论文标题
线性收敛的高斯 - 纽顿亚级别方法，用于解决问题的问题
A linearly convergent Gauss-Newton subgradient method for ill-conditioned problems
论文作者
论文摘要
我们分析了一种优化复合功能$ H \ Circ C $的预处理下级别方法，其中$ h $是本地Lipschitz功能，$ c $是平滑的非线性映射。我们证明，当$ c $满足恒定的排名属性，而$ h $是半牙，并且在$ c $的图像上是敏锐的，方法是线性收敛的。与标准的亚级别方法相反，其甲骨文的复杂性在重新聚集$ c $的情况下是不变的。
We analyze a preconditioned subgradient method for optimizing composite functions $h \circ c$, where $h$ is a locally Lipschitz function and $c$ is a smooth nonlinear mapping. We prove that when $c$ satisfies a constant rank property and $h$ is semismooth and sharp on the image of $c$, the method converges linearly. In contrast to standard subgradient methods, its oracle complexity is invariant under reparameterizations of $c$.
