1
数据矩阵为对角线时套索问题的闭式解
\newcommand{\diag}{\operatorname{diag}}我们遇到了问题:\ min_ {w \ in \ mathbb {R} ^ {d}} \ left(\ frac {1} {n} \ sum_ {i = 1} ^ {n} \ left(\ langle w,x_ {i} \ rangle-y_ {i} \ right)^ {2} +2 \ lambda || w || _1 \ right),minw∈Rd(1n∑i=1n(⟨w,xi⟩−yi)2+2λ||w||1),minw∈Rd(1n∑i=1n(⟨w,xi⟩−yi)2+2λ||w||1),\min_{w\in\mathbb{R}^{d}}\left( \frac{1}{n}\sum_{i=1}^{n} \left( \langle w,x_{i}\rangle-y_{i} \right)^{2} +2\lambda||w||_1\right), 并假设:∑i=1nxixTi=diag(σ21,...,σ2d).∑i=1nxixiT=diag(σ12,...,σd2).\sum_{i=1}^nx_ix_i^T=\diag(\sigma_1^2,...,\sigma_d^2). 在这种情况下是否有封闭形式的解决方案? …