We present randUBV, a randomized algorithm for matrix sketching based on the block Lanzcos bidiagonalization process. Given a matrix $\bf{A}$, it produces a low-rank approximation of the form ${\bf UBV}^T$, where $\bf{U}$ and $\bf{V}$ are approximately orthogonal and $\bf{B}$ is block bidiagonal. It is closely related to the randQB algorithms of Yu, Gu, and Li (2018) in that the entries of $\bf{B}$ are incrementally generated and the Frobenius norm approximation error may be efficiently estimated. Our algorithm is therefore suitable for the fixed-precision problem, and so is designed to terminate as soon as a user input error tolerance is reached. Numerical experiments suggest that the block Lanczos method can be competitive with or superior to algorithms that use power iteration, even when $\bf{A}$ has significant clusters of singular values.
翻译:我们提出了基于 Lanzcos feriagonalization 过程的矩阵草图随机算法 RandusUBVVV。 根据一个矩阵 $\bf{A} $, 它生成了一种以$\bf UBV}T$为单位的低排序近似值, 美元和$\bf{U}美元大致是正方形的, 美元和$\bf{B}$是方形的。 它与 u、 Gu 和 Li (2018) 的 RandQB 算法密切相关, $\bf{B} 的条目是递增生成的, Frobenius 规范近似差可能是有效的估计。 因此, 我们的算法适合固定精度问题, 并因此设计在用户输入错误容忍度达到后立即终止。 数字实验表明, Lanczos 块方法可以与使用电动算法具有竞争力或优越性, 即使$\bf{A} $ 有相当的奇数。