Mode-wise Tensor Decompositions: Multi-dimensional Generalizations of CUR Decompositions
Mode-wise Tensor Decompositions: Multi-dimensional Generalizations of CUR Decompositions
复制标题
DOI：
--
复制
发表时间：
2021-03
期刊：
影响因子：
--
通讯作者：
HanQin Cai;Keaton Hamm;Longxiu Huang;D. Needell
中科院分区：
文献类型：
--
作者：
HanQin Cai;Keaton Hamm;Longxiu Huang;D. Needell
低阶张量近似是现代机器学习和数据科学中的基本工具。在本文中，我们研究了两个主要的张量CUR近似，即Chidori和纤维CUR的特征，扰动分析，和一个有效的采样策略。我们刻画了低多线性秩张量的精确张量CUR分解。我们还提出了理论误差界的张量CUR近似时（对抗或高斯）噪声出现。此外，我们表明，低成本的均匀采样是足够的张量CUR近似，如果张量有一个不相干的结构。经验性能评估，与合成和真实世界的数据集，建立张量CUR近似的速度优势，比其他国家的最先进的低多线性秩张量近似。
Low rank tensor approximation is a fundamental tool in modern machine learning and data science. In this paper, we study the characterization, perturbation analysis, and an efficient sampling strategy for two primary tensor CUR approximations, namely Chidori and Fiber CUR. We characterize exact tensor CUR decompositions for low multilinear rank tensors. We also present theoretical error bounds of the tensor CUR approximations when (adversarial or Gaussian) noise appears. Moreover, we show that low cost uniform sampling is sufficient for tensor CUR approximations if the tensor has an incoherent structure. Empirical performance evaluations, with both synthetic and real-world datasets, establish the speed advantage of the tensor CUR approximations over other state-of-the-art low multilinear rank tensor approximations.