1. Identity statement | |
Reference Type | Conference Paper (Conference Proceedings) |
Site | sibgrapi.sid.inpe.br |
Identifier | 8JMKD3MGPEW34M/49KDKC2 |
Repository | sid.inpe.br/sibgrapi/2023/08.11.19.30 |
Last Update | 2023:08.11.19.30.50 (UTC) rogeriouff@yahoo.com.br |
Metadata Repository | sid.inpe.br/sibgrapi/2023/08.11.19.30.50 |
Metadata Last Update | 2024:02.17.04.05.19 (UTC) administrator |
DOI | 10.1109/SIBGRAPI59091.2023.10347043 |
Citation Key | MoraesEvPeToFeMa:2023:HeReLi |
Title | Heuristics to reduce linear combinations of activation functions to improve image classification |
Format | On-line |
Year | 2023 |
Access Date | 2024, May 15 |
Number of Files | 1 |
Size | 201 KiB |
|
2. Context | |
Author | 1 Moraes, Rogério Ferreira de 2 Evangelista, Raphael dos S. 3 Pereira, Andre Luiz da S. 4 Toledo, Yanexis Pupo 5 Fernandes, Leandro A. F. 6 Martí, Luis |
Affiliation | 1 Universidade Federal Fluminense (UFF), Niterói, Brazil 2 Universidade Federal Fluminense (UFF), Niterói, Brazil 3 Universidade Federal Fluminense (UFF), Niterói, Brazil 4 Universidade Federal Fluminense (UFF), Niterói, Brazil 5 Universidade Federal Fluminense (UFF), Niterói, Brazil 6 Inria Chile Research Center, Las Condes, Chile |
Editor | Clua, Esteban Walter Gonzalez Körting, Thales Sehn Paulovich, Fernando Vieira Feris, Rogerio |
e-Mail Address | rogeriouff@yahoo.com.br |
Conference Name | Conference on Graphics, Patterns and Images, 36 (SIBGRAPI) |
Conference Location | Rio Grande, RS |
Date | Nov. 06-09, 2023 |
Book Title | Proceedings |
Tertiary Type | Full Paper |
History (UTC) | 2023-08-11 19:30:50 :: rogeriouff@yahoo.com.br -> administrator :: 2024-02-17 04:05:19 :: administrator -> rogeriouff@yahoo.com.br :: 2023 |
|
3. Content and structure | |
Is the master or a copy? | is the master |
Content Stage | completed |
Transferable | 1 |
Keywords | learned activation function trainable activation function linear combination of activation functions |
Abstract | Image classification is one of the classical problems in computer vision, and CNNs (Convolutional Neural Networks) are widely used for this task. However, the choice of a CNN can vary depending on the chosen dataset. In this context, we have trainable activation functions that are crucial in CNNs and adapt to the data. One technique for constructing these functions is to write them as a linear combination of other activation functions, where the coefficients of this combination are learned during training. However, if we have a large number of activation functions to combine, the computational cost can be very high, and manually testing and choosing these functions may be impractical, depending on the number of available activation functions. To alleviate the difficulty of choosing which activation functions should be part of the linear combination, we propose two heuristics: Linear Combination Approximator by Coefficients (LCAC) and Major and Uniform Coefficient Extractor (MUCE). Our heuristics provide an efficient selection of a subset of activation functions so that their results are better or equivalent to the linear combination that uses all 34 available activation functions in our experiments (C34), considering the image classification problem. Compared to the C34 function, the LCAC function was better or equivalent in 62.5%, and the MUCE function in 87.5% of the conducted experiments. |
doc Directory Content | access |
source Directory Content | there are no files |
agreement Directory Content | |
|
4. Conditions of access and use | |
data URL | http://urlib.net/ibi/8JMKD3MGPEW34M/49KDKC2 |
zipped data URL | http://urlib.net/zip/8JMKD3MGPEW34M/49KDKC2 |
Language | en |
Target File | Moraes-paper50.pdf |
User Group | rogeriouff@yahoo.com.br |
Visibility | shown |
|
5. Allied materials | |
Mirror Repository | sid.inpe.br/banon/2001/03.30.15.38.24 |
Host Collection | sid.inpe.br/banon/2001/03.30.15.38 |
|
6. Notes | |
Empty Fields | archivingpolicy archivist area callnumber contenttype copyholder copyright creatorhistory descriptionlevel dissemination documentstage edition electronicmailaddress group holdercode isbn issn label lineage mark nextedition nexthigherunit notes numberofvolumes orcid organization pages parameterlist parentrepositories previousedition previouslowerunit progress project publisher publisheraddress readergroup readpermission resumeid rightsholder schedulinginformation secondarydate secondarykey secondarymark secondarytype serieseditor session shorttitle sponsor subject tertiarymark type url versiontype volume |
|
7. Description control | |
e-Mail (login) | rogeriouff@yahoo.com.br |
update | |
|