The present work takes into account the compactness and efficiency of Recurrent Neural Networks (RNNs) for solving Natural Language Processing (NLP) tasks. RNNs are a class of Artificial Neural Networks (ANNs). Compared to Feed-forward Neural Networks (FNNs), RNN architecture is cyclic, i.e. the connection between nodes form cycles. This subtle difference has actually a huge impact on solving sequence-based problems, e.g. NLP tasks.In particular, the first advantage of RNNs regards their ability to modellong-range time dependencies, which is a very desirable property for natural languagedata, where word’s meaning is highly dependent on its context. The second advantage of RNNs is that are flexible and accept as input many different datatypes and representation. This is again the case of natural language data, whichcan come in different sizes, e.g. words with different lengths, and types, e.g. sequences or trees.
Design compact and efficient recurrent neural networks for natural language processing tasks
Genchi, Walter
2019/2020
Abstract
The present work takes into account the compactness and efficiency of Recurrent Neural Networks (RNNs) for solving Natural Language Processing (NLP) tasks. RNNs are a class of Artificial Neural Networks (ANNs). Compared to Feed-forward Neural Networks (FNNs), RNN architecture is cyclic, i.e. the connection between nodes form cycles. This subtle difference has actually a huge impact on solving sequence-based problems, e.g. NLP tasks.In particular, the first advantage of RNNs regards their ability to modellong-range time dependencies, which is a very desirable property for natural languagedata, where word’s meaning is highly dependent on its context. The second advantage of RNNs is that are flexible and accept as input many different datatypes and representation. This is again the case of natural language data, whichcan come in different sizes, e.g. words with different lengths, and types, e.g. sequences or trees.File | Dimensione | Formato | |
---|---|---|---|
tesi_Genchi.pdf
accesso aperto
Dimensione
2.22 MB
Formato
Adobe PDF
|
2.22 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/28372