.png){kind=icon}
.png){kind=icon}
Return
Broad learning system based ensemble deep model
.png){kind=icon}
PRE
AI
DOI：10.1007/s00500-022-07004-z.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Broad learning system (BLS) demonstrates a novel structure of neural networks based on random vector functional link network (RVFL), which has a faster modeling speed, better generalization ability, higher regression accuracy for solving regression tasks. To improve the feature representation capacity of BLS and guarantee the training efficiency, a novel ensemble deep model based on broad learning system is proposed, named as EDBLS. Unlike other deep or ensemble models of BLS, the results of EDBLS can be calculated by training the model only once. Meanwhile, the performance of BLS mainly depends on its structure, it is the same as EDBLS, we need to select network parameters for different real-world problems. As the lower efficiency of grid search, it is no longer suitable for EDBLS with more parameters. Therefore, coot optimization algorithm (COOT) is applied to determine the network structure of EDBLS for regression problems, it is important to note that the COOT is only a tool for determining the structure of EDBLS. Finally, the experimental results on MNIST, NORB and some regression benchmark datasets show that EDBLS has a better classification and regression performance, COOT is beneficial to select the network parameters of EDBLS for regression.
Keywords:
Broad learning system
Ensemble deep learning
Coot optimization algorithm
Neural network
Optimization
Journal
IF:
2.5
Papers:
1.0W
Citations:
2.1W
Organization
No organization information available