cs / machine-learning

gradient and Heissan

gradient and Heissan

gradient 的链式法则

Screenshot 2025-02-11 at 11.43.16

求 Heissan

Note:

对于 f:Rn→Rf:\mathbb{R}^n \rightarrow \mathbb{R}, $∇f:Rn→Rn \nabla f: \mathbb{R}^n \rightarrow \mathbb{R^n} $ 把一个点的值映射到这个点的导数的转置

而 Hessian 则是: $Hf(x)=D(∇f)(x) H_f(x) = D(\nabla f) (x) 可以写作: 可以写作: Hf(x)=∂∂x(∇f)(x) H_f(x) = \frac{\partial}{\partial x}(\nabla f) (x) $ image-20250211123059651

matrix-valued 函数

![Screenshot 2025-02-13 at 15.05.34](gradient and Heissan.assets/Screenshot 2025-02-13 at 15.05.34.png)

Screenshot 2025-02-13 at 15.06.26 Screenshot 2025-02-13 at 15.07.06

![Screenshot 2025-02-13 at 15.07.24](gradient and Heissan.assets/Screenshot 2025-02-13 at 15.07.24.png)

![Screenshot 2025-02-13 at 15.07.36](gradient and Heissan.assets/Screenshot 2025-02-13 at 15.07.43.png)