人工智能导论¶
Bias vs. Variance¶
2022年4月23日，2023年10月12日。
Mean squared error = variance + bias squared.
\(y\) is the true value, and \(\hat{y}\) is a random variable or a list of variables.
\[
\begin{split}
\overline{(\hat y - y)^2}
&= \overline{(\hat y)^2 + y^2 - 2 \hat y y} \\
&= \overline{(\hat y)^2} + {y^2} - 2 \overline{\hat y} y \\
&= \overline{(\hat y)^2} - \overline{\hat y}^2
+ \qty(\overline{\hat y}^2 + y^2 - 2 \overline{\hat y} y) \\
&= \overline{(\hat y - \overline{\hat y})^2}
+ \qty(\overline{\hat y} - y)^2. \\
\end{split}
\]
Dual Formulation¶
Lagrange Multiplier and Dual Formulation · SVM (gitbooks.io)