Multidimensional Connection Cloud Model Coupled with Improved CRITIC Method for Evaluation of Eutrophic Water

The degree of eutrophication in the water environment is deepening. For the appropriate treatment of eutrophication, it is essential to evaluate it accurately. However, the evaluation of eutrophication has not been well solved because it is full of uncertainty. Herein, a multidimensional connection cloud model, combined with the improved CRITIC (Criteria Importance Through Inter-criteria Correlation) method, was put forward here to assess water eutrophication and depict the randomness, ambiguity, and interaction of evaluation factors. First, an improved CRITIC was adopted to determine indicator weight so that the correlation among different indicators and more information were depicted. Secondly, a multidimensional connection cloud was simulated to characterize fuzzy indicators and ambiguous classification boundary values according to classification criteria. Next, the connection degree was calculated relative to the evaluation standard. The eutrophication grade was specified under the “maximum connection degree” principle. At last, the effectiveness and practicality of the model proposed here were affirmed by two cases and comparisons with supplementary methods. The results suggest that the proposed model can avoid shortcomings of the original CRITIC method and cloud model, and make the assessment result more realistic.

Structural Damage Detection Using Supervised Nonlinear Support Vector Machine

Damage detection, using vibrational properties, such as eigenfrequencies, is an efficient and straightforward method for detecting damage in structures, components, and machines. The method, however, is very inefficient when the values of the natural frequencies of damaged and undamaged specimens exhibit slight differences. This is particularly the case with lightweight structures, such as fiber-reinforced composites. The nonlinear support vector machine (SVM) provides enhanced results under such conditions by transforming the original features into a new space or applying a kernel trick. In this work, the natural frequencies of damaged and undamaged components are used for classification, employing the nonlinear SVM. The proposed methodology assumes that the frequencies are identified sequentially from an experimental modal analysis; for the study propose, however, the training data are generated from the FEM simulations for damaged and undamaged samples. It is shown that nonlinear SVM using kernel function yields in a clear classification boundary between damaged and undamaged specimens, even for minor variations in natural frequencies.

Computer Vision-Based Patched and Unpatched Pothole Classification Using Machine Learning Approach Optimized by Forensic-Based Investigation Metaheuristic

During the phase of periodic asphalt pavement survey, patched and unpatched potholes need to be accurately detected. This study proposes and verifies a computer vision-based approach for automatically distinguishing patched and unpatched potholes. Using two-dimensional images, patched and unpatched potholes may have similar shapes. Therefore, this study relies on image texture descriptors to delineate these two objects of interest. The texture descriptors of statistical measurement of color channels, the gray-level cooccurrence matrix, and the local ternary pattern are used to extract texture information from image samples of asphalt pavement roads. To construct a classification model based on the extracted texture-based dataset, this study proposes and validates an integration of the Support Vector Machine Classification (SVC) and the Forensic-Based Investigation (FBI) metaheuristic. The SVC is used to generalize a classification boundary that separates the input data into two class labels of patched and unpatched potholes. To optimize the SVC performance, the FBI algorithm is utilized to fine-tune the SVC hyperparameters. To establish the hybrid FBI-SVC framework, an image dataset consisting of 600 samples has been collected. The experiment supported by the Wilcoxon signed-rank test demonstrates that the proposed computer vision is highly suitable for the task of interest with a classification accuracy rate = 94.833%.

Source-free Domain Adaptation via Avatar Prototype Generation and Adaptation

We study a practical domain adaptation task, called source-free unsupervised domain adaptation (UDA) problem, in which we cannot access source domain data due to data privacy issues but only a pre-trained source model and unlabeled target data are available. This task, however, is very difficult due to one key challenge: the lack of source data and target domain labels makes model adaptation very challenging. To address this, we propose to mine the hidden knowledge in the source model and exploit it to generate source avatar prototypes (i.e. representative features for each source class) as well as target pseudo labels for domain alignment. To this end, we propose a Contrastive Prototype Generation and Adaptation (CPGA) method. Specifically, CPGA consists of two stages: (1) prototype generation: by exploring the classification boundary information of the source model, we train a prototype generator to generate avatar prototypes via contrastive learning. (2) prototype adaptation: based on the generated source prototypes and target pseudo labels, we develop a new robust contrastive prototype adaptation strategy to align each pseudo-labeled target data to the corresponding source prototypes. Extensive experiments on three UDA benchmark datasets demonstrate the effectiveness and superiority of the proposed method.

CF-CNN: Coarse-to-Fine Convolutional Neural Network

In this paper, we present a coarse-to-fine convolutional neural network (CF-CNN) for learning multilabel classes. The basis of the proposed CF-CNN is a disjoint grouping method that first creates a class group with hierarchical association, and then assigns a new label to a class belonging to each group so that each class acquires multiple labels. CF-CNN consists of one main network and two subnetworks. Each subnetwork performs coarse prediction using the group labels created by the disjoint grouping method. The main network includes a refine convolution layer and performs fine prediction to fuse the feature maps acquired from the subnetwork. The generated class set in the upper level has the same classification boundary to that in the lower level. Since the classes belonging to the upper level label are classified with a higher priority, parameter optimization becomes easier. In experimental results, the proposed method is applied to various classification tasks to show a higher classification accuracy by up to 3% with a much smaller number of parameters without modification of the baseline model.

Robust Ordinal Regression: User Credit Grading with Triplet Loss-Based Sampling

With the development of social media sites, user credit grading, which served as an important and fashionable problem, has attracted substantial attention from a slew of developers and operators of mobile applications. In particular, multi-grades of user credit aimed to achieve (1) anomaly detection and risk early warning and (2) personalized information and service recommendation for privileged users. The above two goals still remained as up-to-date challenges. To these ends, in this article, we propose a novel regression-based method. Technically speaking, we define three natural ordered categories including BlockList , GeneralList , and AllowList according to users’ registration and behavior information, which preserve both the global hierarchical relationship of user credit and the local coincident features of users, and hence formulate user credit grading as the ordinal regression problem. Our method is inspired by KDLOR ( kernel discriminant learning for ordinal regression ), which is an effective and efficient model to solve ordinal regression by mapping high-dimension samples to the discriminant region with supervised conditions. However, the performance of KDLOR is fragile to the extreme imbalanced distribution of users. To address this problem, we propose a robust sampling model to balance distribution and avoid overfit or underfit learning, which induces the triplet metric constraint to obtain hard negative samples that well represent the latent ordered class information. A step further, another salient problem lies in ambiguous samples that are noises or located in the classification boundary to impede optimized mapping and embedding. To this problem, we improve sampling by identifying and evading noises in triplets to obtain hard negative samples to enhance robustness and effectiveness for ordinal regression. We organized training and testing datasets for user credit grading by selecting limited items from real-life huge tables of users in the mobile application, which are used in similar problems; moreover, we theoretically and empirically demonstrate the advantages of the proposed model over established datasets.

Local Similarity-Based Fuzzy Multiple Kernel One-Class Support Vector Machine

One-class support vector machine (OCSVM) is one of the most popular algorithms in the one-class classification problem, but it has one obvious disadvantage: it is sensitive to noise. In order to solve this problem, the fuzzy membership degree is introduced into OCSVM, which makes the samples with different importance have different influences on the determination of classification hyperplane and enhances the robustness. In this paper, a new calculation method of membership degree is proposed and introduced into the fuzzy multiple kernel OCSVM (FMKOCSVM). The combined kernel is used to measure the local similarity between samples, and then, the importance of samples is determined based on the local similarity between training samples, so as to determine the membership degree and reduce the impact of noise. The proposed membership requires only positive data in the calculation process, which is consistent with the training set of OCSVM. In this method, the noise has a smaller membership value, which can reduce the negative impact of noise on the classification boundary. Simultaneously, this method of calculating membership has a higher efficiency. The experimental results show that FMKOCSVM based on proposed local similarity membership is efficient and more robust to outliers than the ordinary multiple kernel OCSVMs.

Partial Retraining Substitute Model for Query-Limited Black-Box Attacks

Black-box attacks against deep neural network (DNN) classifiers are receiving increasing attention because they represent a more practical approach in the real world than white box attacks. In black-box environments, adversaries have limited knowledge regarding the target model. This makes it difficult to estimate gradients for crafting adversarial examples, such that powerful white-box algorithms cannot be directly applied to black-box attacks. Therefore, a well-known black-box attack strategy creates local DNNs, called substitute models, to emulate the target model. The adversaries then craft adversarial examples using the substitute models instead of the unknown target model. The substitute models repeat the query process and are trained by observing labels from the target model’s responses to queries. However, emulating a target model usually requires numerous queries because new DNNs are trained from the beginning. In this study, we propose a new training method for substitute models to minimize the number of queries. We consider the number of queries as an important factor for practical black-box attacks because real-world systems often restrict queries for security and financial purposes. To decrease the number of queries, the proposed method does not emulate the entire target model and only adjusts the partial classification boundary based on a current attack. Furthermore, it does not use queries in the pre-training phase and creates queries only in the retraining phase. The experimental results indicate that the proposed method is effective in terms of the number of queries and attack success ratio against MNIST, VGGFace2, and ImageNet classifiers in query-limited black-box environments. Further, we demonstrate a black-box attack against a commercial classifier, Google AutoML Vision.

Evidence integration and decision-confidence are modulated by stimulus consistency

Evidence-integration is a normative algorithm for choosing between alternatives with noisy evidence, which has been successful in accounting for a vast amount of behavioral and neural data. However, this mechanism has been challenged as tracking integration boundaries sub-serving choice has proven elusive. Here we first show that the decision boundary can be monitored using a novel, model-free behavioral method, termed Decision-Classification Boundary. This method allowed us to both provide direct support for evidence-integration contributions and to identify a novel integration-bias, whereby incoming evidence is modulated based on its consistency with evidence from preceding time-frames. This consistency bias was supported in three cross-domain experiments, involving decisions with perceptual and numerical evidence, which showed that choice-accuracy and decision confidence are modulated by stimulus consistency. Strikingly, despite its seeming sub-optimality, this bias fosters performance by enhancing robustness to integration noise. We argue this bias constitutes a new form of micro-level, within-trial, confirmation bias and discuss implications to broad aspects of decision making.