This HTML5 document contains 26 embedded RDF statements represented using HTML+Microdata notation.

The embedded RDF content will be recognized by any processor of HTML5 Microdata.

Namespace Prefixes

Prefix	IRI
dcterms	http://purl.org/dc/terms/
n2	https://kar.kent.ac.uk/id/eprint/
n18	https://kar.kent.ac.uk/74240/
wdrs	http://www.w3.org/2007/05/powder-s#
dc	http://purl.org/dc/elements/1.1/
n6	http://purl.org/ontology/bibo/status/
rdfs	http://www.w3.org/2000/01/rdf-schema#
n15	https://kar.kent.ac.uk/id/subject/
n14	https://demo.openlinksw.com/about/id/entity/https/raw.githubusercontent.com/annajordanous/CO644Files/main/
n3	http://eprints.org/ontology/
n16	https://kar.kent.ac.uk/id/event/
bibo	http://purl.org/ontology/bibo/
rdf	http://www.w3.org/1999/02/22-rdf-syntax-ns#
n9	https://kar.kent.ac.uk/id/eprint/74240#
n4	https://kar.kent.ac.uk/id/document/
n8	https://kar.kent.ac.uk/id/
xsdh	http://www.w3.org/2001/XMLSchema#
n13	https://demo.openlinksw.com/about/id/entity/https/www.cs.kent.ac.uk/people/staff/akj22/materials/CO644/
n11	https://kar.kent.ac.uk/id/person/

Statements

Subject Item

n2:74240

rdf:type

n3:EPrint bibo:AcademicArticle n3:ConferenceItemEPrint bibo:Article

rdfs:seeAlso

n18:

n3:hasDocument

n4:3178592 n4:3178707 n4:3178593 n4:3178704 n4:3178705 n4:3178706

n3:hasSubmitted

n4:3178592

dc:hasVersion

n4:3178592

dcterms:title

Relating RNN layers with the spectral WFA ranks in sequence modelling

wdrs:describedby

n13:export_kar_RDFN3.n3 n14:export_kar_RDFN3.n3

dcterms:date

2019-05-30

dcterms:creator

n11:ext-m.grzes@kent.ac.uk n11:ext-fl207@kent.ac.uk

bibo:status

n6:peerReviewed n6:unpublished

bibo:abstract

We analyse Recurrent Neural Networks (RNNs) to understand the significance of multiple LSTM layers. We argue that the Weighted Finite-state Automata (WFA) trained using a spectral learning algorithm are helpful to analyse RNNs. Our results suggest that multiple LSTM layers in RNNs help learning distributed hidden states, but have a smaller impact on the ability to learn long-term dependencies. The analysis is based on the empirical results, however relevant theory (whenever possible) was discussed to justify and support our conclusions.

dcterms:isPartOf

n8:repository

dcterms:subject

n15:QA76.76

bibo:authorList

n9:authors

bibo:presentedAt

n16:ext-cef8bf803b56cf2195a7991783f3af43