.png){kind=icon}
.png){kind=icon}
Return
Irregular convolution strategy based tensorized type-2 single layer feedforward network
.png){kind=icon}
PRE
AI
DOI：10.1007/s13042-023-01825-6.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Tensorized type-2 single layer feedforward network extends the single layer feedforward network with tensorized type-2 fuzzy structure. In the tensorized type-2 single layer feedforward network, type-2 fuzzy sets are used to generate tensor with lower membership function, principal membership function and upper membership function. Other compositions of the single layer feedforward network, such as defuzzification results, weighted averaged results and different type reduction results could also be formed by the tensorized fuzzy construction method. In thus doing, the type-reduction or defuzzification approach is the unimportance procedure in the fuzzy network operation and construction. To deeply unveil the implicit information hidden in the type-2 fuzzy sets, cross-shaped convolution with irregular convolution kernel is used to form the tensor. The named irregular convolution kernel based tensorized type-2 single layer feedforward network adopts an iterative tensor equation solving algorithm with tensor inequality constraint (Huang and Ma in Linear Multilinear Algebra, 1-24, 2021, https://doi.org/10.1080/03081087.2021.19541 40). Finally, the effectiveness of different convolution kernels for irregular convolution strategy based tensorized type-2 single layer feedforward network are tested. Comparisons are carried out on several benchmark datasets, and five different type-reduction methods for the irregular convolution strategy based tensorized type-2 single layer feedforward network are compared. Results show that the proposed learning method could be improved with this new information extraction strategy.
Keywords:
Tensorized type-2 fuzzy sets
Single layer feedforward network
Irregular kernels
Tensor regression
Journal
IF:
2.7
Papers:
3.1K
Citations:
5.6K