Search

Human-object interaction detection is one of the key issues of scene understanding. It has widespread applications in advanced computer vision technology. However, due to the diversity of human postures, the uncertainty of the shape and size in objects, as well as the complexity of the relationship between people and objects. It is very challenging to detect the interaction relationship between people and objects. To solve this problem, this paper proposes a multi-scale attention fusion method to adapt to people and objects of different sizes and shapes. This method increases the range of attention which can more accurately judge the relationships between people and objects. Besides, we further propose a weighting mechanism to better characterize the interaction between people and close objects and express people’s intention of interaction. We evaluated the proposed method on HICO-DET and V-COCO datasets, which has verified its effectiveness and flexibility as well as has achieved a certain improvement in accuracy.

Today’s networks carry a large amount of data, it is common for traffic to exceed the capacity of the network links, this event is called congestion. Therefore, the development of congestion control algorithms has become a key task in improving the performance of networks. BBR is a congestion control algorithm that continuously measures bottlenecks and RTT to establish the size of the congestion window. WBBR is an algorithm that incorporates a weighting factor to BBR to achieve fairness. The present work proposes an implementation of WBBR in ns-3 and performs a comparative evaluation with BBR since it has not been done before. The implemented module was validated with the previous results available in the literature. The results show that WBBR is fairer than BBR achieving load balancing, without increasing the RTT values.

Recommender systems (RS) play a key role in e-commerce by preselecting presumably interesting products for customers. Hybrid RSs using a weighted average of individual RSs’ predictions have been widely adopted for improving accuracy and robustness over individual RSs. While for regression tasks, approaches to estimate optimal weighting schemes based on individual RSs’ out-of-sample errors exist, there is scant literature in classification settings. Class prediction is important for RSs in e-commerce, as here item purchases are to be predicted. We propose a method for estimating weighting schemes to combine classifying RSs based on the variance-covariance structures of the errors of individual models' probability scores. We evaluate the approach on a large real-world ecommerce data set from a European telecommunications provider, where it shows superior accuracy compared to the best individual model as well as a weighting scheme that averages the predictions using equal weights.

We consider the use of multipreconditioning to solve linear systems when more than one preconditioner is available but the optimal choice is not known. In particular, we consider a selective multipreconditioned GMRES algorithm where we incorporate a weighting that allows us to prefer one preconditioner over another. Our target application lies in the simulation of incompressible two-phase flow. Since it is not always known if a preconditioner will perform well within all regimes found in a simulation, we also consider robustness of the multipreconditioning to a poorly performing preconditioner. Overall, we obtain promising results with the approach.

The investigated method enables precise shaping of acoustic radiation of a vibrating plate, i.e. it allows one to relocate or create resonances and anti-resonances for selected frequencies, simultaneously altering their acoustic radiation efficiency in a desired manner. The method can be very beneficial for plates used as noise barriers, both in passive and active applications. The acoustic radiation shaping method involves mounting several additional ribs and masses to the plate surface at locations followed from an optimization process (sensors and actuators can also be included, if active control is considered). The optimization process requires a model of the vibroacoustic system, a cost function corresponding to the considered objective, and an optimization algorithm. In this paper, an introduction of A-weighting to the cost functions, which reflects a human perception of the noise radiated by or transmitted through the plate, is investigated. It follows from the analysis of obtained results that the introduction of A-weighting can provide even 8 dBA better passive noise attenuation.

Relational data models are usually used by researchers and companies when a single-table model is not enough to describe their system. Then, when it comes to classification, there are mainly two options: apply the corresponding relational version of classification algorithms or use a propositionalization technique to transform the relational database into a single-table representation before classification. In this work, we evaluate a fast and simple propositionalization algorithm called Wordification. This technique uses the table name, attribute name and value to create a feature. Each feature is treated as a word and the instances of the database are represented by a Bag-Of-Words (BOW) model. Then, a weighting scheme is used to weight the features for each instance. The original implementation of Wordification only explored the TF-IDF, the term-frequency and the binary weighting schemes. However, works in the text classification and data mining fields show that the proper choice of weighting schemes can boost classification. Therefore, we empirically experimented different term weighting approaches with Wordification. Our results show that the right combination of weighting scheme and classification algorithm can significantly improve classification performance.

In Dynamic Classifier Selection (DCS) techniques, test sample is classified only by the most competent classifiers. Hence, the major problem in DCS is to find the measures by which competence of classifiers in a pool can be calculated to find out the most competent classifiers. To tackle these issues, we suggest a Framework for Dynamic Ensemble Selection (DES) that uses more than one criterion to calculate the base classifier’s competence level. The framework has three major steps. In first step, training data is used to create a pool consisting of different classifiers. In second step meta-classifier training is performed by extracting meta-features from training data. In third step meta-classifier uses meta-features extracted from test sample to perform an ensemble selection and to predict the final output. In this paper, we suggest some improvements in second step (training) and last step (generalization) of the framework. In training phase, four different models are used as meta-classifiers. While in generalization phase, dynamic weighting scheme is used where meta-classifiers will dynamically assign weights to selected competent classifiers based on their competence level and final decision will be aggregated using a weighting voting scheme. The modifications purposed in this paper altogether enhance performance and accuracy of the framework in contrast with other dynamic selection techniques in literature.

In this paper we suggest an algorithm for solving a multiobjective stochastic linear programming problem with normal multivariate distributions. The problem is first transformed into a deterministic multiobjective problem introducing the expected value criterion and an utility function. The obtained problem is reduced to a monobjective quadratic problem using a weighting method. This last problem is solved by DC algorithm.

We propose a logistic regression model combined with a weighting estimation procedure that incorporates a tuning parameter. We analyse predictive performance indicators. Results show that the parameter defining the weights can be used to improve predictive accuracy, at least when the original predictive value is distant from the response average. We use a publicly available data set to illustrate our method and we discuss the potential benefits of this methodology in the decision to purchase full coverage motor insurance versus a basic insurance product.

Class imbalance has been proven to seriously hinder the precision of many standard learning algorithms. To solve this problem, a number of methods have been proposed, for example, the distance-based balancing ensemble method that learns the unbalanced dataset by converting it into multiple balanced subsets on which sub-classifiers are built. However, the class-imbalance problem is usually accompanied by other data-complexity problems such as class overlap, small disjuncts, and noise instance. Current algorithms developed for primary unbalanced-data problems cannot address the complex-data problems at the same time. Some of these algorithms even exacerbate the class-overlap and small-disjuncts problems after trying to address the complex-data problem. On this account, this study proposes a dynamic ensemble selection decision-making (DESD) method. The DESD first repeats the random-splitting technique to divide the dataset into multiple balanced subsets that contain no or few class-overlap and small-disjunct problems. Then, the classifiers are built on these subsets to compose the candidate classifier pool. To select the most appropriate classifiers from the candidate classifier pool for the classification of each query instance, we use a weighting mechanism to highlight the competence of classifiers that are more powerful in classifying minority instances belonging to the local region in which the query instance is located. Tests with 15 standard datasets from public repositories are performed to demonstrate the effectiveness of the DESD method. The results show that the precision of the DESD method outperforms other ensemble methods.

[EN] The kNN algorithm has three main advantages that make it appealing to the community: it is easy to understand, it regularly offers competitive performance and its structure can be easily tuning to adapting to the needs of researchers to achieve better results. One of the variations is weighting the instances based on their distance. In this paper we propose a weighting based on the Newton's gravitational force, so that a mass (or relevance) has to be assigned to each instance. We evaluated this idea in the kNN context over 13 benchmark data sets used for binary and multi-class classification experiments. Results in F1 score, statistically validated, suggest that our proposal outperforms the original version of kNN and is statistically competitive with the distance weighted kNN version as well., This research was partially supported by CONACYT-Mexico (project FC-2410). The work of Paolo Rosso has been partially funded by the SomEMBED TIN2015-71147-C2-1-P MINECO research project.

Registration of pre-operative 3-D volumes to intra-operative 2-D X-ray images is important in minimally invasive medical procedures. Rigid registration can be performed by estimating a global rigid motion that optimizes the alignment of local correspondences. However, inaccurate correspondences challenge the registration performance. To minimize their influence, we estimate optimal weights for correspondences using PointNet. We train the network directly with the criterion to minimize the registration error. We propose an objective function which includes point-to-plane correspondence-based motion estimation and projection error computation, thereby enabling the learning of a weighting strategy that optimally fits the underlying formulation of the registration task in an end-to-end fashion. For single-vertebra registration, we achieve an accuracy of \(0.74\pm 0.26\) mm and highly improved robustness. The success rate is increased from 79.3% to 94.3% and the capture range from 3 mm to 13 mm.

Analyzing the travel behaviors and patterns of air passengers have always been of great significance to the air travel industry. Understanding the demands and interests of passengers behind their behaviors is a crucial and fundamental task for many applications. However, this task is challenging due to the lack of customer information, data sparsity, and the long-tail distribution. In this paper, we investigate the problem of heterogeneous item recommendation by learning representations of items and passengers in a shared latent space. Specifically, we first establish a heterogeneous information network (HIN) through statistical analysis, where the edges represent the interactions between different nodes. Each node also contains some auxiliary attribute information that describes its travel behavior or that of its passenger groups. Then we devise a joint matrix factorization model to learn node representations based on the HIN, where both the heterogeneous edges and the node attributes are incorporated into the learning process. Moreover, a weighting strategy is further utilized to deal with the long-tail distribution of passenger behaviors based on the implicit feedback information. Experimental results conducted on a real-world passenger name record (PNR) dataset demonstrate the effectiveness of the proposed method.

Discovering social communities of web users through clustering analysis of heterogeneous link associations has drawn much attention. However, existing approaches typically require the number of clusters a priori, do not address the weighting problem for fusing heterogeneous types of links, and have a heavy computational cost. This chapter studies the commonly used social links of users and explores the feasibility of the proposed heterogeneous data co-clustering algorithm GHF-ART, as introduced in Sect. 3.6, for discovering user communities in social networks. Contrary to the existing algorithms proposed for this task, GHF-ART performs real-time matching of patterns and one-pass learning, which guarantees its low computational cost. With a vigilance parameter to restrain the intra-cluster similarity , GHF-ART does not need the number of clusters a priori. To achieve a better fusion of multiple types of links, GHF-ART employs a weighting algorithm, called robustness measure (RM) , to incrementally assess the importance of all the feature channels for the representation of data objects of the same class. Extensive experiments have been conducted on two social network datasets to analyze the performance of GHF-ART. The promising results compare GHF-ART with existing methods and demonstrate the effectiveness and efficiency of GHF-ART. The content of this chapter is summarized and extended from [11] (Copyright ©2014 Society for Industrial and Applied Mathematics. Reprinted with permission. All rights reserved).

In this exploratory article, we draw attention to the common formal ground among various estimators such as the belief functions of evidence theory and their relatives, approximation quality of rough set theory, and contextual probability. The unifying concept will be a general filter function composed of a basic probability and a weighting which varies according to the problem at hand. To compare the various filter functions we conclude with a simulation study with an example from the area of item response theory.

Recently, the link prediction (LP) problem has attracted much attention from both scientific and industrial communities. This problem tries to predict whether two not linked nodes in a network will connect in the future. Several studies have been proposed to solve it. Some of them compute a compatibility degree (link strength) between connected nodes and apply similarity metrics between non-connected nodes in order to identify potential links. However, despite the acknowledged importance of temporal data for the LP problem, few initiatives investigated the use of this kind of information to represent link strength. In this paper, we propose a weighting criterion that combines the frequency of interactions and temporal information about them in order to define the link strength between pairs of connected nodes. The results of our experiment with weighted and non-weighted similarity metrics in ten co-authorship networks present statistical evidences that confirm our hypothesis that weighting links based on temporal information may, in fact, improve link prediction.

In this paper, we propose a new Light Field Image (LFI) compression scheme based on Multiview Video plus Depth (MVD) coding architecture. Through LF function analysis, we preliminarily estimate depth map according to the concept of Epipolar Plane Image (EPI). Such a rough estimation causes some error pixels within initial depth map, so we design a weighting mean filter to smooth the inaccurate region. The final estimated depth maps can be encoded by MVD coding standard jointly with a small number of viewpoint images in LFI, so as to improve compression efficiency of LFI. Ultimately, massive experiments are conducted on 4 LFIs to verify the effectiveness of the proposed compression scheme. The simulated results demonstrate that our LFI compression scheme can achieve a high LFI compression performance and outperform the-state-of-art coding solution.

We study the problem of tracking or disturbance rejection for signals beyond the Nyquist frequency in the sampled-data context. Given the well-established sampling theorem, one is inclined to infer that it is possible to track or reject only those signals below the Nyquist frequency. However, such a high-frequency signal may be detected as an aliased signal, and this opens a new possibility in tracking and regulation. This article shows that a multirate processing with a suitable choice of a weighting function enables tracking or rejection of high-frequency signals beyond the Nyquist frequency. An example is given to illustrate the result.

Evaluating the similarity between two objects in heterogeneous information network is a significant part of information science. The existing meta-structure based similarity measures only consider one meta-structure, which leads to a loss of accuracy. Based on the meta-structure, this paper proposes a weighted method to tackle the problem. We put forward a weighting algorithm that determines the value of weight to each meta-structure according to the set of the user’s preferences, and to compute the similarity value, we convert meta-structure into meta-path and use a novel meta-path based similarity measure StruSim. The top-k similarity research experiment is conducted to prove the effectiveness of the novel method. Using the measure nDCG, we conclude that StruSim performs better than PathSim, HeteSim, and AvgSim. And the multiple meta-structure methods are better than BSCSE and unweighted meta-path based methods. At last, we propose an interpolation and derivation method to search the optimal bias factor in StruSim to achieve a better performance.

A series of strategies for designing support structure to fabricate overhanging features in additive manufacturing (AM) is proposed. The focus of this study is to maximize the external support stiffness while requiring supports that can be easily removed, minimizing the time required to add the supports, and using the least amount of material. These design requirements become necessities not only for adding supports to improve processing and post-processing efficiency, but also for reducing the cost of supports while maintaining specimen geometry. A repulsion index (RI) is proposed for satisfying the easy removal requirement and minimizing the size of artifacts left on the specimen surface; a weighting function is applied to quantify the time consumption to build the supports. The proposed RI and cost due to additional material and time consumption in adding the support are formulated within a multi-objective topological optimization constructed by the simple isotropic material with penalization method, continuous approximation of material distribution, and method of moving asymptotes. Numerical simulations demonstrated that rational and cost effective support layouts can be determined by the proposed cost-based formulation. This allows designers to find design solutions with a compromise between the deformation and the cost of support structure.

In this paper the issue of continuous systems estimation, insensitive to certain perturbations, is presented and discussed. Such an approach has rational advantages, especially when robust schemes are used to assist a target system responsible for industrial diagnostics. This requires that estimated model parameters are generated on-line, and their values are reliable and to a great extent accurate. Practical hints are suggested to challenge the consistency problem of estimates. Namely, the technique of instrumental variables can improve the asymptotic behavior of estimators. With a weighting mechanism, in turn, tracking the time-varying parameters of non-stationary processes is realistic. Yet, evident insensitivity to destructive outliers in the measurement data is guaranteed by the applied estimation routine in the sense of the least sum of absolute errors. Finally, premises for a proper selection of persistently exciting input signals, as well as the directions of further research are summarized in the paper.

With big data-technologies on the rise, new fields of application appear in terms of analyzing data to find new relationships for improving process understanding and stability. Manufacturing companies oftentimes cope with a high number of deviations but struggle to solve them with less effort. The research project BigPro aims to develop a methodology for implementing counter measures to disturbances and deviations derived from big data. This paper proposes a methodology for practitioners to assess predefined counter measures. It consists of a morphology with several criterions that can have a certain characteristic. Those are then combined with a weighting factor to assess the feasibility of the counter measure for prioritization.

The description of Cumulative Prospect Theory (CPT) includes three important parts: a value function over outcomes, v(⋅ ); a weighting function over cumulative probabilities, w(⋅ ); CPT-utility as unconditional expectation of the value function v under probability distortion w. In this paper we consider the problem of choosing an CPT-investor’s portfolio in the case of complete market. The problem of finding the optimal portfolio for CPT-investor is to maximize the unconditional expectation of the value function v under probability distortion w over terminal consumption, subject to budget constraint on initial wealth. We find the optimal payoffs for CPT-investor for the classic Black-Scholes environment assuming that there are a single lognormally distributed stock and a risk free bond. We compare the optimal payoffs of CPT-investor with the optimal payoffs of the investor that maximizes expected power utility over terminal payoffs, subject to budget constraint on initial wealth.

This chapter introduces two well-studied and useful functions: the ordered weighted averaging (OWA) operator and the discrete Choquet integral. The OWA is defined with respect to a weighting vector (similarly to the weighted arithmetic mean), however the weights are allocated to inputs based on their relative size. The discrete Choquet integral includes the OWA as a special case, however in general it is used to aggregate data when independence of the inputs is violated or cannot be assumed, i.e. some of the variables are correlated or interact with one another. We give an overview of these functions along with their properties and some indices used to interpret their behavior. As with the previous chapters, we include worked examples, questions and a guide on how to implement and use these functions in the R programming language.

Salient object detection in video has attracted enormous research efforts for its wide applicability. But there are still some issues in restraining the disturbance of background, which make it difficult to detect salient object in complex scenarios. Inspired by the hypothesis of center prior in image domain, we novelly introduced the concept of dynamic attention center in video. The distance between specific regions and this center is used as a weighting term to restrain the influence of background disturbance and obtain more accurate spatial and temporal saliency maps. Besides, we develop a dynamic fusion method to combine the temporal and spatial saliency map, leading to higher spatiotemporal consistency. The experiments on Freiburg-Berkeley Motion Segmentation Dataset show that our method outperforms several state-of-art methods on subjective visual perception and objective measurements.

Statistical shape models (SSMs) play an important role in medical image analysis. A sufficiently large number of high quality datasets is needed in order to create a SSM containing all possible shape variations. However, the available datasets may contain corrupted or missing data due to the fact that clinical images are often captured incompletely or contain artifacts. In this work, we propose a weighted Robust Principal Component Analysis (WRPCA) method to create SSMs from incomplete or corrupted datasets. In particular, we introduce a weighting scheme into the conventional Robust Principal Component Analysis (RPCA) algorithm in order to discriminate unusable data from meaningful ones in the decomposition of the training data matrix more accurately. For evaluation, the proposed WRPCA is compared with conventional RPCA on both corrupted (63 CT datasets of the liver) and incomplete datasets (15 MRI datasets of the human foot). The results show a significant improvement in terms of reconstruction accuracy on both datasets.

In this work, a new lattice L determined by the class of weighting triangles as a base of L-fuzzy subsets is proposed. Furthermore, extended orders and operators which are obtained by means of fuzzy binary relations \(F_{\triangle }\) associated to a weighting triangle are included. Moreover, some new expressions have been defined for Extended Ordered Weighted Averaging operators, and Extended Aggregation functions.

A novel Weighted Unscented Kalman Filtering method is introduced for neural activity estimation from electroencephalographic signals. The introduction of a weighting stage improves the solution by extracting relevant information directly from the measured data. Besides, a discrete nonlinear state space model representing the brain neural activity is used as a physiological constraint in order to improve the estimation. Moreover, time-varying parameters are considered which allow describing adequately healthy and pathological activity even for localized epilepsy events. Performance of the new method is evaluated in terms of introduced error measurements by application to simulated EEG data over several noise conditions. As a result, a considerable improvement over linear estimation approaches is found.

In this paper, we present an extension of k nearest neighbors method so it can perform imputation/classification from datasets with low quality data. The method performs a weighting of neighbors based on their imperfection and distance of classes. Thus the method allows us explicitly to indicate the average degree of imperfection of the neighbors that it is accepted to carry out the imputation/classification and the average distance of classes to the class of example to impute/classify that it is allowed. We carry out several experiments with both real-world and synthetic datasets with low quality data to test the proposed method.

The assessment of the weights of objective function plays an important role in a multi-objective process. This paper discusses a weighting method for linear fractional programming to solve possibilistic programming of the multi-objective decision-making problem. The minimal and maximal values of the objective function are utilized in the determination the weight value. This analysis concludes that it is worthwhile to pursue proposed solution approach to the multi-objective evaluation scheme, which addresses some limitation to determine the weight values.

The online multi-instance learning tracking (MIL) algorithm is known for its ability of alleviating tracking drift by training classifiers with positive and negative bag. However, the increased computational complexity results in time consuming due to the lack of consideration of sampling importance when collecting training samples. Additionally, the MIL method, as a 2D feature-based tracking algorithm, performs unsteadily when the object changes poses or rotates seriously. In this paper, a histogram-based feature similarity measurement is employed as a weighting strategy to select positive samples. Benefited from profitable depth information, the tracking algorithm we proposed achieves higher tracking performance. For computational efficiency, a compressive sensing method is adopted to extract features and reduce dimensionality. Experimental results demonstrate that our algorithm is better in robustness, accuracy, efficiency than three state-of-the-art methods on challenging video sequences.

Although several image classification methods have been proposed in the literature, most of them require an offline training phase to generate appropriate classifiers. In this paper, we show how combining MPEG-7 color descriptors can achieve an accurately image classification while bypassing the training phase. More specifically, we illustrate how the combination of Compact Color, Dominant Color and Color Layout was used to extract Global Similarity feature for input into ascending hierarchical classification. We introduce a weighting process to compute the global similarity from descriptors. Our experimental evaluation of the proposed classification method shows its significant effectiveness and high reliability.

Evaluating the perceived audiovisual quality of entire video calls requires to take into account the effects of time-varying degradations. Models of time integration have been developed which assess the fluctuating quality of transmitted speech and predict the call-final judgement based on a weighting technique of momentary quality ratings. In this chapter, the applicability of such models to the audiovisual case will not only be evaluated, but the models will be improved as well. To that end, a subjective test methodology is developed which allows the user perception evaluation of time-varying quality of 90 s long sequences organized in a simulated conversational structure. These sequences emulate the course of an entire video call and are composed of independent short samples (9 s). Audiovisual impairments are temporally distributed over the long sequences by applying different levels of impairment to the short samples, hence producing different predefined temporal quality profiles. A separate assessment of the perceived quality of the short samples on one hand and of the long sequences on the other hand allows to compare the perceptual effects between speech and audiovisual stimuli. It also allows the evaluation and optimization of existing call quality models that predict the quality at the end of a (simulated) speech conversation. These models proved to enhance the prediction accuracy in comparison to the plain average, and an optimization of the model parameters further refines the correlation of the estimates with the subjective data. The optimized models showed a higher correlation and a lower prediction error on independent test data.

In smart environments, developers can choose from a large variety of sensors supporting their use case that have specific advantages or disadvantages. In this work we present a benchmarking model that allows estimating the utility of a sensor technology for a use case by calculating a single score, based on a weighting factor for applications and a set of sensor features. This set takes into account the complexity of smart environment systems that are comprised of multiple subsystems and applied in non-static environments. We show how the model can be used to find a suitable sensor for a use case and the inverse option to find suitable use cases for a given set of sensors. Additionally, extensions are presented that normalize differently rated systems and compensate for central tendency bias. The model is verified by estimating technology popularity using a frequency analysis of associated search terms in two scientific databases.

This paper develops a novel approach to find the plane in a 3D fetal ultrasound scan which corresponds to the 2D diagnostic plane used in cranial ultrasound of a neonate to allow image-based biomarkers to be tracked from pre-birth through the first weeks of post-birth life. We propose a method based on regression forests (RF) with important algorithm design considerations taken into account to provide an accurate plane-finding solution. Specifically, the new method constrains the RF method by 1) using informative voxels and voxel informative strength as a weighting within the training stage objective function u, and 2) introducing regularization of the RF by proposing a geometrical feature within the training stage. Results on clinical data indicate that the new automated method is more reproducible than manual plane finding.

Adaptive filters are now becoming increasingly studied for their suitability in application to complex and non-stationary signals. Many adaptive filters utilise a reference input, that is used to form an estimate of the noise in the target signal. In this paper we discuss the application of adaptive filters for high electromyography contaminated electroencephalography data. We propose the use of multiple referential inputs instead of the traditional single input. These references are formed using multiple EMG sensors during an EEG experiment, each reference input is processed and ordered through firstly determining the Pearson’s r-squared correlation coefficient, from this a weighting metric is determined and used to scale and order the reference channels according to the paradigm shown in this paper. This paper presents the use and application of the Adaptive-Multi-Reference (AMR) Least Means Square adaptive filter in the domain of electroencephalograph signal acquisition.