Search

Deep neural network-based object detectors are vulnerable to adversarial examples. Among existing works to fool object detectors, the camouflage-based method is more often adopted due to its adaptation to multi-view scenarios and non-planar objects. However, most of them can still be easily observed by human eyes, which limits their application in the real world. To fool human eyes and object detectors simultaneously, we propose a differential evolution based dual adversarial camouflage method. Specifically, we try to obtain the camouflage texture by the two-stage training, which can be wrapped over the surface of the object. In the first stage, we optimize the global texture to minimize the discrepancy between the rendered object and the scene background, making human eyes difficult to distinguish. In the second stage, we design three loss functions to optimize the local texture, which is selected from the global texture, making object detectors ineffective. In addition, we introduce the differential evolution algorithm to search for the near-optimal areas of the object to attack, improving the adversarial performance under certain attack area limitations. Experimental results show that our proposed method can obtain a good trade-off between fooling human eyes and object detectors under multiple specific scenes and objects., Competing Interests: Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper., (Copyright © 2023. Published by Elsevier Ltd.)

Radial basis function neural networks (RBFNNs) are widely used in nonlinear function approximation. One of the challenges in RBFNN modeling is determining how to effectively optimize width parameters to improve approximation accuracy. To solve this problem, a width optimization method, concurrent subspace width optimization (CSWO), is proposed based on a decomposition and coordination strategy. This method decomposes the large-scale width optimization problem into several subspace optimization (SSO) problems, each of which has a single optimization variable and smaller training and validation data sets so as to greatly simplify optimization complexity. These SSOs can be solved concurrently, thus computational time can be effectively reduced. With top-level system coordination, the optimization of SSOs can converge to a consistent optimum, which is equivalent to the optimum of the original width optimization problem. The proposed method is tested with four mathematical examples and one practical engineering approximation problem. The results demonstrate the efficiency and robustness of CSWO in optimizing width parameters over the traditional width optimization methods.

In practical engineering, the layout optimization technique driven by the thermal performance is faced with a severe computational burden when directly integrating the numerical analysis tool of temperature simulation into the optimization loop. To alleviate this difficulty, this paper presents a novel deep learning surrogate-assisted heat source layout optimization method. First, two sampling strategies, namely the random sampling strategy and the evolving sampling strategy, are proposed to produce diversified training data. Then, regarding mapping between the layout and the corresponding temperature field as an image-to-image regression task, the feature pyramid network (FPN), a kind of deep neural network, is trained to learn the inherent laws, which plays as a surrogate model to evaluate the thermal performance of the domain with respect to different input layouts accurately and efficiently. Finally, the neighborhood search-based layout optimization (NSLO) algorithm is proposed and combined with the FPN surrogate to solve discrete heat source layout optimization problems. A typical two-dimensional heat conduction optimization problem is investigated to demonstrate the feasibility and effectiveness of the proposed deep learning surrogate-assisted layout optimization framework. [ABSTRACT FROM AUTHOR]

The reliability analysis of structural systems is generally difficult when limit state function (LSF) is implicitly defined especially by black-box models in engineering. To balance analysis efficiency and accuracy, an improved approach named BIS-FC is proposed in this paper based on the Beta-spherical importance sampling (BIS) framework with the innovative concept of critical region defined by combination with First Order Reliability Method (FORM). The critical region is defined by the hyper-tangent plane of LSF at the Most Probable Point (MPP) and its parallel hyperplanes according to the convex or concave features of LSF, wherein the samples are of both high occurrence probability and high misjudgment risk due to the linearization assumption of FORM. BIS-FC only conducts LSF analysis for the BIS samples located in the critical region, and the other samples are directly identified as safe or failure according to the linearized hyperplanes. Thus large computational cost can be saved compared to the original BIS, and meanwhile, the analysis accuracy can be greatly enhanced compared to FORM. An iterative process is proposed to properly define the critical region, based on which reliability is analyzed sequentially until the stopping criteria for desired estimation error level and stable convergence are satisfied. The algorithms of BIS-FC for both single and multiple MPP situations are developed and testified with six numerical examples and one satellite structural engineering problem. The results demonstrate the effectiveness of BIS-FC regarding good balance between efficiency and accuracy. [ABSTRACT FROM AUTHOR]

Satellite layout optimization design problems with complicated performance constraints are studied in this paper. In addition to the traditional geometric constraint, system centroid constraint, inertia angles constraint and dynamics performance, more complex design factors based on practical engineering requirements are considered, including thermal performance, CMA (compatibility, maintainability and accessibility) constraint and the special rules of the placement of special components. An enhanced interference algorithm based on finite-circle method (FCM) is proposed to handle CMA constraint, which can effectively control the distance between any two components. Moreover, to provide system engineers with the information of design space about dynamics and thermal performance, the accelerated particle swarm optimization (APSO) is proposed to optimize the constructed layout model globally. Finally, the feasibility and effectiveness of the proposed methodology are validated by a numerical test and an engineering example. [ABSTRACT FROM AUTHOR]

Abstract: To assess the on-orbit servicing (OOS) paradigm and optimize its utilities by taking advantage of its inherent flexibility and responsiveness, the OOS system assessment and optimization methods based on lifecycle simulation under uncertainties are studied. The uncertainty sources considered in this paper include both the aleatory (random launch/OOS operation failure and on-orbit component failure) and the epistemic (the unknown trend of the end-used market price) types. Firstly, the lifecycle simulation under uncertainties is discussed. The chronological flowchart is presented. The cost and benefit models are established, and the uncertainties thereof are modeled. The dynamic programming method to make optimal decision in face of the uncertain events is introduced. Secondly, the method to analyze the propagation effects of the uncertainties on the OOS utilities is studied. With combined probability and evidence theory, a Monte Carlo lifecycle Simulation based Unified Uncertainty Analysis (MCS-UUA) approach is proposed, based on which the OOS utility assessment tool under mixed uncertainties is developed. Thirdly, to further optimize the OOS system under mixed uncertainties, the reliability-based optimization (RBO) method is studied. To alleviate the computational burden of the traditional RBO method which involves nested optimum search and uncertainty analysis, the framework of Sequential Optimization and Mixed Uncertainty Analysis (SOMUA) is employed to integrate MCS-UUA, and the RBO algorithm SOMUA-MCS is developed. Fourthly, a case study on the OOS system for a hypothetical GEO commercial communication satellite is investigated with the proposed assessment tool. Furthermore, the OOS system is optimized with SOMUA-MCS. Lastly, some conclusions are given and future research prospects are highlighted. [Copyright &y& Elsevier]

Abstract: This paper presents a comprehensive review of Uncertainty-Based Multidisciplinary Design Optimization (UMDO) theory and the state of the art in UMDO methods for aerospace vehicles. UMDO has been widely acknowledged as an advanced methodology to address competing objectives of aerospace vehicle design, such as performance, cost, reliability and robustness. However the major challenges of UMDO, namely the computational complexity and organizational complexity caused by both time-consuming disciplinary analysis models and UMDO algorithms, still greatly hamper its application in aerospace engineering. In recent years there is a surge of research in this field aiming at solving these problems. The purpose of this paper is to review these existing approaches systematically, highlight research challenges and opportunities, and help guide future efforts. Firstly, the UMDO theory preliminaries are introduced to clarify the basic UMDO concepts and mathematical formulations, as well as provide a panoramic view of the general UMDO solving process. Then following the UMDO solving process, research progress of each key step is separately surveyed and discussed, specifically including uncertainty modeling, uncertainty propagation and analysis, optimization under uncertainty, and UMDO procedure. Finally some conclusions are given, and future research trends and prospects are discussed. [Copyright &y& Elsevier]

Fuzzy linear regression with crisp inputs and fuzzy output data constitutes an important modeling problem. Basic strategies used to solve this problem, i.e., the possibilistic method and the least squares method, together with their extensions, have some drawbacks. The possibilistic methods put emphasis on an inclusion property while the least squares methods focus on a central tendency property. Therefore, many researchers work on combining these two methods to obtain a better performance. In this paper, in contrast to most existing techniques which treat fuzzy linear regression as an optimization problem, we set the problem of constructing a fuzzy linear regression model in Bayesian statistics and propose a new fuzzy linear regression method based on approximate Bayesian computation (ABC). The method applies the likelihood-free inference algorithm ABC to generate independent samples of unknown model coefficients from Bayesian posterior distribution. This overcomes difficulty of defining likelihood function in fuzzy environment. By adjusting a prior distribution and a threshold of the ABC algorithm, the proposed approach can flexibly balance the inclusion property of the possibilistic methods and the central tendency property of the least squares methods. The convergence property of the proposed ABC algorithm is verified by a numerical example. Two measuring criteria, i.e., a distance metric and a degree of fitting index, which indicate the central tendency property and the inclusion property, respectively, are introduced to evaluate the quality of regression results. Three numerical examples are applied to show the performances of the proposed method. The numerical results are also compared with those obtained by some classical and recently proposed approaches. Additionally, a practical engineering application example is used to illustrate effectiveness of the proposed method. • Set the problem of constructing a fuzzy linear regression model in Bayesian statistics. • Apply the likelihood-free inference algorithm approximate Bayesian computation. • Applied to a practical engineering application example. [ABSTRACT FROM AUTHOR]

The phenomenon of adversarial examples has been revealed in variant scenarios. Recent studies show that well-designed adversarial defense strategies can improve the robustness of deep learning models against adversarial examples. However, with the rapid development of defense technologies, it also tends to be more difficult to evaluate the robustness of the defensed model due to the weak performance of existing manually designed adversarial attacks. To address the challenge, given the defensed model, the efficient adversarial attack with less computational burden and lower robust accuracy is needed to be further exploited. Therefore, we propose a multi-objective memetic algorithm for auto adversarial attack optimization design, which realizes the automatic search for the near-optimal adversarial attack towards defensed models. Firstly, the more general mathematical model of auto adversarial attack optimization design is constructed, where the search space includes not only the attacker operations, magnitude, iteration number, and loss functions but also the connection ways of multiple adversarial attacks. In addition, we develop a multi-objective memetic algorithm combining NSGA-II and local search to solve the optimization problem. Finally, to decrease the evaluation cost during the search, we propose a representative data selection strategy based on sorting cross entropy loss values of images output by models. Experiments on CIFAR10, CIFAR100, and ImageNet datasets show that our method can efficiently find near-optimal adversarial attacks with lower robust accuracy and less time cost, which can provide more reliable and efficient robustness evaluation for defensed models. [ABSTRACT FROM AUTHOR]

• Multistate compression and inference algorithms are applicable to any complex systems. • Given the evidence, backward inference algorithm can update the probability distributions of all nodes. • The potential application of the proposed algorithms in the reliability-based optimization for complex engineering systems. As the number of complex multistate systems' components increases, one major challenge to analyze the reliabilities of complex multistate systems by Bayesian network (BN) is that the memory storage requirements (MSRs) of conditional probability table (CPT) increase exponentially. When the components reach a certain amount, the MSRs of CPT will exceed the computer's random access memory (RAM). To solve this problem, this two-part paper proposes a novel multistate compression algorithm to compress the CPT so that the MSRs of CPT can be reduced apparently. In this Part I, an independent multistate inference algorithm is proposed to perform the inference of BN based on the compressed CPT for the complex multistate independent systems. Given the evidence of system, the backward inference algorithm is proposed to update the probability distributions of compoents. The above proposed algorithms can be generally applied to any complex multistate independent system without constraints on system structure and state configurations. In addition, the Part II studies the application of compression idea in the complex multistate dependent systems. Finally, two case studies are used to validate the performance of the proposed algorithms. [ABSTRACT FROM AUTHOR]