推荐星级：
- 1
- 2
- 3
- 4
- 5
基于交错组卷积的高效深度神经网络
资料介绍
Reducing the dimensionality of data with neural networks, Science,2006
· Fast learning algorithms for Restricted Boltzmann machine ImageNet Classification with dee-i conVolutional neural networks, NIPS,-2012.
· Dramatic performance improvement
· ImageNet, GPU
Two architecture design paths Going deeper
·Stack multiple blocks:vGG
·Improve information flow by skip connections
·GoogleNet,Highway,ResNet,Deeply-Fused Nets,FractalNets,DenseNets,Merge-and-run
Balance condition:To have minimum total #parameters,three conditions hold:
1)#parameters for(L-1)group convolutions are the same
2)#branches for(L-1) group convolutions are the same
|3)#channels in each branch are the same
部分文件列表
| 文件名 | 大小 |
| 基于交错组卷积的高效深度神经网络.pdf | 7M |
全部评论(0)