Search

The adaptive neurofuzzy inference system (ANFIS) is a structured multioutput learning machine that has been successfully adopted in learning problems without noise or outliers. However, it does not work well for learning problems with noise or outliers. High-accuracy real-time forecasting of traffic flow is extremely difficult due to the effect of noise or outliers from complex traffic conditions. In this study, a novel probabilistic learning system, probabilistic regularized extreme learning machine combined with ANFIS (probabilistic R-ELANFIS), is proposed to capture the correlations among traffic flow data and, thereby, improve the accuracy of traffic flow forecasting. The new learning system adopts a fantastic objective function that minimizes both the mean and the variance of the model bias. The results from an experiment based on real-world traffic flow data showed that, compared with some kernel-based approaches, neural network approaches, and conventional ANFIS learning systems, the proposed probabilistic R-ELANFIS achieves competitive performance in terms of forecasting ability and generalizability.

Classification of wheat after harvest is a challenging task. Several techniques are prevalent for measuring the geometrical parameters of the wheat and for scrutinizing the wheat seeds. Parameters used for the classifier includes area, perimeter, compactness, length of kernel, width of kernel, asymmetry coefficient, length of kernel groove. Neural network pattern recognizer is used for the classification of the wheat seeds in this paper. The database for the Wheat seed is downloaded from the UCI machine learning repository. Confusion matrix and ROC is computed as the performance measure. Results show that neural network pattern recognizer is an apt tool for classification of wheat seeds with an overall accuracy of 96.7%.

Given a publicly available pool of machine learning models constructed for various tasks, when a user plans to build a model for her own machine learning application, is it possible to build upon models in the pool such that the previous efforts on these existing models can be reused rather than starting from scratch? Here, a grand challenge is how to find models that are helpful for the current application, without accessing the raw training data for the models in the pool. In this paper, we present a two-phase framework. In the upload phase, when a model is uploading into the pool, we construct a reduced kernel mean embedding (RKME) as a specification for the model. Then in the deployment phase, the relatedness of the current task and pre-trained models will be measured based on the value of the RKME specification. Theoretical results and extensive experiments validate the effectiveness of our approach.

Alzheimer's disease (AD) is a neurodegenerative disease of the human brain that affects neurotransmitters, tissue, and neurons that impair the senses, memories, and behaviors. Still, now there is no remedy for Alzheimer's disease. Even so, prescribed drugs can help reduce the development of the disease. That's why Alzheimer's early detection is very essential for treatment, and further research. Very limited numbers of trained samples and the higher volume of feature descriptions are the major difficulties in early diagnosis of Alzheimer's disease using different classification strategies. In this article, we proposed and related Alzheimer's disease early diagnostic method using Mild Cognitive Impairment (MCI), Structural Magnetic Resonance (sMR) imaging for AD-discrimination and healthy control participants (HC) with Import Vector Machine (IVM), Regularized Extreme Learning Machine (RELM) and a Support vector machine (SVM).The greedy score-based strategy for choosing essential function vectors is used. Furthermore, a discriminatory, kernel-based method is taken to treat dynamic data transformations. For volume sMR scan image data from Alzheimer's disease neuroimaging initiative (ADNI) repositories, we compare the performance of these classification models. An ADNI datasets experimental study reveals that RELM can greatly enhance the accuracy for classification of AD from MCIs as well as HC individuals along with feature selection methodology.

Semisupervised learning (SSL) has been extensively studied in related literature. Despite its success, many existing learning algorithms for semisupervised problems require specific distributional assumptions, such as ``cluster assumption'' and ``low-density assumption,'' and thus, it is often hard to verify them in practice. We are interested in quantifying the effect of SSL based on kernel methods under a misspecified setting. The misspecified setting means that the target function is not contained in a hypothesis space under which some specific learning algorithm works. Practically, this assumption is mild and standard for various kernel-based approaches. Under this misspecified setting, this article makes an attempt to provide a theoretical justification on when and how the unlabeled data can be exploited to improve inference of a learning task. Our theoretical justification is indicated from the viewpoint of the asymptotic variance of our proposed two-step estimation. It is shown that the proposed pointwise nonparametric estimator has a smaller asymptotic variance than the supervised estimator using the labeled data alone. Several simulated experiments are implemented to support our theoretical results.

By training different models and averaging their predictions, the performance of the machine-learning algorithm can be improved. The performance optimization of multiple models is supposed to generalize further data well. This requires the knowledge transfer of generalization information between models. In this article, a multiple kernel mutual learning method based on transfer learning of combined mid-level features is proposed for hyperspectral classification. Three-layer homogenous superpixels are computed on the image formed by PCA, which is used for computing mid-level features. The three mid-level features include: 1) the sparse reconstructed feature; 2) combined mean feature; and 3) uniqueness. The sparse reconstruction feature is obtained by a joint sparse representation model under the constraint of three-scale superpixels' boundaries and regions. The combined mean features are computed with average values of spectra in multilayer superpixels, and the uniqueness is obtained by the superposed manifold ranking values of multilayer superpixels. Next, three kernels of samples in different feature spaces are computed for mutual learning by minimizing the divergence. Then, a combined kernel is constructed to optimize the sample distance measurement and applied by employing SVM training to build classifiers. Experiments are performed on real hyperspectral datasets, and the corresponding results demonstrated that the proposed method can perform significantly better than several state-of-the-art competitive algorithms based on MKL and deep learning.

Background Pulse diagnosis (wrist pulse signal) is a well-known traditional technique used for a health examination. It has the potential to detect cardiac and non-cardiac diseases. Objective A study was conducted to investigate human emotions using wrist pulse signal assessment. The aim was to categorize anxiety, boredom, physical pain, and reference state by processing and analysis of acquired signals. Method A protocol was designed to induce emotions. Data were acquired from 24 healthy volunteers. Signals were processed and further analyzed using paired t-test and Analysis of Variance (ANOVA). Machine learning algorithms, Linear Discriminant Function (LDF), Quadratic Discriminant Function (QDF), and Support Vector Machine using kernel Gaussian radial basis function (RBF-SVM) were used to evaluate significant features and classify the emotions. Results Computing significant plus ranked features performed better over randomly selected features for pairwise emotion classification. Here, the QDF classifier outperforms LDF. Additionally, ANOVA validated the effectiveness of statistically prominent features to classify emotional states. Ratio_Pulse_Strength, total_power, Spectral Entropy, and meancd5 came out as the four most significant features to classify the emotion “Anxiety”, “Boredom”, “Pain”, and “Reference” with positive prediction rate of 100%, 73%, 100%, and 86% respectively using RBF-SVM in the user-independent model. Conclusion Previously, WPS has been used mainly to detect physical abnormality in the human body. The results endorse the potential of user-independent human emotion detection using a wrist pulse signal. The present work was focusing on a few emotional states. Results are encouraging and may be well applied to many more states.

Binary diffing is a process to discover the patched content between two binary programs. Previous research on binary diffing approach it as a function matching problem to formulate an initial 1:1 mapping between functions, and later a sequence matching ratio is computed to classify two functions being an exact match, a partial match or no-match. The accuracy of existing techniques is best only when detecting exact matches and they are not efficient in detecting partially changed functions; especially those with minor patch. These drawbacks are due to two major challenges (i) In 1:1 mapping phase, using a strict policy to match function features (ii) In classification phase, considering an assembly snippet as a normal text and using sequence matching for similarity comparison. An instruction has a unique structure i.e. mnemonics and registers have specific position in an instruction and also have semantic relationship, which makes assembly code different from general text. Sequence matching performs best for general text but it fail to detect structural and semantic changes at an instruction level thus, its use for classification produces many false results. In this research, we have addressed the aforementioned underlying challenges by proposing a two-fold solution. For 1:1 mapping phase, we have proposed computationally inexpensive features, which are compared with a distance based selection criteria to map the similar functions and filter unmatched functions. For classification phase, we have proposed a Siamese classification embedding neural network where each branch is an attention based distributed learning neural networkthat learn the semantic similarity among assembly instructions, learn to highlight the changes at an instruction level and a final stage fully connected layer learn to accurately classify two 1:1 mapped function either an exact or a partial match. We have used x86 kernel binaries for training and achieved 99% classification accuracy; which is higher than existing binary diffing techniques and tools.

In this paper we present a new classification method based on Dictionary Learning (DL). The main contribution consists of a kernel version of incoherent DL, derived from its standard linear counterpart. We also propose an improvement of the AK-SVD algorithm concerning the representation update. Our algorithms are tested on several popular databases of classification problems.

Multiple kernel clustering methods have been quite successful recently especially concerning the multi-view clustering of complex datasets. These methods simultaneously learn a multiple kernel metric while clustering in an unsupervised setting. With the motivation that some minimal supervision can potentially increase their effectiveness, we propose a Multiple Kernel Transfer Clustering (MKTC) method that can be described in terms of two tasks: a source task, where the multiple kernel metric is learned, and a target task where the multiple kernel metric is transferred to partition a dataset. In the source task, we create a weakly supervised multi-instance subset of the dataset, where a set of data instances are together provided some labels. We put forth a Multiple Kernel Multi-Instance $k$ -Means (MKMIKM) method to simultaneously cluster the multi-instance subset while also learning a multiple kernel metric under weak supervision. In the target task, MKTC transfers the multiple kernel metric learned by MKMIKM to perform unsupervised single-instance clustering of the entire dataset in a single step. The advantage of using a multi-instance setup for the source task is that it requires reduced labeling effort to guide the learning of the multiple kernel metric. Our formulations lead to a significantly lower computational cost in comparison to the state-of-the-art multiple kernel clustering algorithms, making them more applicable to larger datasets. Experiments over benchmark computer vision datasets suggest that MKTC can achieve significant improvements in clustering performance in comparison to the state-of-the-art unsupervised multiple-kernel clustering methods and other transfer clustering methods.

This paper presents a recursive feature elimination (RFE) mechanism to select the most informative genes with a least square kernel extreme learning machine (LSKELM) classifier. Describing the generalization ability of LSKELM in a way that is related to small norm of weights, we proposed a ranking criterion to evaluate the importance of genes by the norm of weights obtained by LSKELM network. The proposed method is called LSKELM-RFE algorithm, which first employs the original genes to build a LSKELM classifier, and then ranks the genes according to their importance given by the norm of LSKELM network output weights, and finally removes a least important gene. Benefiting from the random mapping mechanism of the extreme learning machine (ELM) kernel, there are no parameter of LSKELM-RFE needs to be manually tuned. A comparative study among our proposed algorithm and other two famous RFE algorithms has shown that LSKELM-RFE outperforms other RFE algorithms in both the computational cost and generalization ability.

Online multiple kernel learning (OMKL) has provided an attractive performance in nonlinear function learning tasks. Leveraging a random feature (RF) approximation, the major drawback of OMKL, known as the curse of dimensionality, has been recently alleviated. These advantages enable RF-based OMKL to be considered in practice. In this article, we introduce a new research problem, named stream-based active MKL (AMKL), in which a learner is allowed to label some selected data from an oracle according to a selection criterion. This is necessary for many real-world applications as acquiring a true label is costly or time consuming. We theoretically prove that the proposed AMKL achieves an optimal sublinear regret O(√T) as in OMKL with little labeled data, implying that the proposed selection criterion indeed avoids unnecessary label requests. Furthermore, we present AMKL with an adaptive kernel selection (named AMKL-AKS) in which irrelevant kernels can be excluded from a kernel dictionary ``on the fly.'' This approach improves the efficiency of active learning and the accuracy of function learning. Via numerical tests with real data sets, we verify the superiority of AMKL-AKS, yielding a similar accuracy performance with OMKL counterpart using a fewer number of labeled data.

The tremendous growth in deep learning (DL) applications has created an exponential demand for computing power, which leads to the rise of AI-specific hardware. Targeted towards accelerating computation-intensive deep learning applications, AI hardware, including but not limited to GPGPU, TPU, ASICs, etc., have been adopted ubiquitously. As a result, domainspecific CAD tools play more and more important roles and have been deeply involved in both the design and compilation stages of modern AI hardware. Recently, ISPD 2020 contest introduced a special challenge targeting at the physical mapping of neural network workloads onto the largest commercial deep learning accelerator, CS-1 Wafer-Scale Engine (WSE). In this paper, we proposed CU.POKer, a high-performance engine fullycustomized for WSE’s DNN workload placement challenge. A provably optimal placeable kernel candidate searching scheme and a data-flow-aware placement tool are developed accordingly to ensure the state-of-the-art quality on the real industrial benchmarks. Experimental results on ISPD 2020 contest evaluation suites demonstrated the superiority of our proposed framework over not only the state-of-the-art (SOTA) placer but also the conventional heuristics used in general floorplanning.

This article presents the design of an event-trigger-based networked robust dissipative tracking control systems. An event-triggered mechanism (ETM) is proposed by introducing the probability density distribution of communication delay. Thus, it can yield less conservative results. The tracking error system is transferred by a distributed communication delay system wherein the delay probability density is viewed as the kernel of the distributed delay. The threshold of the proposed ETM is designed as a dynamic parameter to adapt the transmission of the control signal. By applying the Lyapunov method, sufficient conditions are achieved to ensure the stability of the tracking error system with strictly dissipative performance, which includes $H_{\infty }$ performance and passive performance. Based on the dissipativity analysis condition, a co-design method with both tracking control and the ETM is derived in a united framework. Experiments on networked dc motor tracking system are provided to show the effectiveness of the presented method.

A key energy consumption in steel metallurgy comes from an iron ore sintering process. Enhancing carbon utilization in this process is important for green manufacturing and energy saving and its prerequisite is a time-series prediction of carbon efficiency. The existing carbon efficiency models usually have a complex structure, leading to a time-consuming training process. In addition, a complete retraining process will be encountered if the models are inaccurate or data change. Analyzing the complex characteristics of the sintering process, we develop an original prediction framework, that is, a weighted kernel-based fuzzy C-means (WKFCM)-based broad learning model (BLM), to achieve fast and effective carbon efficiency modeling. First, sintering parameters affecting carbon efficiency are determined, following the sintering process mechanism. Next, WKFCM clustering is first presented for the identification of multiple operating conditions to better reflect the system dynamics of this process. Then, the BLM is built under each operating condition. Finally, a nearest neighbor criterion is used to determine which BLM is invoked for the time-series prediction of carbon efficiency. Experimental results using actual run data exhibit that, compared with other prediction models, the developed model can more accurately and efficiently achieve the time-series prediction of carbon efficiency. Furthermore, the developed model can also be used for the efficient and effective modeling of other industrial processes due to its flexible structure.

Positron emission tomography (PET) imaging is used to track biochemical processes in the human body. PET image quality is limited by noise and several methods have been implemented to improve the quality. Kernel-based image reconstruction is among the methods implemented to increase PET image quality and commonly uses a Gaussian kernel to include spatial correlations from image priors into the forward projection model of PET. Unfortunately, the Gaussian kernel tends to smooth details in the reconstructed image. To reduce noise without losing contrast details, a different kernel is needed. Wavelet kernels can be more efficient than the Gaussian kernel in reducing noise while keeping contrast details 1 by better separating signal from noise and thus it does not over smooth peak values in the final reconstructed images. In this work, we evaluate a wavelet kernel for kernel-based PET image reconstruction. For this goal, a wavelet kernel approach has been tested using simulated brain data, physical phantom data and patient data. Reconstruction results are presented and discussed in detail comparing the wavelet kernel method with the Gaussian kernel method. Our results suggest that a wavelet kernel performs better in contrast recovery for phantoms and also results in higher signal-to-noise ratio (SNR) for real patient data.

This work proposes a novel data-driven robust model predictive control (DDRMPC) framework for automatic control of greenhouse in-door climate. The framework integrates dynamic control models of greenhouse temperature, humidity, and CO₂ concentration level with data-driven robust optimization models that accurately and rigorously capture uncertainty in weather forecast error. Data-driven uncertainty sets for ambient temperature, solar radiation, and humidity are constructed from historical data by leveraging a machine learning approach, namely, support vector clustering with weighted generalized intersection kernel. A training-calibration procedure that tunes the size of uncertainty sets is implemented to ensure that data-driven uncertainty sets attain an appropriate performance guarantee. In order to solve the optimization problem in DDRMPC, an affine disturbance feedback policy is utilized to obtain tractable approximations of optimal control. A case study of controlling temperature, humidity, and CO₂ concentration of a semiclosed greenhouse in New York City is presented. The results show that the DDRMPC approach ends up with 14% and 4% lower total cost than rule-based control and robust model predictive control with L₁-norm-based uncertainty set, respectively. The constraint violation probability, which is the percentage of time that the greenhouse system states violate the constraint throughout the whole growing period, for DDRMPC is only 0.39%. Hence, the proposed DDRMPC framework can prevent the greenhouse climate from becoming harmful to plants and fruits. In conclusion, the proposed DDRMPC approach can improve the greenhouse climate control performance and reduce cost compared with other control strategies.

While various biomimetic robotic haptic sensors are often utilized to measure multifarious physical interactions to identify material properties under human exploratory procedures (EPs), traditional methods are unable to fuse well multimodal measurements under EPs. In order to solve this problem, an innovative hybrid joint group kernel sparse coding model (HJGKSC) for material recognition under EPs is proposed. First, a series of feature representations according to the characteristics of different measures are introduced. Second, we propose an innovative hybrid fusion framework based on statistical learning with kernel sparsity to effectively fuse the measurements from robotic EPs. Third, a matched optimization algorithm is constructed to deal with the innate optimization problem in the proposed framework. Finally, we develop an experiment based on three different classification levels (coarse, medium, fine) with the public dataset containing 184 material classes to verify the proposed method and compare with some other similar kernel methods. Experimental results show that the accuracy of coarse, medium and fine classification of the proposed hybrid fusion framework is 98.4%, 96.2% and 98.4%, respectively. The framework can be extended easily to other fusion tasks with multiple measurements.

Touch modality identification has attracted increasing attention due to its importance in human-robot interactions. There are three issues involved in the tactile perception for the touch modality identification, including the high dimensionality of tactile signals, complex tensor morphology of tactile sensing units, and the misalignment among different tactile time-series samples. In this article, we propose a novel kernel-based approach to deal with these three issues in a unified framework. Specifically, the techniques, including sparse principal component analysis and subsampling, are employed to reduce the feature dimension. Then, a singular value decomposition (SVD)-based kernel is proposed to preserve the spatial information of the tactile sensing elements. The sample misalignment issue is addressed via the employment of a global alignment kernel. Moreover, the merits of these two kernels are fused through an ideal regularized composite kernel, which simultaneously takes the label information of the training set into consideration. The effectiveness of the proposed kernel-based approach is verified on a public touch modality data set with a comprehensive comparison with the competing methods.

Computer-assisted pathology analysis is an emerging field in health informatics and extremely important for effective treatment. Herein, we demonstrated the ability of the pre-trained Xception model for magnification-dependent breast cancer histopathological image classification in contrast to handcrafted approaches. The Xception model and SVM classifier with the ‘radial basis function’ kernel has achieved the best and consistent performance with the accuracy of 96.25%, 96.25%, 95.74%, and 94.11% for 40X, 100X, 200X and 400X level of magnification, respectively. A comparison with existing state-of-the-art techniques has been conducted based on accuracy, recall, precision, F1 score, area under ROC and precision–recall curve evaluation metrics.

Loop closure detection (LCD) is of significant importance in simultaneous localization and mapping. It represents the robot's ability to recognize whether the current surrounding corresponds to a previously observed one. In this paper, we conduct this task in a two-step strategy: candidate frame selection and loop closure verification. The first step aims to search semantically similar images for the query one using features obtained by Key.Net with HardNet. Instead of adopting the traditional Bag-of-Words strategy, we utilize the aggregated selective match kernel to calculate the similarity between images. Subsequently, based on the potential property of motion field in the LCD scene, we propose a novel feature matching method, i.e., exploiting the smoothness prior and learning the motion field for an image pair in a reproducing kernel Hilbert space (RKHS), to implement loop closure verification. Concretely, we formulate the learning problem into a Bayesian framework with latent variables indicating the true/false correspondences and a mixture model accounting for the distribution of data. Furthermore, we propose a locality-driven mechanism to enhance the local relevance of motion vectors and term the algorithm as locality-driven accurate motion field learning (LAL). To satisfy the requirement of efficiency in the LCD task, we use a sparse approximation and search a suboptimal solution for the motion field in the RKHS, termed as LAL*. Extensive experiments are conducted on public datasets for feature matching and LCD tasks. The quantitative results demonstrate the effectiveness of our method over the current state-of-the-art, meanwhile showing its potential for long-term visual localization. The codes of LAL and LAL* are publicly available at https://github.com/KN-Zhang/LAL.

There are two types of important information in a polarimetric synthetic aperture radar (PolSAR) image: spatial features in two dimensions and polarimetric characteristics in the scattering dimension. Considering both polarimetric and spatial information is important for PolSAR image classification. Convolutional kernels show superior performance for extraction of spatial information from two dimensions of an image in convolutional neural networks (CNNs). But learning CNNs needs large enough training sets to achieve the optimum weights of kernels while there are not usually sufficient training samples for PolSAR images. To deal with this difficulty, a convolutional kernel‐based covariance descriptor (CKCD) is introduced for PolSAR image classification in this study. To extract contextual characteristics, compatible with the original image, the fixed‐valued convolutional kernels randomly selected from the image are used, which do not require any learning, and so do not need any training samples. To include more local spatial information and find the relation among the polarimetric features, the covariance descriptor is constructed on the extracted feature maps. Then, the polarimetric‐contextual features are given to a support vector machine with a matrix logarithm‐based kernel. Finally, the guided filter is applied to the initial classification map to result a smoothed classification map with preserved edges. The experiments on three real PolSAR images show superiority of the proposed CKCD method compared to several PolSAR classification methods such as 2DCNN and 3DCNN in small sample size situations.

The present study provided the first-time comprehensive evaluation of 12 advanced statistical and machine learning (ML) algorithms for the Soil Moisture (SM) estimation from dual polarimetric Sentinel-1 radar backscatter. The ML algorithms namely support vector machine (SVM) with linear, polynomial, radial and sigmoid kernel, random forest (RF), multi-layer perceptron (MLP), radial basis function (RBF), Wang and Mendel’s (WM), subtractive clustering (SBC), adaptive neuro fuzzy inference system (ANFIS), hybrid fuzzy interference system (HyFIS), and dynamic evolving neural fuzzy inference system (DENFIS) were used. Extensive field samplings were performed for collection of in-situ SM data and other parameters from the selected sites for seven different dates and at two different locations (Varanasi and Guntur District, India), concurrent to Sentinel-1 overpasses. The backscattering coefficients were considered as input variables and SM as output variable for the training, validation and testing of the ML algorithms. The site at Varanasi was used for the training, validation and testing of the models. On the other hand, the Guntur site was used as an independent site for checking the model performance, before finalizing the algorithms. The performances of different trained algorithms were evaluated in terms of correlation coefficient (r), root mean square error (RMSE) (in m3/m3) and bias (in m3/m3). The study identified the RF, SBC and ANFIS as the top three best performing models with comparable and promising SM estimation. In order to test the robustness of these best models (RF, SBC and ANFIS), further performance analysis was performed to the independent datasets of the Varanasi and Guntur test sites, which indicates that the performance of these three models were consistent and SBC can be recommended as the best among all for SM estimation.

In this article, a novel fault detection and process monitoring method referred to as artificial neural correlation analysis (ANCA) is proposed. Because nonlinear characteristics are common in complex industrial processes, the classic canonical correlation analysis (CCA) always perform poorly. Many scholars have noticed the nonlinear problem of the process and have also proposed some improved schemes, such as the kernel method. However, the selection of suitable parameters in the kernel method is extremely difficult, so most of the kernel learning methods are slightly unsatisfactory. Considering that the artificial neural network (ANN) can well extract the required feature components from the nonlinear data, we combined ANN and CCA from their respective principles, and proposed a new nonlinear monitoring method and the detailed gradient descent method derivation for the ANCA network is presented. In addition, we have designed two indices to monitor the changes of process variables and performance indicators. Finally, a numerical example, the Tennessee Eastman benchmark, and the Zhoushan thermal power plant process illustrate the superiority of the proposed method.

Regional as well as global cropland database at a high spatial resolution like 10 m is not available. This study aims to integrate the Sentinel-1 synthetic aperture radar (SAR) data and Sentinel-2 optical data in a more robust way to map the croplands at 10 m spatial resolution. It also examines the effectiveness of different derived products from the SAR and multispectral data. The VH and VV polarizations of the SAR data were used to generate the 80 textural parameters at four different kernel sizes. The multispectral bands of the optical data were used to create nine different spectral indices. The SAR and optical data were classified separately and integrated with the other derived products, using the random forest classifier, to identify the most potential way to use the two datasets for LULC classification. The classification accuracy was the least when only the Sentinel 1 data was used, with an overall accuracy of 59.48% and a kappa value of 0.51. The best accuracy (97.94% overall accuracy and 0.98 kappa value) was achieved by combining the Sentinel 1 and 2 data and their optimal textures and spectral indices. Overall, this study suggests the suitability of the SAR data along with optical data to map the croplands with higher accuracy.

OpenCL programs typically employ complex storage models and diverse data types as well as manifest various memory access patterns, which make it challenging to detect the performance problems effectively. However, few research efforts have been dedicated to cope with this challenge so far. In this paper, we introduce CVFuzz, a domain-independent tool that can effectively detect and locate algorithmic complexity vulnerabilities in OpenCL kernels. The key enabling idea is leveraging automatically generated pathological inputs to trigger the worst-case behavior during the execution of OpenCL kernels. Our approach takes advantage of the metrics such as code coverage and run time to guide the generation of inputs that can slow down the execution of a given OpenCL kernel. We evaluate CVFuzz on more than 250 real-world OpenCL kernels. The evaluation results demonstrate that the inputs generated by CVFuzz are effective in detecting the worst-case time algorithmic complexity and optimization vulnerabilities.

In the field of data analysis clustering is considered to be a major tool. Application of clustering in various field of science, has led to advancement in clustering algorithm. Traditional clustering algorithm have lot of defects, while these defects have been addressed but no clustering algorithm can be considered as superior. A new approach based on Kernel Fuzzy C-means clustering using teaching learning-based optimization algorithm (TLBO-KFCM) is proposed in this paper. Kernel function used in this algorithm improves separation and makes clustering more apprehensive. Teaching learning-based optimization algorithm discussed in the paper helps to improve clustering compactness. Simulation using five data sets are performed and the results are compared with two other optimization algorithms (genetic algorithm GA and particle swam optimization PSO). Results show that the proposed clustering algorithm has better performance. Another simulation on same set of data is also performed, and clustering results of TLBO-KFCM are compared with teaching learning-based optimization algorithm with Fuzzy C- Means Clustering (TLBO-FCM).

One fundamental task of hyperspectral imaging is spectral unmixing. In this case, the conventional pure pixel-based hyperspectral image classification (HSIC) may not work effectively for mixed pixels. This article presents a kernel-based approach to hyperspectral mixed pixel classification (HMPC) which includes two nonlinear mixed pixel classifiers, kernel constrained energy minimization (KCEM) and kernel linearly constrained minimum variance (KLCMV) to replace the widely used pure pixel-based support vector machine (SVM) classifier. Interestingly, what the binary-class and multiclass SVM classifiers are to pure pixel-based HSIC can be similarly derived for what a single-class KCEM detector and a multiclass KLCMV detector are to HMPC. In particular, the commonly used discrete classification map-based hard classification measures, average accuracy (AA) and overall accuracy (OA) for performance evaluation can be further generalized to real-valued mixed class abundance fractional map-based soft classification measures via 3-D receiver operating characteristic (3-D ROC) analysis-derived detection measures. Extensive experiments are conducted to demonstrate the utility of HMPC where KCEM/KLCMV not only significantly improve the classification performance of CEM/LCMV-based classifiers but also outperform many existing spectral-spatial classification methods.

The kernel principal component analysis (KPCA) is widely used as a fault monitoring tool for complex nonlinear chemical processes in recent years. The cumulative contribution rate that extracts the kernel principal is usually obtained relied on a fixed model, which cannot be employed for time-varying chemical processes. Hence, a novel adaptive kernel principal component analysis (AKPCA) integrating grey relational analysis (GRA) (AKPCA-GRA) is proposed to dynamically monitor the fault occurrence. A moving window integrating the threshold method is used to adaptively extract the kernel principal for chemical processes. Then the corresponding T2 and Q statistics calculated by the selected kernel principal based on the AKPCA decides whether the fault has occurred. Moreover, the GRA method is used to analyze and calculate the correlation coefficient of abnormal features obtained based on the APKCA method, which provides the operational guidance for the nonlinear chemical process to find out the variable causing the fault. Finally, the proposed method is verified using the Tennessee Eastman (TE) process. The case results demonstrate that the proposed method outperforms the KPCA, the KPCA based on the threshold and the moving window principal component analysis, the support vector machine (SVM) and the Logistic Regression (LR) in terms of the missed alarm rate (MAR) and the false alarm rate (FAR), which can effectively analyze the variables causing the fault.

Improving the performance of nonlinear unmixing has become an active topic among the remote sensing applications. Usually, the noise levels of hyperspectral images (HSIs) vary with different bands. However, this fact is generally ignored and may, to some extent, result in a degradation of the unmixing results. Nonetheless, valuable spatial information that provides a great potential for improving the performance has seldom been considered in the current nonlinear unmixing. In this letter, we propose a novel kernel-based nonlinear unmixing model in which the band-wise noise characterization and the spatial relationships of HSIs are incorporated to solve the above problems. Firstly, the noise levels of different bands are estimated based on the results of superpixel segmentation, and then they are used to characterize the roles of different bands in the unmixing process. To exploit the spatial relationships in the superpixels, a regional l₁ -norm regularization is proposed and incorporated into the unmixing model. Experimental results on both synthetic and real hyperspectral datasets demonstrate the superiority of the proposed model compared to the state-of-the-art nonlinear unmixing methods.

Radar images with high spatial resolution are not affected by illumination or meteorological conditions, which effectively compensate for the shortcomings of optical images and passive microwave images. Thus, active microwave remote sensing technology has advantages in snow depth (SD) research. Machine learning algorithms (MLAs), which do not need to consider complex physical models, have increasingly been applied to SD research. Considering the snow conditions of different underlying surfaces, it is very important to select the appropriate parameter combinations (PC) reflecting the SD information for MLAs. In this study, C-band SAR data with 20 m spatial resolution, groundbased SD observation data from meteorological stations, and field measurement data in Northeast China were used to construct and validate the SD estimation method. Two parameter selection methods including the correlation coefficient method and the machine learning (ML) fusion method were proposed to discuss the influence of different PC on SD estimation. Then, XGBoost, random forest (RF), linear support vector regression (LSVR), and kernel support vector regression (KSVR) were applied to estimate SD based on the selected PC, and evaluate the accuracies of SD estimation using different MLAs. The results demonstrated that the PC selected using the correlation coefficient method and XGBoost algorithm could achieve the best SD results in the study area. Combining RPC-C with XGBoost algorithm, in cropland areas, the average values of mean absolute error (MAE) and root mean squared error (RMSE) were 1.75 cm and 2.58 cm, respectively. Combining RPC-F with XGBoost algorithm, in forest areas, the average values of MAE and RMSE were 3.12 cm and 5.07 cm, respectively. The research of this paper can select the optimal PC for MLAs and effectively improve the accuracy of SD estimation using C-band SAR data.

A massive core computing architecture is developed for accelerating arbitrary calculations in fully parallel with high speed and low cost. The proposed architecture is reconfigurable in fine-grained (arbitrary functions), mid-grained (flexible function feature, accuracy, and number of operands), and coarse-grained (organization of cores). By implementing a large scale of novel bisection neural network (BNN) on hardware, the re-configuration is conducted by partitioning entire BNN into any specific pieces without redundancy. Each piece of BNN retrieves the arbitrary function approximately. By reconfiguring the BNN topology in software, we can easily adjust dimensions of the computing kernel without rewiring, and achieve a wide range of trade-offs between accuracy and efficiency in hardware. In this manner, the multi-grained reconfigurable accelerator (MuGRA) is achieved. Since MuGRA is flexible in all grained levels, various configurations for each validation are demonstrated with rich options of performance-cost matrix. From the FPGA implementation results, compared with other traditional function approximation methods, our method provides fewer parameter storage requirements. The comparison against related works proves that our accelerator effectively reduces the calculation latency with slight accuracy loss.

Hyperspectral images (HSIs) are used in a large number of real-world applications. HSI classification (HSIC) is a challenging task due to high interclass similarity, high intraclass variability, overlapping, and nested regions. The 2-D convolutional neural network (CNN) is a viable classification approach since HSIC depends on both spectral-spatial information. The 3-D CNN is a good alternative for improving the accuracy of HSIC, but it can be computationally intensive due to the volume and spectral dimensions of HSI. Furthermore, these models may fail to extract quality feature maps and underperform over the regions having similar textures. This work proposes a 3-D CNN model that utilizes both spatial-spectral feature maps to improve the performance of HSIC. For this purpose, the HSI cube is first divided into small overlapping 3-D patches, which are processed to generate 3-D feature maps using a 3-D kernel function over multiple contiguous bands of the spectral information in a computationally efficient way. In brief, our end-to-end trained model requires fewer parameters to significantly reduce the convergence time while providing better accuracy than existing models. The results are further compared with several state-of-the-art 2-D/3-D CNN models, demonstrating remarkable performance both in terms of accuracy and computational time.

Background: Breast cancer is the development of a malignant tumor in the breast of human beings (especially females). If not detected at the initial stages, it can substantially lead to an inoperable construct. It is a reason for majority of cancer-related deaths throughout the world. Objectives: The main aim of our study is to diagnose the breast cancer at early stage so that required treatment can be provided for survival. The tumor is classified as malignant or benign accurately at early stage using a novel approach that includes an ensemble of Genetic Algorithm for feature selection and kernel selection for SVM-Classifier. Methods: The proposed GA-SVM (Genetic Algorithm – Support Vector Machine) algorithm in this paper optimally selects the most appropriate features for training with the SVM classifier. Genetic Programming is used to select the features and the kernel for the SVM classifier. Genetic Algorithm operates by exploring the optimal layout of features for breast cancer, thus, subjugating the problems faced in exponentially immense feature space. Results: The proposed approach accounts for a mean accuracy of 98.82% by using the Wisconsin Diagnostic Breast Cancer (WDBC) dataset available on UCI with the training and testing ratio being 50:50 respectively. Conclusion: The results prove that our proposed model outperforms the previously designed models for breast cancer diagnosis. The outcome assures that the GA-SVM model may be used as an effective tool in assisting the doctors for treating the patients. Alternatively, it may be utilized as an alternate opinion in their eventual diagnosis.

Nowadays, machine learning technology is needed in the medical field. therefore, this research is useful for solving problems in the medical field by using machine learning. Many cases of colorectal cancer are diagnosed late. When colorectal cancer is detected, the cancer is usually well developed. Machine learning is an approach that is part of artificial intelligence and can detect colorectal cancer early. This study discusses colorectal cancer detection using twin support vector machine (SVM) method and kernel function i.e. linear kernels, polynomial kernels, RBF kernels, and gaussian kernels. By comparing the accuracy and running time, then we will know which method is better in classifying the colorectal cancer dataset that we get from Al-Islam Hospital, Bandung, Indonesia. The results showed that polynomial kernels has better accuracy and running time. It can be seen with a maximum accuracy of twin SVM using polynomial kernels 86% and 0.502 seconds running time.

Halodoc is a telemedicine-based healthcare application that connects patients with health practitioners such as doctors, pharmacies, and laboratories. There are some comments from halodoc users, both positive and negative comments. This indicates the public's concern for the Halodoc application so it is necessary to analyze the sentiment or comments that appear on the Halodoc application service, especially during the COVID-19 pandemic in order for Halodoc application services to be better. The Naïve Bayes Classifier (NBC) and Support Vector Machine (SVM) algorithms are used to analyze the public sentiment of Halodoc's telemedicine service application users. The negative category sentiment classification result was 12.33%, while the positive category sentiment was 87.67% from 5,687 reviews which means that the positive review sentiment is more than the negative review sentiment. The accuracy performance of the Naive Bayes Classifier Algorithm resulted in an accuracy rate of 87.77% with an AUC value of 57.11% and a G-Mean of 40.08%, while svm algorithm with KERNEL RBF had an accuracy value of 86.1% with an AUC value of 60.149% and a G-Mean value of 49.311%. Based on the accuracy value of the model can be known SVM Kernel RBF model better than NBC on classifying the review of user sentiment of halodoc telemedicine service

We introduce a simulation method for dynamic portfolio valuation and risk management building on machine learning with kernels. We learn the dynamic value process of a portfolio from a finite sample of its cumulative cash flow. The learned value process is given in closed form thanks to a suitable choice of the kernel. We show asymptotic consistency and derive finite sample error bounds under conditions that are suitable for finance applications. Numerical experiments show good results in large dimensions for a moderate training sample size.

The subspace clustering methods for motion segmentation are widely used in the field of computer vision. However, the existing methods ignore the low-rank property of motion trajectory with nonlinear structure and are sensitive to non-Gaussian noise. To this end, we seek to improve the performance of motion segmentation by effectively modeling some important characteristics of the motion trajectories, such as nonlinear structure and contained non-Gaussian noise. Specifically, we propose to use kernel function to model motion trajectory, design a variant of the correntropy-induced metric to measure noise, and integrate the block diagonal regularizer into the kernel subspace clustering to strengthen the block diagonal structure of the learned affinity matrix. More importantly, we propose a unified rank-constrained block diagonal subspace clustering method for motion segmentation, which can handle not only rigid body motion segmentation, but also non-rigid motion segmentation. And we further extend this method to deal with various noises in motion data, such as missing trajectories, corrupted trajectory and outlying trajectory. An effective algorithm HQ& AM, which is integrated by Half-quadratic theory and alternating minimization, is designed to optimize these models. Experimental results on several commonly used motion datasets indicate the effectualness and robustness of our methods.

Researchers often face the problem of how to address missing data. Multiple imputation is a popular approach, with multiple imputation by chained equations (MICE) being among the most common and flexible methods for execution. MICE iteratively fits a predictive model for each variable with missing values, conditional on other variables in the data. In theory, any imputation model can be used to predict the missing values. However, if the predictive models are incorrectly specified, they may produce biased estimates of the imputed data, yielding inconsistent parameter estimates and invalid inference. Given the set of modeling choices that must be made in conducting multiple imputation, in this paper we propose a data-adaptive approach to model selection. Specifically, we adapt MICE to incorporate an ensemble algorithm, Super Learner, to predict the conditional mean for each missing value, and we also incorporate a local kernel-based estimate of variance. We present a set of simulations indicating that this approach produces final parameter estimates with lower bias and better coverage than other commonly used imputation methods. These results suggest that using a flexible machine learning imputation approach can be useful in settings where data are missing at random, especially when the relationships among the variables are complex.

© 2021, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.Size, mass, and shape attributes play a significant role in the quality assessment and post-harvest technologies of agricultural products. Pistachio is widely consumed worldwide, and Turkey has 3rd place in world pistachio production. In this study, physical attributes of 6 different pistachio cultivars (Beyaz Ben, Keten gömleği, Kirmizi, Siirt, Tekin, Uzun) were determined and machine learning algorithms (Multilayer Perceptron (MLP), k-Nearest Neighbor (kNN), Random Forest (RF), Gaussian processes (GP)) were used for mass prediction of these pistachio cultivars. Siirt and Tekin cultivars had the greatest gravitational and dimensional attributes. Among the pistachio cultivars, Kirmizi and Uzun had the greatest shape index and elongation values. Keten gömleği and Beyaz cultivars had the lowest averages of mass and area attributes both for nuts and kernels. Kernel and nut mass of pistachio had significant correlations with volume, geometric mean diameter, and projected and surface area (p < 0.01). Present findings revealed that Gaussian processes had the greatest correlation coefficients (0.976 for nut mass and 0.948 for kernel mass prediction) and the lowest RMSE values (0.038 for nut and 0.029 for kernel mass prediction). This algorithm was respectively followed by Multilayer Perceptron and Random Forest algorithms. Present findings revealed that Gaussian processes, Multilayer Perceptron, and Random Forest algorithms could potentially be used for mass prediction of pistachio cultivars.

Dementia due to Alzheimer’s disease has become a global health burden. Early detection of one of its potential prodromal state, mild cognitive impairment (MCI), appears to be critical for early interventions. The present study, therefore, addressed the challenging issue of classifying individuals with MCI vs. healthy controls (HC) based on resting-state electroencephalography (EEG). Specifically, we design a feature extraction and feature fusion framework to extract discriminative features. In the first step, we extracted the sample entropy (SampEn) and Katz’s fractal dimension (KFD) features from their corresponding best electrodes selected by a wrapper-based algorithm. Next, the two best feature sets were fused by a principal component analysis (PCA) method. The eigen-complexity pattern (ECP) feature extracted by this framework was subsequently fed into the proposed MCI detector, the conformal kernel-based fuzzy support vector machine (CKF-SVM). The CKF-SVM not only adopts a fuzzified penalty strategy to avoid the overfitting often observed in conventional SVM due to outliers, but also applies a conformally transformed kernel to further increase the class separability. The results carried out on 51 participants (24 MCI, 27 HC) and leave-one-participant-out cross-validation (LOPO-CV) show that the ECP feature combined with a simple linear discriminant analysis classifier achieved an accuracy of 84.31%, higher than the one by either of the two complexity measures (SampEn and KFD), as well as the ones by spectral powers of different scalp regions. The results also show that CKF-SVM outperformed SVM and other classifiers commonly used in the MCI-HC classification studies. A high LOPO-CV accuracy 90.19% (sensitivity = 87.50%, specificity = 92.60%) was achieved by using the ECP feature and the CKF-SVM classifier. These results suggested that the proposed approach has a potential for developing a sensitive EEG-based computer-aided diagnosis (CAD) system that may, in the future, provide an objective measure to assist physicians’ diagnose of MCI and even a neurofeedback brain–computer interface (BCI) system for monitoring the intervention response of individuals with MCI.

Path tracking control of intelligent vehicles has to deal with the difficulties of model uncertainties and nonlinearities. As a class of adaptive optimal control methods, reinforcement learning (RL) has received increasing attention in solving difficult control problems. However, feature representation and online learning ability are two major problems to be solved for learning control of uncertain dynamic systems. In this article, we propose a multi-kernel online RL approach for path tracking control of intelligent vehicles. In the proposed approach, a multiple kernel feature learning framework is designed for online learning control based on dual heuristic programming (DHP) and the new online learning control algorithm is called multi-kernel DHP (MKDHP). In MKDHP, instead of the expert knowledge for selecting and fine-tuning of a suitable kernel function, only a set of basic kernel functions is required to be predefined and the multi-kernel features can be learned for value function approximation in the critic. The simulation studies on path tracking control for intelligent vehicles have been conducted under $S$ -curve and urban road conditions. The results demonstrated that compared with other typical path tracking controllers for intelligent vehicles, such as the linear quadratic regulator (LQR), the pure pursuit controller and the ribbon-based controller, the proposed multi-kernel learning controller can achieve better performance in terms of tracking precision and smoothness.

Extreme learning machine (ELM) has fast learning speed and perfect performance, at the same time, ELM provides a unified learning framework with a widespread type of feature mappings which can be applied in multiclass classification applications directly. These advantages make ELM become one of the best classification algorithms, and ELM has attracted great attention in supervised learning and semi-supervised learning. However, noise and outliers of data are usually existed in the real world, which will affect the performance of ELM. To improve the robustness and classification performance of ELM, we propose the Kernel Risk-Sensitive Mean p-power Loss Based Hyper-graph Regularized Robust Extreme Learning Machine (KRP-HRELM) method. On the one side, as a nonlinear similarity measure defined in the reproducing kernel space, the kernel risk-sensitive mean p-power loss (KRP) can effectively weaken the negative effects caused by noise and outliers. Therefore, the KRP is introduced into ELM to enhance its robustness. Then, the application of hyper-graph can help the ELM to explore higher-order geometric structures among more sampling points, thereby obtaining more comprehensive data information. In addition, to obtain a more sparsity network model, the L2,1-norm is used to constrain the output weight. On the other side, improving the practical application ability of KRP-HRELM is also the focus of our research, so KRP-HRELM is extended to semi-supervised learning, which is called the semi-supervised KRP-HRELM (SS-KRP-HRELM). Notably, the results of the robustness experiment have proved that our method has extraordinary robustness. At the same time, by using four evaluation measures such as accuracy, recall, precision, and F1-measure, to evaluate the classification results, we can find that our method has obtained better classification performance than other advanced methods.

We investigate the computational issues related to the memory size in the estimation of quadratic covariation, taking into account the specifics of financial ultra-high-frequency data. In multivariate price processes, we consider both contamination by the market microstructure noise and the non-synchronicity of the observations. We formulate a multi-scale, flat-top realized kernel, non-flat-top realized kernel, pre-averaging and modulated realized covariance estimators in quadratic form and fix their bandwidth parameter at a constant value. This allows us to operate with limited memory and formulate this estimation as a streaming algorithm. We compare the performance of the estimators with fixed bandwidth parameter in a simulation study. We find that the estimators ensuring positive semidefiniteness require much higher bandwidth than the estimators without this constraint.

In this paper, support vector machines (SVMs), least squares SVMs (LSSVMs), relevance vector machines (RVMs), and probabilistic classification vector machines (PCVMs), are compared on sixteen binary and multiclass medical datasets. Particular emphasis is put on the comparison among the commonly used Gaussian radial basis function (GRBF) kernel, and the relatively new generalized min–max (GMM) kernel and exponentiated-GMM (eGMM) kernel. Since most medical decisions involve uncertainty, a postprocessing approach based on Platt’s method and pairwise coupling is employed to produce probabilistic outputs for prediction uncertainty assessment. The extensive empirical study illustrates that the SVM classifier using the tuning-free GMM kernel (SVM-GMM) shows good usability and broad applicability, and exhibits competitive performance against some state-of-the-art methods. These results indicate that SVM-GMM can be used as the first-choice method when selecting an appropriate kernel-based vector machine for medical diagnosis. As an illustration, SVM-GMM efficiently achieves a high accuracy of 98.92% on the thyroid disease dataset consisting of 7200 samples.