- Mlpconv layer with "micronetwork" with each conv layer to compute more abstract features for local patches（带有“微网”的MLPCCONV层的每个CONV层计算局部补丁的更多抽象特征）
- Micronetwork uses multilayer perceptron (FC, i.e. 1*1 conv layers)（“微网”使用多层感知器(FC,例如,1*1卷积层)）
- Precursor to GoogLeNet and ResNet "bottleneck" layers（是GoogLeNet和ResNet的“瓶颈”层的先驱）
- Philosophical inspiration for GoogLeNet（是GoogLeNet的哲学启发）