Recurrent Neural Networks with Long Short-Term Memory (LSTM) make use of gating mechanisms to mitigate exploding and vanishing gradients when learning long-term dependencies. For this reason, LSTMs and other gated RNNs are widely adopted, being the standard de facto for many sequence modeling tasks. Although the memory cell inside the LSTM contains essential information, it is not allowed to influence the gating mechanism directly. In this work, we improve the gate potential by including information coming from the internal cell state. The proposed modification, named Working Memory Connection, consists in adding a learnable nonlinear projection of the cell content into the network gates. This modification can fit into the classical LSTM gates without any assumption on the underlying task, being particularly effective when dealing with longer sequences. Previous research effort in this direction, which goes back to the early 2000s, could not bring a consistent improvement over vanilla LSTM. As part of this paper, we identify a key issue tied to previous connections that heavily limits their effectiveness, hence preventing a successful integration of the knowledge coming from the internal cell state. We show through extensive experimental evaluation that Working Memory Connections constantly improve the performance of LSTMs on a variety of tasks. Numerical results suggest that the cell state contains useful information that is worth including in the gate structure.

Working Memory Connections for LSTM / Landi, Federico; Baraldi, Lorenzo; Cornia, Marcella; Cucchiara, Rita. - In: NEURAL NETWORKS. - ISSN 0893-6080. - 144:(2021), pp. 334-341. [10.1016/j.neunet.2021.08.030]

Working Memory Connections for LSTM

Landi, Federico;Baraldi, Lorenzo;Cornia, Marcella;Cucchiara, Rita
2021

Abstract

Recurrent Neural Networks with Long Short-Term Memory (LSTM) make use of gating mechanisms to mitigate exploding and vanishing gradients when learning long-term dependencies. For this reason, LSTMs and other gated RNNs are widely adopted, being the standard de facto for many sequence modeling tasks. Although the memory cell inside the LSTM contains essential information, it is not allowed to influence the gating mechanism directly. In this work, we improve the gate potential by including information coming from the internal cell state. The proposed modification, named Working Memory Connection, consists in adding a learnable nonlinear projection of the cell content into the network gates. This modification can fit into the classical LSTM gates without any assumption on the underlying task, being particularly effective when dealing with longer sequences. Previous research effort in this direction, which goes back to the early 2000s, could not bring a consistent improvement over vanilla LSTM. As part of this paper, we identify a key issue tied to previous connections that heavily limits their effectiveness, hence preventing a successful integration of the knowledge coming from the internal cell state. We show through extensive experimental evaluation that Working Memory Connections constantly improve the performance of LSTMs on a variety of tasks. Numerical results suggest that the cell state contains useful information that is worth including in the gate structure.
2021
144
334
341
Working Memory Connections for LSTM / Landi, Federico; Baraldi, Lorenzo; Cornia, Marcella; Cucchiara, Rita. - In: NEURAL NETWORKS. - ISSN 0893-6080. - 144:(2021), pp. 334-341. [10.1016/j.neunet.2021.08.030]
Landi, Federico; Baraldi, Lorenzo; Cornia, Marcella; Cucchiara, Rita
File in questo prodotto:
File Dimensione Formato  
2021_NN_LSTM.pdf

Open Access dal 05/09/2023

Tipologia: AAM - Versione dell'autore revisionata e accettata per la pubblicazione
Dimensione 433.86 kB
Formato Adobe PDF
433.86 kB Adobe PDF Visualizza/Apri
1-s2.0-S0893608021003439-main (1).pdf

Accesso riservato

Tipologia: VOR - Versione pubblicata dall'editore
Dimensione 1.01 MB
Formato Adobe PDF
1.01 MB Adobe PDF   Visualizza/Apri   Richiedi una copia
Pubblicazioni consigliate

Licenza Creative Commons
I metadati presenti in IRIS UNIMORE sono rilasciati con licenza Creative Commons CC0 1.0 Universal, mentre i file delle pubblicazioni sono rilasciati con licenza Attribuzione 4.0 Internazionale (CC BY 4.0), salvo diversa indicazione.
In caso di violazione di copyright, contattare Supporto Iris

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11380/1252276
Citazioni
  • ???jsp.display-item.citation.pmc??? 12
  • Scopus 214
  • ???jsp.display-item.citation.isi??? 160
social impact