面向含噪数据流的鲁棒在线学习算法

基分类器

Logistic

Linear SVM

BernoulliNB

Perceptron

PassiveAggressiveClassifier

计算公式

Ramp_loss

Calculate_Weight

Parameter:

$$ \eta > 0 $$

Initialize:

$$ w_1 = (1/d,...,1/d) $$

$x^4$

Update rule

$$

\forall i,w_{t+1}[i] = \frac{w_t[i]e^{-\eta z_t[i]}}{\sum_jw_t[j]e^{-\eta z_t[j]}} \quad

\ z_t[i] = \left{
\begin{array}{lr}
0 \quad h_i(x) = y& \
1 \quad h_i(x) \neq y &
\end{array}
\right. \ h_i(x)为第i个base_model的预测标签 \ \eta 为初始指定参数

$$