Softmax-loss layer：输出 loss 值

$L=-\sum{j} y{j} \log p_{j}$

layer { 
	name: "loss" 
    type: "SoftmaxWithLoss" 
    bottom: "ip1" 
    bottom: "label" 
    top: "loss" 
} 

Softmax layer: 输出似然值

$p{j}=\frac{e{j}^{o}}{\sum{k} e^{o{k}}}$

layers { 
    bottom: "cls3_fc" 
    top: "prob" 
    name: "prob" 
    type: “Softmax" 
}