arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2412.02597math.NAcs.NA
收藏

快速随机Kronecker张量分解:算法与误差分析

Fast randomized Kronecker tensor decomposition: algorithms and error analysis

Research Center of the Artificial Intelligence Institute, Innopolis University(因诺波利斯大学人工智能研究院研究中心);Lab of Machine Learning and Knowledge Representation, Innopolis University(因诺波利斯大学机器学习与知识表示实验室);Peoples’ Friendship University of Russia(俄罗斯人民友谊大学);Department of Teleinformatics Engineering, Federal University of Ceara(塞阿拉联邦大学电信信息工程系);School of Information and Communication Engineering, University of Electronic Science and Technology of China (UESTC)(电子科技大学信息与通信工程学院)

Salman Ahmadi-Asl, Naeim Rezaeian, Andre L. F. de Almeida, Yipeng Liu

AI总结:

该研究针对Kronecker张量分解问题,提出用含过采样和幂迭代的随机SVD替代确定性SVD的快速算法,能显著加快分解速度,经理论分析和实验验证,相比确定性方法有数量级加速,且可应用于多种领域。

摘要:

本文提出了快速随机算法来计算Kronecker张量分解(KTD),通过在TTr1SVD框架中用包含过采样和幂迭代的随机奇异值分解(SVD)替代确定性SVD序列。所提算法能比现有最先进的确定性方法更快地将给定张量分解为KTD格式。主要思路是利用随机化降低计算复杂度并保持精度可控。给出了详细理论分析,包括考虑通过TTr1SVD树结构的误差传播的递归误差界。证明了期望的Frobenius范数误差由尾能量之和乘以随幂迭代次数指数衰减的因子界定。在合成和真实数据集上的大量模拟表明与确定性方法相比有几个数量级的加速,并应用于张量补全、视频/图像压缩、图像去噪和图像超分辨率。

英文摘要:

This paper proposes fast randomized algorithms for computing the Kronecker Tensor Decomposition (KTD) by replacing the sequence of deterministic SVDs in the TTr1SVD framework with randomized SVDs incorporating oversampling and power iterations. The proposed algorithms can decompose a given tensor into the KTD format significantly faster than existing state-of-the-art deterministic methods. Our principal idea is to use randomization to reduce computational complexity while maintaining controlled accuracy. A detailed theoretical analysis is presented, including a recursive error bound that accounts for error propagation through the TTr1SVD tree structure. We prove that the expected Frobenius norm error is bounded by a sum of tail energies multiplied by factors that decay exponentially with the number of power iterations. Extensive simulations on synthetic and real-world datasets demonstrate several orders of magnitude acceleration compared to the deterministic approach, with applications to tensor completion, video/image compression, image denoising, and image super-resolution.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑