Neurocomputing最新文献

While addressing the problem of imbalanced data classification, most existing resampling methods primarily focus on balancing class distribution. However, they often overlook class overlap and fail to adequately consider the feature distributions of different classes. Consequently, when resampling is performed under such conditions, samples within areas of overlap remain susceptible to misclassification, failing to substantially improve overall performance. To address these shortcomings, we propose a novel data resampling technique, Nearest Neighbors and Density-based Undersampling (NDU). This method employs within-class k-nearest neighbors and between-class probability densities to design a weight assignment strategy. Leveraging this strategy, we establish an exclusive metric, the F_factor, to evaluate the importance of majority class samples in overlap areas. Subsequently, NDU promotes a gradient-based segmented undersampling strategy, which applies varying degrees of undersampling to majority class samples across segmented regions. Through experiments on binary imbalanced datasets with class overlap, we evaluate the efficiency of diverse resampling methods concerning classification performance. The results demonstrate that our proposed method effectively addresses class overlap challenges.

This paper is concerned with the distributed fusion filtering algorithm design problem for multi-sensor nonlinear networked systems (MSNNSs) subject to multiplicative noises, random varying parameter matrix and missing measurements (MMs). In particular, we utilize the Bernoulli random variable with certain statistical features to describe and characterize the MMs phenomenon. By introduce a fictitious noise, the effects from process noise as well as random varying parameter matrix are addressed and a new nonlinear stochastic networked system is obtained. The primary purpose of this paper is to develop a novel fusion filtering scheme of the distributed way and provide the corresponding boundedness evaluation criterion. Firstly, specific upper bounds of filtering error covariance (FEC) are identified and locally minimized at each sampling instant. Subsequently, based on the obtained local filters, a distributed fusion filtering algorithm is designed via adopting the inverse covariance intersection (ICI) fusion idea. Furthermore, the analysis with respect to the upper bound of local FEC is discussed and examined by proposing a sufficient condition under certain constraints regarding the related parameters. Eventually, with the help of the simulation experiments, the usefulness of the proposed fusion filtering algorithm is illustrated.

This paper introduces a novel model-agnostic algorithm called adaptive ensemble batch multi-input multi-output conformalized quantile regression (AEnbMIMOCQR) that enables forecasters to generate multi-step ahead prediction intervals for a fixed pre-specified miscoverage rate $\alpha$ in a distribution-free manner. Our method is grounded on conformal prediction principles, however, it does not require data splitting and provides close to exact coverage even when the data is not exchangeable. Moreover, the resulting prediction intervals, besides being empirically valid along the forecast horizon, do not neglect heteroscedasticity. AEnbMIMOCQR is designed to be robust to distribution shifts, which means that its prediction intervals remain reliable over an unlimited period of time, without entailing retraining or imposing unrealistic strict assumptions on the data-generating process. Through methodically experimentation, we demonstrate that our approach outperforms other competitive methods on both real-world and synthetic datasets. The code used in the experimental part and a tutorial on how to use AEnbMIMOCQR can be found at the following GitHub repository: https://github.com/Quilograma/AEnbMIMOCQR.

Path planning is one of the most crucial elements in the field of robotics, such as autonomous driving, minimally invasive surgery and logistics distribution. This review begins by summarizing the limitations of conventional path planning methods and recent work on DRL-based path planning methods. Subsequently, the paper systematically reviews the construction of key elements of DRL methods in recent work, with the aim of assisting readers in comprehending the foundation of DRL research, along with the underlying logic and considerations from a practical perspective. Facing issues of sparse rewards and the exploration–exploitation balance during the practical training process, the paper reviews enhancement methods for training efficiency and optimization results in DRL path planning. In the end, the paper summarizes the current research limitations and challenges in practical path planning applications, followed by future research directions.

An Undirected Weighted Network (UWN) can be precisely quantified as an adjacency matrix whose inherent characteristics are fully considered in a Symmetric Nonnegative Latent Factor (SNLF) model for its good representation accuracy. However, an SNLF model uses a sole latent factor matrix to precisely describe the topological characteristic of a UWN, i.e., symmetry, thereby impairing its representation learning ability. Aiming at addressing this issue, this paper proposes an Alternating nonnegative least squares-incorporated Regularized Symmetric Latent factor analysis (ARSL) model. First of all, equation constraints composed of multiple matrices are built in its learning objective for well describing the symmetry of a UWN. Note that it adopts an L₂-norm-based regularization scheme to relax such constraints for making such a symmetry-aware learning objective solvable. Then, it designs an alternating nonnegative least squares-incorporated algorithm for optimizing its parameters efficiently. Empirical studies on four UWNs demonstrate that an ARSL model outperforms the state-of-the-art models in terms of representation accuracy, as well as achieves promising computational efficiency.

In this paper, we present a novel network-based approach, namely Inherently Non-negative Latent Feature Analysis for Diabetes Mellitus Comorbidity Detection (INDM), to enhance the detection and analysis of comorbidities associated with diabetes mellitus. Different from existing methods, INDM is the first computational approach that integrates comorbidity networks of the chronic disease spectrum with patient clinical characteristics. To perform the analytical tasks, the proposed INDM adopts the following core components. First, comorbidity networks representing patients diagnosed solely with hypertension and those with hypertension and diabetes are constructed, following the case-control design that establishes a 1:1 matching in age and gender between two cohorts. Subsequently, the disease set is modeled in the comorbidity network according to the relative risk methodology. This enables nodes and edges in the comorbidity network to represent disease interactions that are derived from the patient-disease bipartite graph. Second, a nonlinear loss function with the capability of inherently non-negative latent feature analysis followed by a comorbidity classifier is adopted to uncover the patterns indicating the diabetes comorbidity in the comorbidity network. The proposed INDM has been rigorously tested on actual diabetes comorbidity datasets. The notable results demonstrate that INDM exhibits superior detection accuracy. Furthermore, the topological structure discovered by the proposed INDM can provide a profound insight into hypertension comorbidity in both the case and control groups.

Existing crowd counting methods are mainly trained and tested in similar scenarios. When the testing and training scenarios of the model are different, the counting accuracy of these methods will sharply decrease, which seriously limits their practical application. To address this problem, we propose a multistage gated fusion network (MGFNet) for cross-scene crowd counting. MGFNet is primarily composed of dynamic gated convolution units (DGCU) and multilevel scale attention blocks (MSAB) modules. Specifically, DGCU uses a dynamic gating path to supplement detailed information to reduce the loss of crowd information and overestimation of background in different scenarios. MSAB calibrates crowd information at different scales and perspectives in different scenes by generating attention maps with discriminative information. In addition, we used a new global local consistency loss to optimize the model to adapt to changes in crowd density and distribution. Extensive experiments on four different types of scene counting benchmarks show that the proposed MGFNet achieves superior cross-scene counting performance.

Recent advances show that Transformer-based models and object detection-based models play an indispensable role in VQA. However, object detection-based models have significant limitations due to their redundant and complex detection box generation process. In contrast, Visual and Language Pre-training (VLP) models can achieve better performance, but require high computing power. To this end, we present Weight-Sharing Hybrid Attention Network (WHAN), a lightweight Transformer-based VQA model. In WHAN, we replace the object detection network with Transformer encoder and use LoRA to solve the problem that the language model cannot adapt to interrogative sentences. We propose Weight-Sharing Hybrid Attention (WHA) module with parallel residual adapters, which can significantly reduce the trainable parameters of the model and we design DWA and BVA modules that can allow the model to perform attention operations from different scales. Experiments on VQA-v2, COCO-QA, GQA, and CLEVR datasets show that WHAN achieves competitive performance with far fewer trainable parameters.

Knowledge distillation (KD) enables a simple model (student model) to perform as a complex model (teacher model) by distilling the knowledge from a pre-trained teacher model. Existing soft-label distillation methods often use a fixed temperature value in the softmax function to prevent overconfidence in the distillation process. However, this approach can lead to the suppression of important ‘dark knowledge’ for non-target classes in difficult samples, while also over-smoothing the confidence values for easier samples. To address this issue, we propose a novel approach called difficulty level-based knowledge distillation (DLKD), which considers the difficulty level of each sample to distill refined knowledge with high or low confidence, depending on the sample’s complexity. Our method calculates the difficulty level based on the Euclidean distance between the teacher model’s predictions and the pruned teacher model’s predictions. Experimental results demonstrate that our DLKD method outperforms state-of-the-art methods on challenging samples, including those with noisy labels or augmented data, achieving superior results on CIFAR-100, FGVR, and ImageNet datasets for image classification.

Combining Convolutional Neural Network(CNN) and Transformer has become one of the mainstream methods for three-dimensional (3D) medical image segmentation. However, the complexity and diversity of target forms in 3D medical images require models to capture complex feature information for segmentation, resulting in an excessive number of parameters which are not conducive to training and deployment. Therefore, we have developed a lightweight 3D multi-target semantic segmentation model. In order to enhance contextual texture connections and reinforce the expression of detailed feature information, we designed a multi-scale and multi-angle feature interaction module to enhance feature representation by interacting multi-scale features from different perspectives. To address the issue of attention collapse in Transformers, leading to the neglect of other detailed feature learning, we utilized local features as dynamic parameters to interact with global features, dynamically grouping and learning critical features from global features, thereby enhancing the model's ability to learn detailed features. While ensuring the segmentation capability of the model, we aimed to keep the model lightweight, resulting in a total of 9.63 M parameters. Extensive experiments were conducted on public datasets ACDC and Brats2018, as well as a private dataset, Temporal Bone CT. The results indicate that our proposed model is more competitive compared to the latest techniques in 3D medical image segmentation.