2017
DOI: 10.48550/arxiv.1701.05923
|View full text |Cite
Preprint
|
Sign up to set email alerts
|

Gate-Variants of Gated Recurrent Unit (GRU) Neural Networks

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1
1

Citation Types

0
5
0
1

Year Published

2018
2018
2024
2024

Publication Types

Select...
4
1

Relationship

1
4

Authors

Journals

citations
Cited by 5 publications
(6 citation statements)
references
References 0 publications
0
5
0
1
Order By: Relevance
“…Such aggregations require manual feature engineering for each specific rule, with an implementation effort comparable to direct implementation of rule algorithms by policy experts. The deep neural network, which we implemented as a recurrent neural network [44][45][46] to capture the temporal nature of historical claims data, also has very poor performance. Besides the well-known problems of this type of network (vanishing or exploding gradient 47 ), we observe that rules defined in policy documents usually have different temporal aggregations (ranging from months to years).…”
Section: Discussionmentioning
confidence: 99%
See 1 more Smart Citation
“…Such aggregations require manual feature engineering for each specific rule, with an implementation effort comparable to direct implementation of rule algorithms by policy experts. The deep neural network, which we implemented as a recurrent neural network [44][45][46] to capture the temporal nature of historical claims data, also has very poor performance. Besides the well-known problems of this type of network (vanishing or exploding gradient 47 ), we observe that rules defined in policy documents usually have different temporal aggregations (ranging from months to years).…”
Section: Discussionmentioning
confidence: 99%
“…The deep neural network baseline uses a Recurrent Neural Network (RNN) architecture [44][45][46] to learn dependencies between elements in a sequence. The sequence used to classify a claim consists of all the preceding claims and related features in chronological order, in addition to the claim being classified.…”
Section: Baseline Modelsmentioning
confidence: 99%
“…Many of these variants have already been validated to produce comparable performance to the standard LSTM RNN in recent publications [9][10][11][12][13][14][15]. The remaining ones are currently being investigated in case studies.…”
Section: Discussionmentioning
confidence: 99%
“…We will describe a diverse set of variant networks with the intended goal of providing a host of choices, balancing parameter-reduction and quantitative performance in (validation-testing) accuracy. We have already demonstrated the quantitative performances of these new network variants in recent publications ( [9][10][11][12][13][14])-albeit for initial datasets. Here, we describe the insight and reasoning into the reduced networks' developments in a comprehesive way [15].…”
Section: The Rationale In Developing the Slim Lstmsmentioning
confidence: 94%
“…Перевагами нейронних мереж є [9][10][11]:  можливість їх навчання та адаптації;  можливість виявлення закономірностей у даних, їх узагальнення, тобто отримання знань з даних, тому не потрібні знання про об'єкт (наприклад, його математична модель);…”
Section: метод динамічного управління буфером запасів на основі м'яки...unclassified