Processing math: 80%

等式约束的优化问题

g(x)是红线,f(x)由等值线构成的曲面。满足约束的极小值点具有以下特点:

f(x)g(x)在该点上梯度方向重合,可能同向或反向。

[1:13:26]等值约束的优化问题转化为解方程组问题

f(x)=λg(x)g(x)=0 

Many Options

Reparameterization
Eliminate constraints to reduce to unconstrained case

Newton’s method
Approximation: quadratic function with linear constraint

Penalty method
Augment objective with barrier term, e.g. f(x)+ρ|g(x)|

Alternating Projection

[1:14:18] C1,C2,C3的代表不同的等式约束。相当于坐标法推广

Augmented Lagrangians

[1:15:18] 把 Constrain 作为目标函数,极小划 权ρ在优化过程中会改变

Alternating Direction Method of Multipliers (ADMM)

[1:15:50] ADMM.把变量分离。把问题分成若干小问题分解为:小规模局部优化问题、闭式解全局优化问题。

min

\wedge _\rho (x,z;\lambda )=f(x)+g(z)+\lambda ^\top (Ax+Bz-c)+\frac{\rho }{2}||Ax+Bz-c||_2^2

https://web.stanford.edu/~boyd/papers/pdf/admm_slides.pdf

The Art of ADMM “Splitting”


本文出自CaterpillarStudyGroup,转载请注明出处。 https://caterpillarstudygroup.github.io/GAMES102_mdbook/