权重Moore-彭罗斯逆与加权最小二乘问题的扰动条件

需积分: 0 0 下载量 44 浏览量 更新于2024-07-16 收藏 395KB PDF 举报
"这篇论文由王淑璠、郑兵和熊志平共同撰写,发表于兰州大学数学与统计学院,探讨了加权Moore-Penrose逆和加权线性最小二乘问题在扰动条件下的特性。文章指出,范数相对条件数是衡量输入数据对小扰动敏感度的关键指标,在数值分析中具有重要地位。作者在这项研究中,关注的是满列秩矩阵的加权Moore-Penrose逆的各种范数相对条件数公式,并得到了线性最小二乘(LS)问题的加权条件数的明确表达式。这些表达式适用于矩阵和向量加权范数的情况,假设rank(A+∆A) = rank(A)。这扩展了之前一些研究者的工作,并且受到了兰州大学的启动基金和甘肃省自然科学基金的支持。" 在本文中,作者深入探讨了加权Moore-Penrose逆的概念,这是一个在线性代数和数值分析中至关重要的工具,特别是在处理不完全或不精确数据时。Moore-Penrose逆是矩阵的广义逆,对于非方阵或奇异矩阵,它提供了类似于逆矩阵的解法。当考虑加权因素时,这可以更准确地反映实际问题中的情况,例如在处理带有不同重要性的观测值的最小二乘问题时。 加权线性最小二乘问题通常表示为寻找一个向量x,使得加权残差向量Ax - b的加权范数最小。这里的加权范数可能是由权重矩阵M定义的,即min_x ∥Ax - b∥_M。在计算和理论中,条件数是评估这类问题稳定性的一个关键指标,因为它量化了数据微小变化对解的影响程度。如果条件数很高,那么问题被认为是高度敏感的,即使微小的数据扰动也可能导致解的大幅变化。 文章的贡献在于给出了当矩阵A受到小扰动∆A时,保持其秩不变(即rank(A+∆A) = rank(A)),如何计算加权线性最小二乘问题的条件数的明确公式。这些结果不仅深化了对矩阵理论的理解,而且对数值方法的实际应用有着直接的指导意义,比如在数据拟合、误差分析和控制系统设计等领域。 这篇论文提供了加权Moore-Penrose逆和加权线性最小二乘问题在数值稳定性和敏感性分析方面的新见解,为未来的研究和工程实践提供了坚实的基础。

For macroscopically anisotropic media in which the variations in the phase stiffness tensor are small, formal solutions to the boundary-value problem have been developed in the form of perturbation series (Dederichs and Zeller, 1973; Gubernatis and Krumhansl, 1975 ; Willis, 1981). Due to the nature of the integral operator, one must contend with conditionally convergent integrals. One approach to this problem is to carry out a “renormalization” procedure which amounts to identifying physically what the conditionally convergent terms ought to contribute and replacing them by convergent terms that make this contribution (McCoy, 1979). For the special case of macroscopically isotropic media, the first few terms of this perturbation expansion have been explicitly given in terms of certain statistical correlation functions for both three-dimensional media (Beran and Molyneux, 1966 ; Milton and Phan-Thien, 1982) and two-dimensional media (Silnutzer, 1972 ; Milton, 1982). A drawback of all of these classical perturbation expansions is that they are only valid for media in which the moduli of the phases are nearly the same, albeit applicable for arbitrary volume fractions. In this paper we develop new, exact perturbation expansions for the effective stiffness tensor of macroscopically anisotropic composite media consisting of two isotropic phases by introducing an integral equation for the so-called “cavity” strain field. The expansions are not formal but rather the nth-order tensor coefficients are given explicitly in terms of integrals over products of certain tensor fields and a determinant involving n-point statistical correlation functions that render the integrals absolutely convergent in the infinite-volume limit. Thus, no renormalization analysis is required because the procedure used to solve the integral equation systematically leads to absolutely convergent integrals. Another useful feature of the expansions is that they converge rapidly for a class of dispersions for all volume fractions, even when the phase moduli differ significantly.

2023-06-02 上传

这一段讲的是什么:Abstract—A recent trojan attack on deep neural network (DNN) models is one insidious variant of data poisoning attacks. Trojan attacks exploit an effective backdoor created in a DNN model by leveraging the difficulty in interpretability of the learned model to misclassify any inputs signed with the attacker’s chosen trojan trigger. Since the trojan trigger is a secret guarded and exploited by the attacker, detecting such trojan inputs is a challenge, especially at run-time when models are in active operation. This work builds STRong Intentional Perturbation (STRIP) based run-time trojan attack detection system and focuses on vision system. We intentionally perturb the incoming input, for instance by superimposing various image patterns, and observe the randomness of predicted classes for perturbed inputs from a given deployed model—malicious or benign. A low entropy in predicted classes violates the input-dependence property of a benign model and implies the presence of a malicious input—a characteristic of a trojaned input. The high efficacy of our method is validated through case studies on three popular and contrasting datasets: MNIST, CIFAR10 and GTSRB. We achieve an overall false acceptance rate (FAR) of less than 1%, given a preset false rejection rate (FRR) of 1%, for different types of triggers. Using CIFAR10 and GTSRB, we have empirically achieved result of 0% for both FRR and FAR. We have also evaluated STRIP robustness against a number of trojan attack variants and adaptive attacks. Index Terms—Trojan attack, Backdoor attack

2023-07-24 上传