Soft Thresholding 求解优化问题例子

软阈值(SoftThresholding)可以求解如下优化问题：

$$\underset{x}{\mathrm{argmin}}||x-B|{|}_2^2+\lambda ||x|{|}_1$$

的解为：

$$soft(B,\lambda /2)=\{\begin{array}{rlrl}B+\lambda /2& ,& B& <-\lambda /2\\ 0& ,& |B|& <\lambda /2\\ B-\lambda /2& ,& B& >\lambda /2\end{array}$$

具体参考证明

例子：

来源于文章Newton, M.A., Polson, N.G. and Xu, J. (2020), Weighted Bayesian bootstrap for scalable posterior distributions. Can J Statistics. https://doi.org/10.1002/cjs.11570

θ w ∗ = arg min ⁡ θ ∈ Θ { w 1 2 ( y − θ ) 2 + λ w 0 ∣ θ ∣ } \theta_w^* = \argmin_{\theta \in \Theta}\{\frac{w_1}{2}(y-\theta)^2+\lambda w_0 |\theta| \} θw∗​=θ∈Θargmin​{2w1​​(y−θ)2+λw0​∣θ∣}

利用soft thresholding解得：

$${\theta }_w^{\ast }=\{\begin{array}{rlrl}y-\lambda w_0/w_1& ,& & ify>\lambda w_0/w_1\\ y+\lambda w_0/w_1& ,& & ify<-\lambda w_0/w_1\\ 0& ,& & if|y|⩽\lambda w_0/w_1\end{array}$$