Noisy Data Points Research Articles

Improvement for deep learning for suicide and depression identification with unsupervised label correction

Abstract. Our study try to address the challenge of accurately identifying depression and suicidal ideation on social media platforms by introducing an enhanced novel methodology for unsupervised feature selection and label correction. Utilizing advanced word embedding models like BERT and the Universal Sentence Encoder, we transform textual content into dense numerical vectors that capture the nuanced emotional context of online discussions. Our approach enhances these embeddings with a deep neural network (DNN) to extract distinctive features, reducing the dimensionality of the data through Principal Component Analysis (PCA). For label correction, we employ clustering techniques including OPTICS, K-medoids, and hierarchical clustering, which are robust against noisy data points. We then train classifiers using CNN, DNN, logistic regression, and random forest algorithms, evaluated with metrics such as accuracy, precision, recall, F1 score, and AUC. This methodology improves the accuracy of classifying depressive and suicidal sentiments to some extent, assisting to utilize the vast data available on social media to advance mental health diagnostics and interventions.

A sparse quantized hopfield network for online-continual memory

An important difference between brains and deep neural networks is the way they learn. Nervous systems learn online where a stream of noisy data points are presented in a non-independent, identically distributed way. Further, synaptic plasticity in the brain depends only on information local to synapses. Deep networks, on the other hand, typically use non-local learning algorithms and are trained in an offline, non-noisy, independent, identically distributed setting. Understanding how neural networks learn under the same constraints as the brain is an open problem for neuroscience and neuromorphic computing. A standard approach to this problem has yet to be established. In this paper, we propose that discrete graphical models that learn via an online maximum a posteriori learning algorithm could provide such an approach. We implement this kind of model in a neural network called the Sparse Quantized Hopfield Network. We show our model outperforms state-of-the-art neural networks on associative memory tasks, outperforms these networks in online, continual settings, learns efficiently with noisy inputs, and is better than baselines on an episodic memory task.

Enhancing class imbalance solutions: A projection-based fuzzy LS-TSVM approach

Class imbalance and noise present significant challenges in numerous real-world classification tasks. The prevalence of an uneven distribution of samples typically results in a bias towards the majority class in Support Vector Machine (SVM) classifiers, compounded by the often inherent noise within these samples. Addressing both class imbalance and noise, we introduce two fuzzy-based methodologies. The first method employs intuitionistic fuzzy membership, resulting in the development of the Robust Energy-based Intuitionistic Fuzzy Least Squares Twin Support Vector Machine (IF-RELSTSVM), a model specifically designed for class imbalance learning. The IF-RELSTSVM model is distinguished by its use of intuitionistic fuzzy scores for both classes, significantly attenuating the detrimental effects of noise and outliers. A distinctive attribute of IF-RELSTSVM is its proficiency in processing noisy data points, whether proximate to or distant from the hyperplane. Additionally, we introduce a novel concept of hyperplane-based fuzzy membership, calculating fuzzy memberships through a projection-based approach. This foundation supports the formulation of a Robust Energy-based Fuzzy Least Squares Twin Support Vector Machine (F-RELSTSVM), also aimed at class imbalance learning. The efficacy of the proposed IF-RELSTSVM and F-RELSTSVM algorithms is rigorously evaluated across several benchmark and synthetic datasets, employing the Area Under the ROC Curve (AUC) as a performance metric. Experimental findings indicate that these algorithms surpass baseline models in the majority of datasets tested. Statistical analyses further validate the significance of our proposed methods, demonstrating their suitability for application in environments characterized by noise and class imbalance. A case study in credit card fraud detection showcases the F-RELSTSVM algorithm achieving an impressive average AUC of 90.84%, thereby outperforming comparable algorithms and highlighting the practical applicability of our approaches in tackling challenging datasets.

The Predictive Power of Monetary Policy on International Stock Market Returns—Evidence From TV-ARMAX Model

In this paper, we apply the time-varying ARMA model with exogenous variable (TV-ARMAX) to examine the predictive power of monetary policy on international stock returns. This method allows time-varying coefficient estimates and uses time-dependent cumulated variation penalty to filter noisy outlier data points. Based on a wide range of 31 countries, our method robustly outperforms other popular methods including the simple linear-regression model (SLM), the vector autoregression and its variants (VAR, TV-VAR, and VARX) and the ARMA model with exogenous variable (ARMAX).

Prediction of ultimate strain and strength of CFRP-wrapped normal and high-strength concrete compressive members using ANN approach

The literature is deficit in predicting the axial strength (AS) and axial strain of carbon fiber reinforced polymer (CFRP)-wrapped normal strength concrete (NSC) and high strength concrete (HSC) compressive members using machine learning techniques. The already proposed models for predicting the AS and axial strain of CFRP-wrapped NSC compressive members were developed using a general regression analysis technique based on a small number of noisy data points by considering the limited parameters of specimens. Therefore, there is a need for a refined and accurate theoretical model for capturing the AS and axial strain of compressive members. The main objective of the current study is to develop the theoretical models for predicting the AS and axial strain of CFRP-wrapped NSC and HSC compressive members using machine learning methods. Two different analysis approaches are employed for securing the objective of the present study. The first approach is using the general regression analysis technique, and the second one is employing artificial neural networks (ANN) modeling. The testing database employed in the analysis consists of the testing results of 364 concrete compressive members subjected to compressive loading. The accuracy of the proposed empirical and ANN models is evaluated and compared on the basis of the testing results. Three statistical indices were employed to determine the performance of the currently proposed models developed in the present study. The ANN models presented the accuracy with the statistical indices R 2 = 0.984, RMSE = 0.112, and MAE = 0.097 for the strength model and R 2 = 0.942, RMSE = 1.211, and MAE = 0.978 for the strain model. The suggested empirical model presented the statistical indices R 2 = 0.90, RMSE = 0.33, and MAE = 2.45 for the strength model and R 2 = 0.80, RMSE = 2.05, and MAE = 5.34 for the strain model. The statistical evaluation showed that the ANN models are more effective and precise than the empirical ones in predicting the AS and axial strain of CFRP-wrapped circular concrete compressive members.

Prediction and uncertainty quantification of SAFARI-1 axial neutron flux profiles with neural networks

In this study, Deep Neural Networks (DNNs) are used to predict the assembly axial neutron flux profiles in the SAFARI-1 research reactor, with quantified uncertainties in the DNN predictions and testing using data from cycles that are unseen in the training process. The training dataset consists of copper-wire activation measurements from the reactor’s historical cycles. Uncertainty Quantification of the regular DNN models’ predictions is performed using Monte Carlo Dropout (MCD) and Bayesian Neural Networks solved by Variational Inference (BNN VI). The regular DNNs, DNNs solved with MCD and BNN VI results agree very well with each other and with the new measured dataset not used in the training process, thus indicating good prediction and generalization capability. The uncertainty bands produced by MCD and BNN VI agree very well and fully envelop the noisy measurement data points.

Deep Cleaner—A Few Shot Image Dataset Cleaner Using Supervised Contrastive Learning

Images are increasingly used for AI-based diagnosis and analysis of many diseases like cervical cancer, mouth cancer, glucose analysis from retina etc. In many cases, data collection is done by specialised camera modules which capture images of affected areas. As with any other sources of data, this process is also error-prone and may contain unwanted objects and regions that may require cleaning by removing them. Outliers in these kinds of dataset may adversely affect the performance of machine learning models. Manually cleaning would be a tedious task, especially when the data is collated from different sources. Hence, cleaning the data before training the model is of utmost importance. In this paper, we propose a Few-Shot learning based model pre-trained in supervised contrastive learning settings to automate the process of data cleaning. Our model learns the dataset distribution and distinguishes the accurate data points from noisy data points. We also show that scaling up the model can greatly improve the Few-Shot performance. On the noisy MobileODT cervical data, which was collected from Kaggle, our model obtained 52% accuracy without cleaning data using an EfficientNet architecture for the classification task. Whereas the same architecture with ROI cropping achieved an accuracy of 76.56% after cleaning through the proposed Deep Cleaner approach that requires only 100 clean images. The proposed approach performs 2.74% better than a denoising auto-encoder, which is considered a powerful anomaly detection technique.

Multi Feature DCR based Drug Compound Selection and Recommendation System for Efficient Decision-Making using Genetic Algorithm

The performance of treating the cardiac diseases is dependent on the kind of drug being selected. There exist numerous decisive support systems which work according to certain characteristics and factors like drug availability, and popularity. Still, they struggle to achieve expected performance in supporting the medical practitioner. To handle this issue, a multi feature drug curing rate based drug compound selection and recommendation system (MDCRSR) is presented. The method utilizes medical histories and data set of various medical organization around the disease considered. Using the traces, the method identifies the drug compounds and features to perform preprocessing which eliminates the noisy data points. Further, the features of the traces are extracted to perform training with genetic algorithm. At the test phase, the method estimates the fitness measure for different drug combination and compounds by measuring their Drug Curing Rate (DCR). The method performs cross over and mutation to produce various populations of drug compounds. According to the curing rate, the drug compound pattern or population is selected and ranked. The ranked results are populated to the medical practitioner. The method improves the performance of recommendation system as well as drug compound selection.

Quick remedy commits and their impact on mining software repositories

Most changes during software maintenance and evolution are not atomic changes, but rather the result of several related changes affecting different parts of the code. It may happen that developers omit needed changes, thus leaving a task partially unfinished, introducing technical debt or injecting bugs. We present a study investigating “quick remedy commits” performed by developers to implement changes omitted in previous commits. With quick remedy commits we refer to commits that (i) quickly follow a commit performed by the same developer, and (ii) aim at remedying issues introduced as the result of code changes omitted in the previous commit (e.g., fix references to code components that have been broken as a consequence of a rename refactoring) or simply improve the previously committed change (e.g., improve the name of a newly introduced variable). Through a manual analysis of 500 quick remedy commits, we define a taxonomy categorizing the types of changes that developers tend to omit. The taxonomy can (i) guide the development of tools aimed at detecting omitted changes and (ii) help researchers in identifying corner cases that must be properly handled. For example, one of the categories in our taxonomy groups the reverted commits, meaning changes that are undone in a subsequent commit. We show that not accounting for such commits when mining software repositories can undermine one’s findings. In particular, our results show that considering completely reverted commits when mining software repositories accounts, on average, for 0.07 and 0.27 noisy data points when dealing with two typical MSR data collection tasks (i.e., bug-fixing commits identification and refactoring operations mining, respectively).

Semi-Supervised Ensemble Learning for Dealing with Inaccurate and Incomplete Supervision

In real-world tasks, obtaining a large set of noise-free data can be prohibitively expensive. Therefore, recent research tries to enable machine learning to work with weakly supervised datasets, such as inaccurate or incomplete data. However, the previous literature treats each type of weak supervision individually, although, in most cases, different types of weak supervision tend to occur simultaneously. Therefore, in this article, we present Smart MEnDR, a Classification Model that applies Ensemble Learning and Data-driven Rectification to deal with inaccurate and incomplete supervised datasets. The model first applies a preliminary phase of ensemble learning in which the noisy data points are detected while exploiting the unlabelled data. The phase employs a semi-supervised technique with maximum likelihood estimation to decide on the disagreement rate. Second, the proposed approach applies an iterative meta-learning step to tackle the problem of knowing which points should be made correct to improve the performance of the final classifier. To evaluate the proposed framework, we report the classification performance, noise detection, and the labelling accuracy of the proposed method against state-of-the-art techniques. The experimental results demonstrate the effectiveness of the proposed framework in detecting noise, providing correct labels, and attaining high classification performance.

Effect of Micro-Patterned Membrane PEFC Dependent on Operating Current Density

1. IntroductionPolymer electrolyte fuel cells (PEFCs) have been widely recognized as a key technology for a future hydrogen-based society, because of their several practical applications such as in fuel cell vehicles, owing to their good operating capability at low temperatures.It is well known that the mass transport and the chemical reactions at the cathode side of PEFCs limit their performance. Heinzmann et al. used electrochemical impedance spectroscopy (EIS) and distribution of relaxation time (DRT) analysis [1] to show that the dominant processes that occur are cathode gas diffusion and oxygen reduction reaction (ORR).One method to improve the ORR is to increase the surface available for the chemical reaction. Jeon et al. fabricated a micropatterned membrane and applied it to a membrane electrode assembly (MEA) [2]. They confirmed that the use of a micropatterned membrane improved the performance of the PEFC.However, the mechanism of this improvement is yet to be clarified in detail. Micropatterns on the membrane should increase the active reaction area (Fig.1(a)). However, patterns on the membrane surface increase the length of the path from the top to the bottom of the catalyst layer, which should hamper oxygen diffusion (Fig.1(b)). We previously performed DRT analysis with micro-patterned MEA to investigate the mechanism of improvement in PEFC performance using a micro-patterned MEA [3]. In this study, a wider range of EIS was performed, and both the positive and negative effects of using a micro-patterned MEA are discussed herein.2. ExperimentA cathode micro-patterned MEA was fabricated, and EIS and DRT analyses were performed. The catalyst ink was prepared using Pt/C (TEC10V50E, Tanaka Kikinzoku Kogyo K.K.), isopropyl alcohol, water, and a 10% ionomer solution. This ink was sprayed onto a 5.0 mm line-patterned membrane (NRE-212, Sigma-Aldrich Co. LLC) to fabricate the MEA. The catalyst loading was adjusted to 0.4 mg/cm2. The cell was assembled using a commercial gas diffusion layer (Sigracet®︎ 29BC, SGL Carbon GmbH). The operation was carried out galvanostatically with a relative humidity of 80%, gas flow rate of 200 ccm, and oxygen partial pressure of 0.2. EIS was measured at 0.02 to 1.00 A, with frequency ranging between 0.1 Hz and 1 MHz. The acquired data were then subjected to DRT analysis using DRT tools by Wan et al. [4]. Noisy data points were excluded before application because of the high sensitivity of DRT analysis to noise.3.Results and discussionFig.2 shows the polarization curve results. Solid lines with circle markers and broken lines with ‘x’ markers indicate the patterned and flat samples, respectively. In the lower current region, where the ORR is dominant, the patterned sample outperforms the conventional flat sample. On the other hand, in the higher current region, where gas diffusion is dominant, the patterned sample suffers steep drops and the flat becomes relatively better.Fig.3 shows the DRT analysis graph. Fig.3(a) stands for patterned sample result and Fig3.(b) for flat sample. The peaks of DRT indicate the polarization resistance at its time dependency. In the high current density region, the peaks of the patterned and the flat samples are significantly different. In particular, at around 101 to102 Hz, an independent peak appears at 101 Hz for the patterned sample as the current density increases. This shows the polarization resistance of gas diffusion, which indicates that the gas diffusion in the patterned sample is significantly worse than that in the flat sample.The micro-patterned MEA has a negative effect on cell performance because of the longer oxygen diffusion path involved. It can be assumed that they might not perform as well as conventional flat MEA under certain conditions that favor good oxygen transportation, such as high current or low oxygen partial pressure operation. Therefore, it is important to consider the physical properties (catalyst layer porosity, ionomer content, etc.) and operating conditions in order to maximize the potential of the micro-patterned MEA.4. ConclusionWe performed a wide range of galvanostatic EIS and DRT analyses of the data from the operation of a cathode micro-patterned PEFC. The micro-patterned membrane is shown to have both a positive and a negative effect. Comprehensive optimization of parameters such as micropattern geometry, catalyst composition, and operating conditions, are required to take full advantage of the surface-patterning method.

A New Validity Index for Fuzzy-Possibilistic C-Means Clustering

In some complicated datasets, due to the presence of noisy data points and outliers, cluster validity indices can give conflicting results in determining the optimal number of clusters. This paper presents a new validity index for fuzzy-possibilistic c-means clustering called Fuzzy-Possibilistic)FP (index, which works well in the presence of clusters that vary in shape and density. Moreover, FPCM like most of the clustering algorithms is susceptible to some initial parameters. In this regard, in addition to the number of clusters, FPCM requires a priori selection of the degree of fuzziness (m) and the degree of typicality (η). Therefore, we presented an efficient procedure for determining an optimal value for and . The proposed approach has been evaluated using several synthetic and real-world datasets. Final computational results demonstrate the capabilities and reliability of the proposed approach compared with several well-known fuzzy validity indices in the literature. Furthermore, to clarify the ability of the proposed method in real applications, the proposed method is implemented in microarray gene expression data clustering and medical image segmentation.

Airfoil profile reconstruction from unorganized noisy point cloud data

Abstract Airfoil blades are typically inspected in sections to verify their conformance to the geometric tolerances specified on their nominal design. To maintain the accuracy of geometric error evaluation, in particular, for the position and orientation errors of the airfoil sections, sectional airfoil profiles should be reconstructed from the inspection data points. This paper presents a new method to automatically reconstruct the airfoil profile from unorganized noisy sectional data points of 3D scanned blades. A three-step airfoil profile reconstruction approach is presented. First, the algorithm thins the scattered set of sectional data points by projecting them onto the local curves fitted to them. For this purpose, a recursive weighted local least-squares scheme is proposed to fit local curves within the measurement uncertainty constraint of inspection data. Then, to order the thinned set of data points, the profile polygon is generated and imperfect nodes are modified by evaluation of the angular deviation of edges. Finally, a closed nonperiodic B-spline curve is fitted to the thinned and ordered set of data points to construct the smooth airfoil profile. A series of case studies have been carried out to demonstrate the effectiveness of the proposed airfoil profile reconstruction method. Implementation results have demonstrated that the proposed method is accurate and robust to noise. In addition to blade inspection, other applications such as repair and adaptive machining of aero-engine blades can equally benefit from the proposed method for automatic airfoil profile reconstruction.

Weighted Multiview Possibilistic C-Means Clustering With L2 Regularization

Since social media, virtual communities and networks rapidly grow, multiview data become more popular. In general, multiview data always contain different feature components in different views. Although these data are extracted in different ways (views) from diverse settings and domains, they are used to describe the same samples, which make them highly related. Hence, applying (single-view) clustering methods for multiview data poses difficulty in achieving desirable clustering results. Thus, multiview clustering methods should be developed that will utilize available multiview information. Most of multiview clustering techniques currently use k-means due to its conceptual simplicity, and use fuzzy c-means (FCM) that the datapoints can belong to more than one cluster based on their membership degrees from 0 to 1. However, the use of k-means or FCM may degrade its performance due to the presence of noise and outliers, especially on large or high-dimensional datasets. The constraint imposed on the membership degrees of k-means and FCM tends to assign a corresponding high membership value to an outlier or a noisy data point. To address these drawbacks, possibilistic c-means (PCM) relaxes the membership constraint of k-means and FCM so that outliers and noisy datapoints can be properly identified. On the other hand, there are various extensions of k-means and FCM for multiview data, but no extension of PCM for multiview data was made in the literature. Thus, we use PCM in our proposed multiview clustering model. In this article, we propose novel weighted multiview PCM algorithms designed for clustering multiview data as well as view and feature weights on PCM approaches, called W-MV-PCM and W-MV-PCM with L2 regularization (W-MV-PCM-L2). In multiview clustering, different views may vary with respect to its importance and each view may contain some irrelevant features. In the proposed algorithms, a learning scheme is constructed to compute for the view weights, and feature weights within each view. This scheme will be able to identify the importance of each view and, at the same time, it will also identify and select relevant features in each view. Comparisons of W-MV-PCM-L2 with existing multiview clustering algorithms are made on both synthetic and real datasets. The experimental results are evaluated using accuracy rate (AR) and external validity indexes, such as Rand index (RI) and normalized mutual information (NMI). The proposed W-MV-PCM-L2 algorithm with comparisons of existing algorithms under criteria of AR, RI, and NMI shows that it is a feasible and effective multiview clustering algorithm.

Distribution of Relaxation Time Analysis of Cathode Micro-Patterned PEFC

1. Introduction Fuel cells have been attracting worldwide attention because they are a key technology for producing electric energy for a future hydrogen-based society. In particular, polymer electrolyte fuel cells (PEFCs) are important among other fuel cells. This is because they can be operated at relatively low temperatures, which means they have many practical applications, e.g., fuel cell vehicles. However, PEFCs still have some problems. One is their performance because they are not as efficient as other forms of fuel cells. There are various approaches for achieving this, but it is common theory that the cathode chemical process limits the performance of PEFCs. Heinzmann et al. applied electrochemical impedance spectroscopy (EIS) and distribution of relaxation time (DRT) analysis to break down polarization resistance in detail [1]. They clearly showed that the dominant process is the cathode gas diffusion and oxygen reduction reaction (ORR). One way to improve the ORR is to increase the chemical reaction surface. Jeon et al. fabricated a micro-patterned membrane and applied it to membrane electrode assembly (MEA) [2]. As a result, they confirmed that the micro-patterned membrane can improve the performance of the entire PEFC. However, the physical phenomenon of its improvement has yet to be clarified. In detail, these patterns can certainly increase the surface area but may decrease mass transportation, especially gas diffusion. In this study, we performed DRT analysis with micro-patterned MEA to investigate the mechanism of performance improvement from using micro-patterned MEA (Fig. 1).2.Experiment A cathode micro-patterned MEA was fabricated, and then EIS and DRT analysis were performed. The patterned membrane was produced by a thermal nanoimprinting process. A Ni mold with a 5.0-mm line pattern was prepared, and the membrane was pressed onto the mold. Through this process, the pattern of the Ni mold was replicated to a Nafion membrane (NRE-212, Sigma-Aldrich Co. LLC) with a thickness of 50 mm. Then, the catalyst ink was prepared with Pt/C (TEC10V50E, Tanaka Kikinzoku Kogyo K.K.), isopropyl alcohol, water, and a 10% ionomer solution. We sprayed this ink onto the patterned membrane to fabricate the MEA. The catalyst loading was approximately 0.4 mg/cm2. Using a commercial gas diffusion layer (Sigracet®︎ 29BC, SGL carbon GmbH), the cell was assembled, and EIS was operated at 0.1 Hz to 1 MHz and at 0.7, 0.65, and 0.60 V. The acquired data were then applied to DRT analysis using the DRT tools by Wan et al. [3]. Noisy data points were excluded before application because the DRT analysis is very sensitive to noise.3.Results and discussion Figure 2 shows a cross-sectional SEM image of the micro-patterned MEA. Clearly, the membrane is patterned via the designed geometry, and the catalyst layer is fully integrated into the micro-pattern. Figure 3 shows the EIS results. The polarization resistance decreased as the potential was decreased, similar to the general behavior of conventional fuel cells. Figure 4 shows the DRT analysis. Two clear peaks are observable at approximately 10 Hz and 100 Hz, which we call P1 and P2, respectively. P1 remained the same or became slightly larger with decreasing potential. However, P2 decreased as the potential decreased. In addition, P2 of 0.60 V indicated that this peak split into two peaks as the potential decreased. P2 appeared as a combined peak of gas diffusion and the oxygen reduction reaction (ORR) upon comparing it to previous DRT analysis research of commercial cells [1]. We could determine which peak represents what physical phenomena if we perform additional DRT analysis while changing the operating environment, as in previous studies. By comparing the analysis of micro-patterned samples with conventional sample, it will be possible to evaluate the effect of the micro-pattern and its physical phenomena.4.Conclusion We performed DRT analysis with a cathode micro-patterned PEFC. DRT analysis has the ability to clarify the physical phenomena of how a patterned MEA increases cell performance with patterned membranes. Our results indicated that there is a peak at approximately 100 Hz, and it splits into two peaks representing gas diffusion and ORR with a lower potential, such as 0.60 V. As mentioned earlier, further work is required to assign physical phenomena to each peak. A comparison of DRT data using a conventional sample can possibly reveal the mechanism of performance increase owing to membrane patterning.

Spectral embedded generalized mean based [formula omitted]-nearest neighbors clustering with S-distance

The spectral clustering algorithm is extensively employed in different aspects, especially in the field of pattern recognition. However, the efficient construction of the neighborhood graph is the main reason for its promising results. Generally, the similarity matrix relies on the applied similarity measure between two data points, selection of k−nearest neighbors (KNN), and approach for the construction of a neighborhood graph. In this study, we integrate S-distance to spectral clustering, which is capable to find out the complex and non-linear cluster structures. Moreover, generalized mean distance-based KNN is proposed to decrease the sensitiveness towards the value of the k. Also, a symmetry-favored KNN method is applied to construct the neighborhood graph, which reduces the impact of outliers and noisy data points. However, spectral clustering faces scalability and speedup issues in the case of large size datasets. Thus, the proposed spectral clustering algorithm is also executed in distributed environments. Several experiments are performed to validate the proposed clustering algorithm on 20 real-world datasets and 3 large size datasets. Experimental results demonstrate that the proposed clustering algorithm outperforms some of the baseline methods in terms of accuracy and clustering error rates. Finally, we conduct Wilcoxon’s Rank-Sum test and illustrate that the proposed spectral clustering algorithm is statistically significant.

A new deterministic heuristic knots placement for B-Spline approximation

In this paper, we propose an adaptive knot placement algorithm for B-Spline curve approximation to dense and noisy 2D data points. The proposed algorithm is based on a heuristic rule for knot placement. It consists in constructing a distribution knot function by blending geometric criteria such as discrete derivatives, discrete angular variations and curvature. It has been successfully compared to three well known methods for approximating various noisy functions and sets of data in handwriting context.

If-SVM: Iterative factoring support vector machine

Support Vector Machine (SVM) is widely applied in classification and regression tasks where support vectors are pursued through convex quadratic programming technique due to its effectiveness and efficiency. However, existing studies ignore the importance of training samples when they are fed into the model. In this paper, we propose a novel Iterative Factoring Support Vector Machine (If-SVM) method. Sample factoring is introduced in our proposed model to measure the significance of each data point, where it can effectively reduce the negative impact of trivial or noisy data points. In this way, our proposed model is concentrates on the critical data points falling around the hyperplane. By introducing this iterative factoring of data points into SVM, the classification accuracy of our proposed method is above that of 1.45% than other comparative methods in image recognition datasets. Experimental results on a variety of UCI demonstrate that, our proposed method has superior performances in decreasing the total number of support vectors than the other state-of-the-art SVM methods. More importantly, our further experiments also illustrate that, the classification performance of the state-of-the-art SVM methods can be improved 1.29% by incorporating our sample factoring idea into their models, which demonstrate our idea is a useful tool to improve the state-of-art SVM models.

Robust low rank representation via feature and sample scaling

Low-rank representation (LRR) is a very competitive technique in various real-world applications for its powerful capability in discovering latent structure of noisy or corrupted data set. However, traditional low-rank models treat each data point and feature equally so that noisy data cannot be detected and suppressed effectively and have obvious deterioration in performance, especially in heavy noisy scenario. In this paper, to address this problem, we develop a method of feature and sample scaling for low rank representation. The importance of data points and their features in both feature and sample spaces are considered, as such, clean data points and noisy data points and their features can be distinguished. In addition, based on the observation that noisy data points are usually deviated far away from the principal projection of the data set, a cosine similarity metric between data vector and the principal projection vector is developed to measure the importance of each sample. Applying our method into two classical low rank models such as Low Rank Representation (LRR) and Bilinear Factorization (BF), we can learn better low-rank structure of clean data, while the outliers or missing data being suppressed. Extensive experimental results on ORL, COIL20 and video surveillance, demonstrate that our proposed method can outperform state-of-the-art low rank methods in image clustering tasks with various levels of corruptions, especially in a heavy noisy scenario.

A new contour method for rapid evaluation of the cross-sectional residual stress distribution in complex geometries using a 3D scanner

This paper presents a new method for contour measurements using a 3D scanner to determine cross-sectional residual stresses in structural components with complex configurations. The proposed method was applied to evaluate the axial residual stress distribution in a dissimilar pipe weld and the obtained results were compared with those obtained using a benchmarked contour measurement machine (CMM) equipped with a confocal laser probe for verification. It reveals a quantitative agreement between the two methods in terms of both the measured contours and the calculated residual stresses. Most importantly, this method allows a significantly reduced measurement time by ≈99 % (≈20 min, with an effective measurement speed of ≈0.2 sec/mm2), compared to the benchmarked method (≈48 hr, with an effective measurement speed of ≈28.7 sec/mm2). Based on stress uncertainty analysis, this method guarantees a longer range of stable fits due to fewer noisy data points than the benchmarked method.