海森矩阵(Hessian Matrix)
Key Words: Hessian Matrix, second order derivatives, convexity, and saddle points
原文链接:Hessian, second order derivatives, convexity, and saddle points
翻译:
Hessian Matrix:二阶导和函数曲率
回忆一下 f f f的梯度 f : R n → R f:\mathbb{R}^n \rightarrow \mathbb{R} f:Rn→R:
f ′ ( x ) = [ δ f δ x 1 , δ f δ x 2 , ⋯ , δ f δ x n ] (1) f'(x) = [\frac{\delta f}{\delta x_1}, \frac{\delta f}{\delta x_2}, \cdots, \frac{\delta f}{\delta x_n} ]\tag{1} f′(x)=[δx1δf,δx2δf,⋯,δxnδf](1)
求 f f f的二阶导意味着,我们可以看到第 i i i个一阶导 δ f δ x i \frac{\delta f}{\delta x_i} δxiδf是如何影响其他所有偏导数的。求二阶导相当于求 f ′ ( x ) f'(x) f′(x)的Jacobian矩阵,结果如公式(2)所示。
H = [ ∂ 2 f ∂ x 1 2 ∂ 2 f ∂ x 1 ∂ x 2 ⋯ ∂ 2 f ∂ x 1 ∂ x n ∂ 2 f ∂ x 2 ∂ x 1 ∂ 2 f ∂ x 2 2 ⋯ ∂ 2 f ∂ x 2 ∂