Narrow Results

Compared to metallic materials, composites exhibit greater uncertainty, with some input variables characterized by limited and sparse data. Traditional nonprobabilistic methods provide only a qualitative range for uncertainty parameters, which is insufficient for quantifying credibility effectively and consequently increases the conservativeness of structural reliability analysis. To quantify uncertainty parameters in composites, a nonprobabilistic interval approach is employed to address the challenges posed by various failure modes and limited sample sizes. Based on Bayesian theory, a truncated normal distribution is utilized as the prior for the interval radius, with new sample points introduced to update this radius. The MCMC method is employed to numerically solve for the posterior distribution, thereby establishing a connection between the nonprobabilistic quantification sets and the credible levels. A nonprobabilistic credible Bayesian reliability model for composites is developed by integrating the Hashin failure criterion with Taylor series expansion for uncertainty propagation analysis. The effectiveness and feasibility of the proposed model are demonstrated through two relevant numerical examples.

Pornographic images are disturbing and malicious contents that are easily available through Internet technology. It has a negative and lasting effect on children who use the Internet; thus, pornography has become a serious threat not only to Internet users but also to society at large. Therefore, developing efficient and reliable tools to automatically filter pornographic contents is imperative. However, the effective interception of pornography remains a challenging issue. In this paper, a four-phase anti-pornography system based on the neural and Bayesian methods of artificial intelligence is proposed. Primitive information on pornography is examined and then used to determine if a given image falls under the pornography category. First, we present a detailed description of preliminary study phase followed by the modeling phase for the proposed skin detector. An anti-pornography system is created in the development phase, which also includes the proposed pornography classifier based on skin detection. Finally, the performance assessment method for the proposed anti-pornography system is discussed in the evaluation phase.

This study proposed a pornography classifier using multi-agent learning as a combination of the Bayesian method using color features extracted from skin detection based on the YCbCr color space and the back-propagation neural network method using shape features also extracted from skin detection. The classification of pornographic images was made more robust to the variation of images despite size engineering problems. Previous studies failed to achieve such robustness. Findings showed that the proposed multi-agent learning-based pornography classifier has produced significant TP and TN average rates (i.e., 96% and 97.33%, respectively). In addition, the proposed classifier has achieved a significantly low average rate of FN and FP (i.e., only 4% and 2.67%, respectively). The implementation of this algorithm is crucial and significant not only in identifying pornography but also in blocking Web sites that covertly promote pornography.

Modal identification aims at identifying the dynamic properties including natural frequency, damping ratio, and mode shape, which is an important step in further structural damage detection, finite element model updating, and condition assessment. This paper presents the work on the investigation of the dynamic characteristics of a long-span cable-stayed bridge-Sutong Bridge by a Bayesian modal identification method. Sutong Bridge is the second longest cable-stayed bridge in the world, situated on the Yangtze River in Jiangsu Province, China, with a total length of 2 088m. A short-term nondestructive on-site vibration test was conducted to collect the structural response and determine the actual dynamic characteristics of the bridge before it was opened to traffic. Due to the limited number of sensors, multiple setups were designed to complete the whole measurement. Based on the data collected in the field tests, modal parameters were identified by a fast Bayesian FFT method. The first three modes in both vertical and transverse directions were identified and studied. In order to obtain modal parameter variation with temperature and vibration levels, long-term tests have also been performed in different seasons. The variation of natural frequency and damping ratios with temperature and vibration level were investigated. The future distribution of the modal parameters was also predicted using these data.

The most commonly used method for sampling damage parameters from the posterior distribution is the Markov chain Monte Carlo (MCMC) method. The population MCMC method as one of the MCMC methods has been utilized for damage identification by some researchers recently. Nevertheless, for the conventional population MCMC methods, these sampling methods often require significant computational resources and tuning of a large number of algorithm parameters. Aiming at the problem of difficulty in selecting the proposal distribution and low computational efficiency in the conventional MCMC method, this paper proposed a simple population Metropolis–Hastings (SP-MH) algorithm for the damage identification, which is realized by exchanging information among chains in a relatively small population and using tuning-free strategy. Then, a numerical cantilever beam and an experimental frame are utilized to verify the effectiveness and feasibility of the proposed algorithm, it can be seen that the convergence rate of the SP-MH algorithm is faster than that of the Differential Evolution Monte Carlo (DE-MC) algorithm, and in a small population state, the SP-MH algorithm can still maintain convergence, saving plenty of computing time for damage identification. The results show that the SP-MH algorithm is feasible and accurate in practice damage identification, and the SP-MH algorithm performs better than the DE-MC algorithm. Compared with the DE-MC algorithm, the SP-MH algorithm is simple and convenient for damage identification due to its tuning-free strategy and relatively small population.

Chile presents a high seismicity and very particular characteristics due to its subductive environment and high convergence rate of the interacting plates. In this paper, ground motion models (GMM) for Arias Intensity and strong ground motion duration are developed based on recently compiled Chilean earthquakes records. The Chilean derived GMM are later compared with global and subduction models. For the duration case, two definitions are considered: significant duration and bracketed duration. Care is taken to consider newly define rupture areas for the major earthquake in the catalog, which allows to redefine the distances to be used in the analysis. The strong motion records are processed using robust techniques and standardized procedures. The model parameters are selected using a Bayesian updating method. This allows to estimate the parameters as well as their corresponding uncertainties recognizing the variability of the coefficients and of the selected model. The derived ground motion predictive model associated is validated through a residual analysis and probability distributions, obtained from the adjustment method. Additionally, for the bracketed duration analysis, a classification method based on Support Vector Machine (SVM) is used, with the aim to incorporate the records with duration of zero seconds. The comparison of the resultant model with the existing ones allows to confirm the differences that the tectonic Chilean environment produces, characterized mainly by a high relative convergence rate between plates and interplate thrust earthquakes with high magnitudes.

Some methods for constructing solutions of cluster analysis problems based on processing a set of solutions for the cluster analysis problem by a set of heuristic clustering algorithms, as well as regression analysis problems using a set of different supervised classification algorithms, are considered. The solution to the problem of cluster analysis is defined as the construction of a set of equivalent information matrices. Two information matrices are called equivalent to each other if they are equaled up to a permutation of the columns. The construction of the optimal committee solution of the problem of cluster analysis by a team of algorithms is reduced to solving one discrete optimization problem on permutations. An upper bound for the variation of the basic functional is obtained. An effective (polynomial) algorithm is proposed for its minimization. Heuristic methods for constructing collective clustering are considered.

The problem of restoring a function (regression recovery) as an application of some function (corrector) to the set of solutions of special supervised classification problems is considered. A preset training sample of feature descriptions of objects is assumed. In this case, all the problems associated with the heterogeneity of the features, their different informative or incomplete sampling, are failed on the classification problems. Based on this sample and the values of the dependent variable, a random collective of classification tasks is constructed. For a recognizable object, the class is computed, to which it belongs according to each algorithm. Then the corrector found earlier is applied to the class numbers. Two types of correctors are considered: the Bayesian and the average linear corrector. In the Bayesian corrector, the naive Bayesian method is used. The average linear corrector consists of calculating a linear combination of the mean values of the dependent value from the calculated values of the classes.

Clustering analysis is a significant technique of data mining. With the rise of lifelong learning, lifelong clustering has become a research topic. Lifelong clustering builds libraries shared among multiple tasks, and these tasks achieve effective information transmission by interacting with the shared knowledge libraries. However, selecting optimal hyper-parameters in the knowledge transfer process often employs the actual clustering division in the dataset as a reference, which is unavailable during the clustering process. Moreover, the hyper-parameters for each task are typically set to constant values because of computational difficulty. Therefore, this paper explores a clustering method based on Bayesian inference, where the parameter setting is priori information, and the clustering divisions obtained by the parameters are posteriori information. In our method, hyper-parameters corresponding to the maximum a posteriori (MAP) probability are selected in each task. Then, we apply this method to Lifelong Spectral Clustering to select hyper-parameters and propose a new algorithm, called Maximum a Posteriori Lifelong Spectral Clustering (MAPLSC). Finally, experiments on several real-world datasets show the effectiveness of our method and the average clustering performance of Lifelong Spectral Clustering is improved.

In dealing with small sample size problem, such as face recognition from images, Null space based LDA and Bayesian method are two effective algorithms. In this paper, we analysis the two algorithms mentioned above from the view of Weighted Projection Approach, and show that they both can be viewed as special cases of Weighted Projection Approach. Then a Parameterized Weighted Projection Approach is proposed. It can combine the advantage of those two algorithms analyzed above and also exhibit good flexibility. Experiments show that the new algorithm is more flexible and can reach better performance.