Repositori institucional URV
| URV's Author/s: | Abdellatif Fatahallah Ibrahim Mahmoud, Hatem / Abdelnasser Mohamed Mahmoud, Mohamed / Puig Valls, Domènec Savi / Romaní Also, Santiago |
| Author, as appears in the article.: | Schwarz Schuler, Joao Paulo; Romani, Santiago; Abdel-Nasser, Mohamed; Rashwan, Hatem; Puig, Domenec |
| Author's mail: | domenec.puig@urv.cat santiago.romani@urv.cat hatem.abdellatif@urv.cat mohamed.abdelnasser@urv.cat |
| Author identifier: | 0000-0002-0562-4205 0000-0001-6673-9615 0000-0001-5421-1637 0000-0002-1074-2441 |
| Journal publication year: | 2021 |
| Publication Type: | Proceedings Paper |
| APA: | Schwarz Schuler, Joao Paulo; Romani, Santiago; Abdel-Nasser, Mohamed; Rashwan, Hatem; Puig, Domenec (2021). Grouped Pointwise Convolutions Significantly Reduces Parameters in EfficientNet. Amsterdam: IOS Press |
| Papper original source: | Frontiers In Artificial Intelligence And Applications. 339 383-391 |
| Abstract: | EfficientNet is a recent Deep Convolutional Neural Network (DCNN) architecture intended to be proportionally extendible in depth, width and resolution. Through its variants, it can achieve state of the art accuracy on the ImageNet classification task as well as on other classical challenges. Although its name refers to its efficiency with respect to the ratio between outcome (accuracy) and needed resources (number of parameters, flops), we are studying a method to reduce the original number of trainable parameters by more than 84% while keeping a very similar degree of accuracy. Our proposal is to improve the pointwise (1x1) convolutions, whose number of parameters rapidly grows due to the multiplication of the number of filters by the number of input channels that come from the previous layer. Basically, our tweak consists in grouping filters into parallel branches, where each branch processes a fraction of the input channels. However, by doing so, the learning capability of the DCNN is degraded. To avoid this effect, we suggest inter-leaving the output of filters from different branches at intermediate layers of consecutive pointwise convolutions. Our experiments with the CIFAR-10 dataset show that our optimized EfficientNet has similar learning capacity to the original layout when training from scratch. |
| Article's DOI: | 10.3233/FAIA210158 |
| Link to the original source: | https://ebooks.iospress.nl/doi/10.3233/FAIA210158 |
| Papper version: | info:eu-repo/semantics/publishedVersion |
| licence for use: | https://creativecommons.org/licenses/by/3.0/es/ |
| Department: | Enginyeria Informàtica i Matemàtiques |
| Licence document URL: | https://repositori.urv.cat/ca/proteccio-de-dades/ |
| Thematic Areas: | Artificial intelligence Ciências agrárias i Comunicació i informació Engenharias iii Engenharias iv General o multidisciplinar Información y documentación Interdisciplinar Medicina ii |
| Keywords: | Classification Cnn Computer vision Dcn Dcnn Deep learning Efficientnet |
| Entity: | Universitat Rovira i Virgili |
| Record's date: | 2024-09-21 |
| Search your record at: |   |
| File | Description | Format | |
|---|---|---|---|
| DocumentPrincipal | DocumentPrincipal | application/pdf |
© 2011 Universitat Rovira i Virgili