Hedge Backpropagation (HBP)

• Illustration of the proposed Online Deep Learning framework using Hedge Backpropagation (HBP).

  使用对冲反向传播(HBP)提出的在线深度学习框架的示例。

• The blue lines represent feedforward flow for computing hidden layer features.

  蓝线表示计算隐含层特征的前馈流。

• The orange lines indicate softmax output followed by the hedging combination at prediction time.

  橙色线表示softmax输出，后面是预测时间的对冲组合。

• The green lines indicate the online updating flows with the hedge backpropagation approach.

  绿线表示采用对冲反向传播方法的在线更新流程。

 

The final prediction of this model is a weighted combination of the predictions of all these classifiers, where the weight of each classifier is denoted by alpha(l)

该模型的最终预测是所有这些分类器预测的加权组合，其中每个分类器的权重表示为

 

for each classifer, the smoothing parameter is used for setting a minimum weight.

Once the weights of the classifier is updated, the weights are calculated  as shown,

 

https://www.ijrte.org/wp-content/uploads/papers/v8i5/E6337018520.pdf

Initialize F(x)=DNN with N hidden layers and N+1 classifiers f(n), 

alpha(n)=1/(N+1),