查看更多>>摘要：Bayesian neural networks (BNNs) are used in many tasks because they provide a probabilistic representation of deep learning models by placing a distribution over the model parameters. Although BNNs are a more robust deep learning paradigm than vanilla deep neural networks, their ability to handle adversarial attacks in practice remains limited. In this study, we propose a novel multi-task adversarial training approach for improving the adversarial robustness of BNNs. Specifically, we first generate diverse and stronger adversarial examples for adversarial training by maximising a multi-task loss. This multi-task loss is a combination of the unsupervised feature scattering loss and supervised margin loss. Then, we find the model parameters by minimising another multi-task loss composed of the feature loss and variational inference loss. The feature loss is defined based on distance parallel to l parallel to(p), which measures the difference between the two feature representations extracted from the clean and adversarial examples. Minimising the feature loss improves the feature similarity and helps the model learn more robust features, resulting in enhanced robustness. Extensive experiments are conducted on four benchmark datasets in white-box and black-box attack scenarios. The experimental results demonstrate that the proposed approach significantly improves the adversarial robustness compared with several state-of-the-art defence methods. (C) 2022 Elsevier Inc. All rights reserved.
查看更多>>摘要：Surrogate-assisted evolutionary algorithms (SAEAs) are promising methods for addressing computationally expensive problems. This paper proposes a multiple surrogates and offspring-assisted differential evolution (MSODE) algorithm for high-dimensional expensive problems. Ensemble models consisting of multiple base models are built based on bagging. The ensemble models contribute toward reducing the variations of predictions and the uncertainty of base models. The algorithm proposes a multiple-offspring evolution strategy in which it generates multiple offspring for each parent individual to enhance the search ability of the population. The appropriate number of offspring was investigated, considering the tradeoff between optimization results and efficiency. MSODE consists of a global prescreening search, local search, and uncertainty prescreening search. The global prescreening search adopts a global ensemble to prescreen promising offspring. The local search accelerates the convergence by searching for the optimum of a local ensemble. The uncertainty prescreening search requires a number of fitness evaluations that slightly deteriorate the results. A comprehensive analysis was conducted to determine the optimal parameter settings. MSODE was compared with meta-heuristic algorithms and SAEAs on a series of benchmark problems. The results show that MSODE behaves better than most algorithms and is competitive against the best ones. (C) 2022 Elsevier Inc. All rights reserved.
查看更多>>摘要：Pairwise comparison matrices are often used in multicriteria decision making (MCDM). The most critical part of this technique is the inconsistency, which emerges for logical reasons but can cause significant problems during the decision making process. Hence it is necessary to keep inconsistency below an acceptable threshold. In order to support the decision maker (DM) in making a rational decision, we must keep in mind the following: Our suggestion should be as close to the DM's original result as possible, moreover it should have as low inconsistency as possible. We have studied various linear programming (LP) models that are used for reducing the inconsistency of pairwise comparison matrices (PCM) (Bozoki et al., 2011, 2015). These models, however aim at fulfilling only one of the previously mentioned two goals at a time. Therefore, the optimal solutions given may differ from each in the other respect, which is not taken as an objective but as a constraint in the model. So, they cannot be considered as equally good optimal solutions from a wider perspective. Based on our experiences concerning these models, we have developed a mixed-integer linear-fractional programming (MILFP) model that takes both mentioned goals as objectives by combining them into a linear-fractional objective function. We also provide the linear analogue (LA) of our MILFP model using an appropriate adaptation and combination of the Charnes-Cooper transformation and Glover's linearization scheme. (C) 2022 The Author(s). Published by Elsevier Inc.
查看更多>>摘要：Fuzzy inference systems, referring to a system that works on fuzzy sets, have been used in many areas such as classification, information order, especially in the field of forecasting. Because fuzzy inference systems are based on certain rules, determining these rules is the most important problem of many well-known fuzzy inference systems in the literature. To overcome this problem, the Type-1 fuzzy regression functions approach uses fuzzy functions instead of relations, unlike many fuzzy inference systems that operate on a rule base and establish a compound relation between the input and output of the system. One of the most important success criteria of Type-1 fuzzy regression functions is the type of clustering method. The Gustafson Kessel clustering algorithm has no limitations unlike the fuzzy clustering algorithm and can recognize ellipsoidal clusters. In this study, the Gustafson Kessel clustering algorithm is used instead of the fuzzy clustering algorithm and thus the membership values of the input set are obtained with the Gustafson Kessel clustering algorithm in the structure of the fuzzy regression functions approach. The analysis results show that the forecasting performance is increased considerably with the use of the Gustafson Kessel clustering algorithm. (C) 2022 Elsevier Inc. All rights reserved.
查看更多>>摘要：The ensemble methods are meta-algorithms that combine several base machine learning techniques to produce one optimal predictive model. Many existing committees of classifiers use feature space partitioning to determine the decision regions in which the selection of base classifiers occur or to learn a diverse ensemble of base classifiers. Furthermore, the division of the feature space into subspaces does not depend on the predictive model that defines the decision boundaries. Therefore, we propose a novel ensemble learning algorithm based on the feature space partitioning defined by decision tree boundaries. In addition, in a proposed framework, the feature subspace selection occurs taking class label imbalance ratio into account. The proposed method defines the ensemble class label based on selected previously feature subspaces and a neighborhood of feature subspaces defined by its reference point. Experimental results indicate that our proposed method is more effective than state-of-the-art ensemble methods on twenty-seven benchmark datasets regarding seven representative classification performance measures. (C) 2022 Elsevier Inc. All rights reserved.
查看更多>>摘要：The stochastic stability analysis for particle swarm optimization (PSO) has met formidable challenges due to the intricate influences on particles' motion from collaboration capability, cognitive competence, selective randomness and environmental complexity. This paper firstly regards the above-mentioned factors as forces acting on particles of PSO: interaction, damping, randomness and external forces. Based on this mechanical analogy, a novel stochastic composite dynamic model (SCDM) for PSO is proposed which can be classified into two parts: diffusion process and drift process. The diffusion process can not be easily obtained by analytical analysis, but its approximate distribution can be experimentally justified to obey Gaussian model, and the drift process is partly determined by its optimization function.To analyze the influence on PSO from the external environment, linear and quadratic optimization functions are chose to simplify the analysis. And the SCMD's stability is discussed via constructing the responding Lyapunov functions. We conclude that the gradient of linear function only decides the uniform ultimate bound while the gradient of quadratic function affects the stable conditions besides the uniform ultimate bound. Furthermore, the stable point of PSO is proved to be the optimum of quadratic function. The convergence time is also discussed to measure PSO's performance. (C) 2021 Elsevier Inc. All rights reserved.
查看更多>>摘要：Understanding collaborative and sequential information plays a major role in next-item recommendation. Although previous models have achieved considerable results, their capabilities have been limited by the basic paradigm of characterizing each user's profile with only his/her sequential behaviors. This paper considers collaborative information over users to solve the problem. Specifically, we first reorganize all interactions as a multiplex directed heterogeneous graph (named sequential graph) to depict each user's sequential behaviors and all users' collaborative interactions. Second, we propose a generic vectorization algorithm to address the challenge of multiplex edges. The algorithm can implement arbitrary graph attention networks on complex graphs without graph simplification. Finally, we propose a novel sequential graph attention network (SGAT) that controls information to attentively propagate through directed heterogeneous edges. Based on the sequential graph and the vectorization algorithm, SGAT infers each user's dynamic preference by capturing high-order sequential information among items and exploiting both latitudinally and longitudinally collaborative information over users. Extensive experiments and ablation studies on various datasets demonstrate the superiority of SGAT. The sequential graph provides an alternative way to organize users' sequential behaviors and the vectorization algorithm can promote the development of graph mining by retaining topological structures of complex graphs. (C) 2022 Elsevier Inc. All rights reserved.
查看更多>>摘要：Most existing subspace-based multiview learning algorithms treat multiview features equally in the process of subspace reconstruction without distinguishing the discrepant contributions from different views. This limitation adversely affects subspace embedding, particularly when significant variances in view-specific feature characteristics are present. In this paper, a unified subspace-based self-weighted multiview learning framework (SSMVL) is proposed for instance retrieval. In contrast to the previous methods in which the variances in multiview features are ignored, the self-weighted learning mechanism is integrated into the multiview subspace learning framework such that the weights of different views are adaptively learned instead of empirically assigned. In addition, the proposed SSMVL approach falls into the unsupervised learning category and is thus independent of massive amounts of labeled data resulting from labor-intensive annotation. In this study, the extension of SSMVL to the Hamming subspace learning paradigm is also explored for efficient retrieval. Experiments on five public benchmarks reveal that the self-weighted learning strategy plays a beneficial role in multiview fusion, and our method achieves superior performance in comparison to state-of-the-art methods. (C) 2022 Elsevier Inc. All rights reserved.
查看更多>>摘要：The optimal selection, ranking and classification of objects are important aspects in the multi-criteria decision-making (MCDM) problem. In this paper, we try to propose a three-way MCDM method with fuzzy complementary preference relation based on additive consistency to select the optimal object, rank objects and classify objects at the same time, which will help decision-makers better make decisions. Firstly, we give the fuzzy complementary judgment matrix based on additive consistency. Then, based on this matrix, we construct the concept of -similarity class, and define a new relative loss function. Subsequently, we describe the detailed decision-making processes of the newly proposed three-way MCDM method. Further, through the trainer selection problem, we verify the feasibility of the method. Comparative analyses show that our method has stronger decision-making function than some existing methods. Experimental analyses show that our method is stable in ranking the considered objects and selecting the optimal object. Moreover, the proposed method can not only provide reasonable ranking decision suggestion for decision-makers, but also can meet the decision-makers' preference in the classification decision. (C) 2022 Elsevier Inc. All rights reserved.
查看更多>>摘要：Missing values are a common phenomenon in modern medical research of complex diseases. The data often contains nominal or categorical variables, for example, single nucleotide polymorphisms (SNPs) in genetic studies. If the missing values are not handled properly, the downstream statistical analysis of incomplete data may be biased. While various imputation methods are available for metrically scaled variables, methods for categorical data are scarce. An imputation method that has been shown to work well for high dimensional metrically scaled variables is the imputation by nearest neighbor methods. In this paper, we propose a weighted nearest neighbors approach to impute missing values in categorical variables in high dimensional datasets. The proposed method explicitly uses the information on the association among attributes. Using different simulation settings, the performance is compared with available imputation methods. A variety of real data sets, containing heart, DNA, and lymphatic cancer, is also used to support the results obtained by simulations. The results show that the weighting of attributes yields smaller imputation errors than existing approaches like random forest and MICE. (C) 2022 Elsevier Inc. All rights reserved.
NSTL
Elsevier