Local minima found in the subparameter space can be effective for ensembles of deep convolutional neural networks
作者:
Highlights:
• Local minima found in the subparameter space can be effective for ensembles of deep convolutional neural networks (CNNs).
• Finding local minima in the subparameter space makes the training stage for ensembles of deep CNNs more affordable.
• Multiple models obtained at the found local minima can be selected to achieve better ensemble results via ensemble selection.
• The selected models for ensemble of deep CNNs can be fused in the subparameter space to reduce the expense at the testing stage.
摘要
•Local minima found in the subparameter space can be effective for ensembles of deep convolutional neural networks (CNNs).•Finding local minima in the subparameter space makes the training stage for ensembles of deep CNNs more affordable.•Multiple models obtained at the found local minima can be selected to achieve better ensemble results via ensemble selection.•The selected models for ensemble of deep CNNs can be fused in the subparameter space to reduce the expense at the testing stage.
论文关键词:Ensemble learning,Ensemble selection,Ensemble fusion,Deep convolutional neural network
论文评审过程:Received 25 September 2019, Revised 1 July 2020, Accepted 7 August 2020, Available online 8 August 2020, Version of Record 16 August 2020.
论文官网地址:https://doi.org/10.1016/j.patcog.2020.107582