Adaptive Weight Matrix Research Articles

A training strategy for enhancing prediction accuracy of high magnitude oceanic environmental factors based on deep learning model

Oceanic environmental factors have marked impacts on marine ecosystems and economic development. Variations in high-magnitude oceanic environmental factors often lead to marine disasters and ecological crises. Previous studies have demonstrated the success of deep learning in marine environmental prediction tasks; however, limited research has specifically targeted high-magnitude oceanic environmental factors. Here, we propose a deep-learning model training strategy that enhances the model's attention to high-magnitude oceanic environmental factors by incorporating an adaptive weight matrix, thereby improving prediction accuracy. We evaluated the proposed training strategy for two prediction tasks: time-series and physics-based predictions. Compared to the baseline model, the prediction error for high-magnitude oceanic environmental factors was reduced by up to 56.93 % using this training strategy, and the stability of the model performance in medium-to long-term forecasts was enhanced. In addition, the suitability of the strategy across diverse scenarios was examined by assessing its robustness and computational overhead. This study provides new insights for more accurate prediction of oceanic environmental factors.

Multimodal transformer with adaptive modality weighting for multimodal sentiment analysis

Multimodal Sentiment Analysis (MSA) constitutes a pivotal technology in the realm of multimedia research. The efficacy of MSA models largely hinges on the quality of multimodal fusion. Notably, when conveying information pertinent to specific tasks or applications, not all modalities hold equal importance. Previous research, however, has either disregarded the importance of modalities altogether or solely focused on the importance of linguistic and non-linguistic modalities while neglecting the importance between non-linguistic modalities. To facilitate effective multimodal information fusion based on the relative importance of modalities, a novel multimodal fusion mode named Multimodal Transformer with Adaptive Modality Weighting (MTAMW) is proposed in this paper. Specifically, we introduce a multimodal adaptive weight matrix that allocates appropriate weights to each modality based on its contribution to sentiment analysis. Furthermore, a multimodal attention mechanism is introduced, utilizing multiple Softmax functions to compute attention weights, thereby efficiently fusion multimodal information via a single-stream Transformer. By meticulously considering the relative importance of each modality during the fusion process, more effective multimodal information fusion is achievable. Extensive experiments on benchmark datasets show that it is superior to or comparable to state-of-the-art methods on MSA tasks. The codes for our experiments are available at https://github.com/Vamos66/MTAMW.

Data-Driven Multimodel Predictive Control for Multirate Sampled-Data Nonlinear Systems

This paper is concerned with the asynchronous control problem of multi-rate sampled-data nonlinear systems. To solve this problem, the data-driven multi-model predictive control (DMMPC) strategy is proposed. First, a multi-model predictive control structure is designed such that each state variable of the multi-rate sampled-data nonlinear system can be controlled synchronously at all sampling instants. In this structure, the fuzzy neural network (FNN) is introduced to build the multi-model. Then, the prediction outputs of each state variable at all sampling instants are obtained to provide control information for the controller. Specially, the objective function with adaptive weight matrix (AWM) is designed to reduce the influence of the prediction error caused by nonlinear fitting on control performance. Then, the optimal control laws are calculated to improve the control precision. Finally, the convergence and stability of DMMPC are proved in detail. The numerical example and industrial application reveal that the proposed DMMPC can obtain considerable control performance for the multi-rate sampled-data nonlinear systems. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —The asynchronous control problem of the multi-rate sampled-data nonlinear system (MRSNS) may degrade the operation performance of closed-loop system. In this paper, a data-driven multi-model predictive control (DMMPC) strategy is designed for MRSNS without mechanism model. The strategy mainly consists of three parts: First, a multi-model prediction structure based on fuzzy neural network (FNN) is established to obtain the prediction output of all variables at each sampling point. The parameters of FNNs are corrected at each sampling point to ensure prediction accuracy. Second, an objective function with an adaptive weight matrix (AWM) is designed to compute the control law, in which AWM is used to reduce the influence of prediction error on control performance. Third, the effectiveness of DMMPC is verified by a numerical example and an industrial application of the wastewater treatment process. The experimental results show that DMMPC can achieve satisfied operation performance in control accuracy.

Robust noise hybrid active contour model for infrared image segmentation using orientation column filters

Infrared (IR) image segmentation plays an important role in many applications of night vision, including pedestrian detection, security monitoring, etc. However, the precision is constrained by edge blur and noise interference from the original infrared imaging. In order to achieve robust segmentation results under noise interference, a hybrid active contour model for the segmentation of targets in images using local feature information and global information is proposed. Based on the concept of orientation columns in the primary visual cortex, orientation column filters are defined, which can effectively extract local feature information with noise robustness. Then a global term with noise robustness is defined, and the adaptive weight matrix is adopted to combine the two to construct a complete signed pressure force (SPF) function. Several experiments demonstrate that the proposed algorithm performs more accurately and robustly on noisy infrared images segmentation compared to typical algorithms.

Early Detection of Alzheimer's Disease Using Cognitive Features: A Voting-Based Ensemble Machine Learning Approach

Early detection of Alzheimer's disease (AD) is vital for adequate control. Machine learning techniques have gained much attraction due to their efficiency in predicting AD using cognitive tests. Ensemble machine learning models are helpful in improving the robustness of the learning system via combining multiple machine learning models. This article proposes a novel ensemble machine learning technique for the early detection of AD. First, a novel feature selection technique referred to as Neighborhood Component Analysis and Correlation-based Filtration (NCA-F) is proposed to select the vital cognitive features from a given dataset. Second, various machine learning classifiers were trained using the proposed NCA-F method. The top classifiers were selected for voting based on the performance results. The voting is performed using an adaptive weight matrix process. The output label of a model is multiplied by the F1 score and represented as weight. The results revealed an accuracy of 93.92% when using adaptive voting, which is better than the accuracy of 90.53% observed when using the traditional artificial neural network method. The proposed technique improved the accuracy of detecting AD at an early stage. Furthermore, the results against a recent study using same features also revealed an improvement of 12.12% in accuracy.

Dynamical Fusion Model With Joint Variational and Deep Priors for Hyperspectral Image Super-Resolution

In this paper, we propose a novel dynamic fusion model (DFM) with joint variational and deep priors for the task of hyperspectral image super-resolution (HISR). The given model can benefit from both the advantages of traditional modeling and deep learning methods, thus achieving significant improvements based on existing deep pre-trained models. Specifically, the given model mainly contains two new designed terms, i.e., the weighted spatial fidelity (WSF) term and the deep fusion (DF) term. The WSF term focuses on the spatial recovery of the low-resolution hyperspectral image through the high-resolution multispectral image without the knowledge of the spectral response matrix, thus the proposed DFM can be viewed as a semi-blind model for HISR. Moreover, the DF term relied upon deep fusion with a designed adaptive weight matrix, which can effectively inject the deep priors into the traditional minimization model. Besides, the proposed DFM can be quickly and effectively solved using the alternating direction method of multipliers. Experimental results on widely used datasets demonstrate the superiority of our approach compared with state-of-the-art HISR methods.

Combining random forest and graph wavenet for spatial-temporal data prediction

The prosperity of deep learning has revolutionized many machine learning tasks (such as image recognition, natural language processing, etc.). With the widespread use of autonomous sensor networks, the Internet of Things, and crowd sourcing to monitor real-world processes, the volume, diversity, and veracity of spatial-temporal data are expanding rapidly. However, traditional methods have their limitation in coping with spatial-temporal dependencies, which either incorporate too much data from weakly connected locations or ignore the relationships between those interrelated but geographically separated regions. In this paper, a novel deep learning model (termed RF-GWN) is proposed by combining Random Forest (RF) and Graph WaveNet (GWN). In RF-GWN, a new adaptive weight matrix is formulated by combining Variable Importance Measure (VIM) of RF with the long time series feature extraction ability of GWN in order to capture potential spatial dependencies and extract long-term dependencies from the input data. Furthermore, two experiments are conducted on two real-world datasets with the purpose of predicting traffic flow and groundwater level. Baseline models are implemented by Diffusion Convolutional Recurrent Neural Network (DCRNN), Spatial-Temporal GCN (ST-GCN), and GWN to verify the effectiveness of the RF-GWN. The Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) are selected as performance criteria. The results show that the proposed model can better capture the spatial-temporal relationships, the prediction performance on the METR-LA dataset is slightly improved, and the index of the prediction task on the PEMS-BAY dataset is significantly improved. These improvements are extended to the groundwater dataset, which can effectively improve the prediction accuracy. Thus, the applicability and effectiveness of the proposed model RF-GWN in both traffic flow and groundwater level prediction are demonstrated.

Adaptive weight matrix and phantom intensity learning for computed tomography of chemiluminescence.

Classic algebraic reconstruction technique (ART) for computed tomography requires pre-determined weights of the voxels for the projected pixel values to build the equations. However, such weights cannot be accurately obtained in the application of chemiluminescence measurements due to the high physical complexity and computation resources required. Moreover, streaks arise in the results from ART method especially with imperfect projections. In this study, we propose a semi-case-wise learning-based method named Weight Encode Reconstruction Network (WERNet) to co-learn the target phantom intensities and the adaptive weight matrix of the case without labeling the target voxel set and thus offers a more applicable solution for computed tomography problems. Both numerical and experimental validations were conducted to evaluate the algorithm. In the numerical test, with the help of gradient normalization, the WERNet reconstructed voxel set with a high accuracy and showed a higher capability of denoising compared to the classic ART methods. In the experimental test, WERNet produces comparable results to the ART method while having a better performance in avoiding the streaks. Furthermore, with the adaptive weight matrix, WERNet is not sensitive to the ensemble intensity of the projection which shows much better robustness than ART method.

Distributed tracking-ADMM approach for chance-constrained energy management with stochastic wind power in smart grid

This paper proposes a distributed strategy to the chance-constrained energy management for smart grid with penetration of stochastic wind power. The energy management model is constructed including chance constraints of spinning reserves for the sake of guaranteeing the maximum utilization of wind power on the basis of reliability. With the available wind power characterizing by Weibull distribution, the chance constraints can be converted into deterministic ones by the derived analytical form of inverse cumulative distribution function. Although the original problem is transformed into a typical convex optimization problem, the tight coupling of constraints presents challenges to the design of distributed strategy. As such, we formulate the problem into a compact form with each generator unit depending on individual decision variables, instead of the common form with decision vector being the collection of all local decision variables. Then, by developing a new initialization method and an adaptive weight matrix selection method, a distributed strategy based on tracking Alternating Direction Method of Multipliers (ADMM) is proposed to solve the model. The simulation results indicate that the proposed distributed strategy achieves comparable performance to the corresponding centralized scenario, and better performance than distributed consensus-based ADMM in the related literature. Moreover, the validity of the proposed distributed strategy is confirmed in one day chance-constrained energy management with stochastic wind power.

Fixed-time constrained acceleration reconstruction scheme for robotic exoskeleton via neural networks

Accurate acceleration acquisition is a critical issue in the robotic exoskeleton system, but it is difficult to directly obtain the acceleration via the existing sensing systems. The existing algorithm-based acceleration acquisition methods put more attention on finite-time convergence and disturbance suppression but ignore the error constraint and initial state irrelevant techniques. To this end, a novel radical bias function neural network (RBFNN) based fixed-time reconstruction scheme with error constraints is designed to realize high-performance acceleration estimation. In this scheme, a novel exponential-type barrier Lyapunov function is proposed to handle the error constraints. It also provides a unified and concise Lyapunov stability-proof template for constrained and non-constrained systems. Moreover, a fractional power sliding mode control law is designed to realize fixed-time convergence, where the convergence time is irrelevant to initial states or external disturbance, and depends only on the chosen parameters. To further enhance observer robustness, an RBFNN with the adaptive weight matrix is proposed to approximate and attenuate the completely unknown disturbances. Numerical simulation and human subject experimental results validate the unique properties and practical robustness.

A Level Set Method for Infrared Image Segmentation Using Global and Local Information

Infrared image segmentation plays a significant role in many burgeoning applications of remote sensing, such as environmental monitoring, traffic surveillance, air navigation and so on. However, the precision is limited due to the blurred edge, low contrast and intensity inhomogeneity caused by infrared imaging. To overcome these challenges, a level set method using global and local information is proposed in this paper. In our method, a hybrid signed pressure function is constructed by fusing a global term and a local term adaptively. The global term is represented by the global average intensity, which effectively accelerates the evolution when the evolving curve is far away from the object. The local term is represented by a multi-feature-based signed driving force, which accurately guides the curve to approach the real boundary when it is near the object. Then, the two terms are integrated via an adaptive weight matrix calculated based on the range value of each pixel. Under the framework of geodesic active contour model, a new level set formula is obtained by substituting the proposed signed pressure function for the edge stopping function. In addition, a Gaussian convolution is applied to regularize the level set function for the purpose of avoiding the computationally expensive re-initialization. By iteration, the object of interest can be segmented when the level set function converges. Both qualitative and quantitative experiments verify that our method outperforms other state-of-the-art level set methods in terms of accuracy and robustness with the initial contour being set randomly.

Cost Sensitive Face Security Verification Based on Limited Expression-Pose Pattern Sparse Representation Model

Face security verification has been recognized as a cost sensitive classification problem. To deal with this problem, many cost sensitive classifiers have been proposed to alleviate the facial variation. However, no suficient attention is paid to the research on sparse representation cost sensitive face verification. In this paper, we proposed a coarse to find face security verification method, called cost sensitive face verification based on limited expression-pose pattern (CSFV_LEP) for security verification task. The main contributions of the proposed method are as follows: (1) a discrimination dictionary is established in a common way to discriminate whether visitor is an internal member; (2) a confirmation dictionary, which contains only limited expression-pose details, is used to confirm whether this is the correct classification. Meanwhile, we use adaptive weight matrix from similarity information to enhance the robustness of the two dictionaries. Experiments show that, the proposed method has high verifiable and ideal secure performance according to accuracy and efficiency, and contribute to reduce cost penalization during the sparse coding stages.

Adaptive weight matrix design and parameter estimation via sparse modeling for MIMO radar

Although sparse representation and sparse recovery algorithms for colocated multiple-input multiple-output (MIMO) radar have received much attention, incoherence of the sensing matrix received few discussions. In this paper, we propose adaptive weight matrix design and parameter estimation via sparse modeling to improve the recovery performance for MIMO radar. First, a sparse framework is formulated for the MIMO array with decoupled transmit weight matrix and steering matrix. Next, a two stage method is proposed to optimize the two matrices to improve the DOA estimation performance. Finally, a sparse recovery approach based on lq (0 < q ≤ 1) norm optimization is developed to estimate the unknown target parameters. Furthermore, the algorithm can be iteratively implemented with the obtained closed-form solution for the optimization problem. The angle-amplitude estimation performance is examined by analyzing the Cramér–Rao lower bound (CRLB). Numerical results demonstrate that significant performance improvement has been achieved by the proposed sensing matrix optimization and adaptive weight matrix design approach.

Adaptive fast consensus algorithm for distributed sensor fusion

In the past few years, the problem of distributed parameter estimation has received a lot of attention, particularly for the applications in sensor networks. This paper focuses on the distributed iterative parameter estimation scheme. An alternative form of the consensus averaging-based algorithm is introduced, in which each node iteratively updates its estimate by adding a weighted sum of its own and its neighbors’ estimate, with the time-varying weight matrices. To improve the convergence of this distributed iterative scheme, an adaptive weight matrix modification algorithm is proposed, in which each node is modeled as an adaptive filter. The weight matrix is modified using the steepest descent method. With the non-persistent measurement data, a linear predictor is designed to provide the reference signal in the adaptive filter. As a result, the intermediate estimate of each node is driven closer to the steady-state estimate. Simulation results show that the proposed algorithm decreases the intermediate estimation error and accelerates the consensus convergence, with the nearly optimal steady-state estimation performance.

Design of two-dimensional rectangular array beamformers with partial adaptivity

This paper considers the problem of designing a two-dimensional (2-D) rectangular array beamformer with partial adaptivity. Using the eigenstructure of the signal data received by a 2-D rectangular array beamformer, we first show that the optimal weight matrix when using full adaptivity can be obtained from a set of singular vectors. Then the design problem of using partial adaptivity is formulated. As a result, the optimal solution for the partially adaptive weight matrix can be found by solving two basic problems, namely determining the proper dimension of the partially adaptive weight matrix and the set of the singular vectors. We develop the detection formulas for the information theoretic criteria AIC and MDL to find the proper dimension. Next, an efficient method is presented so that the optimal solution for the set of the singular vectors can be found analytically. We also investigate the required computational complexity. It is shown that 2-D partially adaptive beamforming using the proposed technique requires less computational complexity than 2-D fully adaptive beamforming using conventional techniques. Moreover, computer simulations demonstrate that the proposed 2-D partially adaptive technique provides satisfactory array performance when compared with conventional fully adaptive techniques.