. While the standard self attention only makes use of the inner product of queries and keys, it is demonstrated that the keys, queries and softmax are not necessary for obtaining the attention score required to capture long-term dependencies in temporal sequences. Through the singular-value decomposition (SVD) on the softmax attention score, we further observe that self attention compresses the contributions from both queries and keys in the space spanned by the attention score. Therefore, our proposed easy-attention method directly treats the attention scores as learnable parameters. This approach produces excellent results when reconstructing and predicting the temporal dynamics of chaotic systems exhibiting more robustness and less complexity than self attention or the widely-used long short-term memory (LSTM) network. We show the improved performance of the easy-attention method in the Lorenz system, a turbulence shear flow and a model of a nuclear reactor.
title Easy attention: A simple attention mechanism for temporal predictions with transformers
topic Machine Learning
url https://arxiv.org/abs/2308.12874