Large-scale multimodal contrastive learning has recently achieved impressive success in learning rich and transferable representations, yet it remains fundamentally limited by the uniform treatment of feature dimensions and the neglect of the intrinsic spectral structure of the learned features. Empirical evidence indicates that high-dimensional embeddings tend to collapse into narrow cones, concentrating task-relevant semantics in a small subspace, while the majority of dimensions remain occupied by noise and spurious correlations. Such spectral imbalance and entanglement undermine model generalization. We propose Spectral Disentanglement and Enhancement (SDE), a novel framework that bridges the gap between the geometry of the embedded spaces and their spectral properties. Our approach leverages singular value decomposition to adaptively partition feature dimensions into strong signals that capture task-critical semantics, weak signals that reflect ancillary correlations, and noise representing irrelevant perturbations. A curriculum-based spectral enhancement strategy is then applied, selectively amplifying informative components with theoretical guarantees on training stability. Building upon the enhanced features, we further introduce a dual-domain contrastive loss that jointly optimizes alignment in both the feature and spectral spaces, effectively integrating spectral regularization into the training process and encouraging richer, more robust representations. Extensive experiments on large-scale multimodal benchmarks demonstrate that SDE consistently improves representation robustness and generalization, outperforming state-of-the-art methods. SDE integrates seamlessly with existing contrastive pipelines, offering an effective solution for multimodal representation learning.

Traffic forecasting is crucial in the realm of intelligent urban planning, playing an essential role in route optimization, arrival time estimation, and the prevention of congestion-related incidents. Deep learning has catalyzed the development of advanced spatio-temporal graph neural networks (STGNNs) for traffic predictions. A key advantage of STGNNs is their capability to capture complex spatial correlations, thereby improving prediction accuracy. Nevertheless, the existing methodologies, including dynamical graph-based networks and attention-based architectures, tend to overlook regular traffic patterns, leading to overly complex neural networks with less interpretability and more computational costs, which could be a bottleneck for practical applications.To address the challenge, we propose a lightweight and user-friendly dynamic adaptive graph generator, termed the Vector-Quantized Graph Generator (VQGG), which can autonomously identify prevalent spatial patterns and construct the corresponding adaptive graphs efficiently for enhanced traffic forecasting. We have extensively tested the VQGG with two benchmark datasets by integrating it as a substitute for the adaptive graphs in baseline models with minor modifications. Extensive experimental results demonstrate that VQGG can reduce forecasting errors by approximately 2.59%, indicating a significant improvement in predictive performance.

Different from the common static and continuous-time dynamic problems of unconstrained/constrained nonlinear optimization, this article aims to investigate a discrete-time dynamic problem of nonlinear optimization with multiple types of constraints, which can be succinctly termed as future multiconstrained nonlinear optimization (FMCNO) problem because of the unknown future. Considering the unique advantages of neural networks with parallelism and fuzzy control systems (FCSs) with adaptivity, a fuzzy-enhanced robust discretized zeroing neural network (FER-DZNN) model is proposed to address the FMCNO problem. Specifically, by introducing a fuzzy factor outputted from an FCS with dual inputs, the FER-DZNN model is designed on the basis of an FER evolution rule and a five-step look-ahead discretization rule. Moreover, theoretical results are provided to indicate the convergence and robustness of the FER-DZNN model under various noises. Finally, two illustrative examples, including an application example to robotic manipulator control, are presented to substantiate the superior convergent and robust performance of the FER-DZNN model under various noises for addressing the FMCNO problem.

In this study, we present a novel approach for predicting interventions for patients in the intensive care unit using a multivariate time series graph convolutional neural network. Our method addresses two critical challenges: the need for timely and accurate decisions based on changing physiological signals, drug administration information, and static characteristics; and the need for interpretability in the decision-making process. Drawing on real-world ICU records from the MIMIC-III dataset, we demonstrate that our approach significantly improves upon existing machine learning and deep learning methods for predicting two targeted interventions, mechanical ventilation and vasopressors. Our model achieved an accuracy improvement from 81.6% to 91.9% and a F1 score improvement from 0.524 to 0.606 for predicting mechanical ventilation interventions. For predicting vasopressor interventions, our model achieved an accuracy improvement from 76.3% to 82.7% and a F1 score improvement from 0.509 to 0.619. We also assessed the interpretability by performing an adjacency matrix importance analysis, which revealed that our model uses clinically meaningful and appropriate features for prediction. This critical aspect can help clinicians gain insights into the underlying mechanisms of interventions, allowing them to make more informed and precise clinical decisions. Overall, our study represents a significant step forward in the development of decision support systems for ICU patient care, providing a powerful tool for improving clinical outcomes and enhancing patient safety.

Event detection in a city is crucial for the government to listen to the voice of the citizens, be aware of the real occurrences in a city, and then make wiser policies. However, in reality some important events with few samples are easily to be overwhelmed by the massive information and hard to be recognized, and additionally the limited word description from the short texts even makes the recognition harder. To address the problems, we propose a knowledge-aware event detector by incorporating the external knowledge to detect the events with few examples. The external knowledge incorporation with different semantic relations is capable to enrich the short texts. In addition, we leverage the representative few shot learning framework to formulate the event detection as the text classification problem. The proposed model is evaluated on two widely event-detection datasets. The experiments show a consistent accuracy improvement. The findings validates that our model with the knowledge infusion is effective to detect the few shot events from the short texts.

Session-Based Recommendation (SBR) is a spotlight research problem. Although many efforts have been made, challenges still exist. The key to unlocking this shackle is the user intention, an intuitive but hard-to-model concept in the anonymous session. Unlike previous research, we suggest mining potential user intention by counting the number of item occurrences in a user session and considering the long interval between item re-interactions. Beyond these, we take user preference, a biased user intention, into account in the prediction stage. Forming these together, we propose a model named user Intention aware Graph Neural Network (Int-GNN) aiming at capturing user intention. Extensive experiments have been conducted on three real-world datasets, and the results show the superiority of our method. The code is available on GitHub: https://github.com/xuguangning1218/IntGNN_ICASSP2023

Stance detection is an important task, which aims to classify the attitude of an opinionated text toward a given target. In this paper, we develop an interpretable neural production system for stance detection (NPS4SD). NPS4SD is an end-to-end deep learning model, which consists of a set of knowledge rules that are applied by binding with specific entities. NPS4SD consists of two main components: a pretrained model for learning the text representation and a variable binding network (VBN) to bind the knowledge rules with text entities. Extensive experiments are conducted to evaluate the effectiveness of the proposed NPS4SD model on three real-world datasets with in-domain, cross-target and zero-shot setups. Experimental results demonstrate that NPS4SD achieves substantially better performance than the strong competitors for the stance detection task.

This paper presents a collaborative control method based on payload-leading for the multi-quadrotor transportation systems. The goal is to keep the relative distance between the quadrotors and the payload as constant as possible during the transportation, so as to ensure the stable attitude of the payload. The control mechanism consists of a guidance control law that generates the common desired velocity for the quadrotors, an internal feedback controller for each quadrotor, and a decentralized formation controller. The stability of the control structure is proved by Lyapunov theory. Finally, the experimental platform of the multi-quadrotor transportation system is built to verify the effectiveness of the control method. Experimental results show that the proposed method has an excellent control effect.

To effectively diminish the kinetic energy dissipation, joint-angle drift, and joint-velocity discontinuity problems, and simultaneously achieve the end-effector position and direction control as well as the avoidance of joint-physical limits, a novel velocity layer weighted multicriteria optimization scheme is proposed, which outperforms the traditional schemes for the robotic manipulators with multiple constraints. Besides, considering that the existing joint-limit conversion strategies are not differentiable everywhere or with relatively complex formulation, a new exponential joint-limit conversion strategy is introduced to facilitate the dynamic quadratic programming reformulation of the proposed scheme. For easier numerical realization and real-time control, aided with a high-precision six-step extrapolated-backward discretization rule, a novel discretized zeroing neural network model is proposed to resolve the proposed scheme, which has higher precision than the existing neural network models. Finally, numerical and physical experiments are conducted to substantiate the efficacy, superiority, and practicability of the proposed scheme and model.

Equivalency is a powerful approach that can transform an original problem into another problem that is relatively more ready to be resolved. In recent years, Zhang neurodynamics equivalency (ZNE), in the form of neurodynamics or recurrent neural networks (RNNs), has been investigated, abstracted, and proposed as a process that can equivalently solve equations at different levels. After long-term research, we have noticed that the ZNE can not only work with equations, but also inequations. Thus, the ZNE of inequation type is proposed, proved, and applied in this study. The ZNE of inequation type can transform different-level bound constraints into unified-level bound constraints. Applications of the jerk-level ZNE of bound constraints, equation constraints, and objective indices ultimately build up effective time-varying quadratic-programming schemes for cyclic motion planning and control (CMPC) of single and dual robot-arm systems. In addition, as an effective time-varying quadratic-programming solver, a projection neural network (PNN) is introduced. Experimental results with single and dual robot-arm systems substantiate the correctness and efficacy of ZNE and especially the ZNE of inequation type. Comparisons with conventional methods also exhibit the superiorities of ZNE.

This research first proposes the general expression of Zhanget al.discretization (ZeaD) formulas to provide an effective general framework for finding various ZeaD formulas by the idea of high-order derivative simultaneous elimination. Then, to solve the problem of future equality-constrained nonlinear optimization (ECNO) with various noises, a specific ZeaD formula originating from the general ZeaD formula is further studied for the discretization of a noise-perturbed continuous-time advanced zeroing neurodynamic model. Subsequently, the resulting noise-perturbed discrete-time advanced zeroing neurodynamic (NP-DTAZN) algorithm is proposed for the real-time solution to the future ECNO problem with various noises suppressed simultaneously. Moreover, theoretical and numerical results are presented to show the convergence and precision of the proposed NP-DTAZN algorithm in the perturbation of various noises. Finally, comparative numerical and physical experiments based on a Kinova JACO2robot manipulator are conducted to further substantiate the efficacy, superiority, and practicability of the proposed NP-DTAZN algorithm for solving the future ECNO problem with various noises.

In this article, the new jerk-level configuration adjustment (JLCA) schemes are proposed to achieve the configuration adjustment of constrained redundant robots. Specifically, by applying the zeroing neurodynamics design rule three times, the new JLCA performance index is first derived; then, together with the joint physical constraints incorporated, the new JLCA schemes are obtained for dual-arm and single-arm redundant robots, respectively. For comparison purposes, three other configuration adjustment schemes are also presented. Moreover, the comparative simulative experiments based on a planar dual-arm redundant robot (i.e., five-link dual-arm robot) and a spatial single-arm redundant robot (i.e., Kinova JACO$^2$robot) are performed to verify the efficacy and superiority of the proposed JLCA schemes, as compared with the three other configuration adjustment schemes. At last, the comparative physical experiments are conducted on the real Kinova JACO$^2$robot to substantiate the practicability and excellent performance of the proposed JLCA scheme for single-arm redundant robots.

The abundant sequential documents such as online archival, social media, and news feeds are streamingly updated, where each chunk of documents is incorporated with smoothly evolving yet dependent topics. Such digital texts have attracted extensive research on dynamic topic modeling to infer hidden evolving topics and their temporal dependencies. However, most of the existing approaches focus on single-topic-thread evolution and ignore the fact that a current topic may be coupled with multiple relevant prior topics. In addition, these approaches also incur the intractable inference problem when inferring latent parameters, resulting in a high computational cost and performance degradation. In this work, we assume that a current topic evolves from all prior topics with corresponding coupling weights, forming the multi-topic-thread evolution . Our method models the dependencies between evolving topics and thoroughly encodes their complex multi-couplings across time steps. To conquer the intractable inference challenge, a new solution with a set of novel data augmentation techniques is proposed, which successfully discomposes the multi-couplings between evolving topics. A fully conjugate model is thus obtained to guarantee the effectiveness and efficiency of the inference technique. A novel Gibbs sampler with a backward–forward filter algorithm efficiently learns latent time-evolving parameters in a closed-form. In addition, the latent Indian Buffet Process compound distribution is exploited to automatically infer the overall topic number and customize the sparse topic proportions for each sequential document without bias. The proposed method is evaluated on both synthetic and real-world datasets against the competitive baselines, demonstrating its superiority over the baselines in terms of the low per-word perplexity, high coherent topics, and better document time prediction.

In this work, the LQ decomposition problem of dynamic matrix is investigated. First, by applying Zhang neural network (ZNN) method as well as the Kronecker-product and vectorization techniques, a ZNN model is proposed to solve the LQ decomposition problem of dynamic matrix. Then, two simulative examples are provided to verify the validity of the proposed ZNN model. Finally, an example of the mobile object localization based on the angle-of-arrival (AoA) technique is provided to illustrate the applicability of the proposed ZNN model.

Because of its extensive appearance and application in scientific research and industrial production, the matrix square root problem has received massive attention and study. In this paper, based on our previous work, by using zeroing neural dynamics (ZND) method, a continuous-time dynamic matrix square root (CTDMSR) model is given at first. Besides, a general ten-instant Zhang et al. discretization (ZeaD) formula is derived, constructed and investigated, and the corresponding theoretical analysis is provided. Next, by applying this general formula to discretize the CTDMSR model, a general ten-instant discrete-time dynamic matrix square root (DTDMSR) model with sixth-order precision is further obtained. For comparison purposes, four DTDMSR models, with the second-, third-, fourth-, and fifth-order precision, are also acquired and presented, respectively, by using other ZeaD formulas. At last, the effectiveness and correctness of the proposed DTDMSR models for dynamic matrix square root finding are further substantiated by numerical experimental results.

In this article, we derive the general explicit linear four-step (ELFS) rule with fifth-order precision systematically, together with a group of specific ELFS rules provided. Afterwards, we formulate and investigate a new and challenging discrete-time dynamic problem with relatively complex structure and future unknownness, which is simply termed future system of bounded inequalities and nonlinear equations (SBINE). With the aid of the general ELFS rule, the general ELFS-type discretized zeroing neural network (DZNN) model is proposed to solve the future SBINE. Moreover, theoretical and numerical results are presented to show the validity and high precision of the proposed general ELFS-type DZNN model. Finally, comparative numerical experiments based on a wheeled mobile robot containing several additional constraints are further performed to substantiate the applicability, validity, and superiority of the proposed general ELFS-type DZNN model.

The online event discovery in social media based documents is useful, such as for disaster recognition and intervention. However, the diverse events incrementally identified from social media streams remain accumulated, ad hoc, and unstructured. They cannot assist users in digesting the tremendous amount of information and finding their interested events. Further, most of the existing work is challenged by jointly identifying incremental events and dynamically organizing them in an adaptive hierarchy. To address these problems, this article proposes d ynamic and h ierarchical C ategorization M odeling (dhCM) for social media stream. Instead of manually dividing the timeframe, a multimodal event miner exploits a density estimation technique to continuously capture the temporal influence between documents and incrementally identify online events in textual, temporal, and spatial spaces. At the same time, an adaptive categorization hierarchy is formed to automatically organize the documents into proper categories at multiple levels of granularities. In a nonparametric manner, dhCM accommodates the increasing complexity of data streams with automatically growing the categorization hierarchy over adaptive growth. A sequential Monte Carlo algorithm is used for the online inference of the dhCM parameters. Extensive experiments show that dhCM outperforms the state-of-the-art models in terms of term coherence, category abstraction and specialization, hierarchical affinity, and event categorization and discovery accuracy.

Future quadratic programming (FQP) is an interesting and challenging topic due to its unknown future information and time-dependent feature. In this paper, a continuous-time zeroing neurodynamics (CTZN) model for quadratic programming is first obtained via zeroing neurodynamics (ZN) method. Then, a general four-step Zhang et al. discretization formula is presented and adopted to discretize the above CTZN model, and thus the general four-step discrete-time ZN (DTZN) algorithm for the FQP is developed. For comparison, a three-step DTZN algorithm and a one-step DTZN algorithm for the FQP are also presented. It is worth noting that there is an important parameter termed stepsize in the DTZN algorithms, which is closely related to their stability. If the value of stepsize is outside its effective interval, the DTZN algorithms are impossible to achieve convergence in terms of residual errors, which leads to failure of the FQP problem solving. By utilizing bilinear transformation and Routh stability criterion, the effective stepsize interval of the general four-step DTZN algorithm is confirmed via theoretical proof. Besides, numerical results substantiate the effectiveness and superiority of the general four-step DTZN algorithm as well as the accuracy of the effective stepsize interval. Finally, the general four-step DTZN algorithm is applied to fulfill the path-tracking control of different robot manipulators, with the effectiveness and superiority further validated.

In this work, time-dependent linear and nonlinear inequalities system (TDLNIS) is studied and solved. First, using zeroing neural network (ZNN) method twice, a continuous time-dependent ZNN (CTDZNN) model is proposed to solve the continuous TDLNIS. Subsequently, explicit linear dual-multistep methods, i.e., explicit linear dual-4-step, dual-3-step, and dual-2-step methods, are presented and studied. Afterwards, by applying the explicit linear dual-4-step method to the proposed CTDZNN model, a 4-step discrete time-dependent ZNN (4S-DTDZNN) model is proposed to solve the discrete TDLNIS. For comparison, 3-step discrete time-dependent ZNN (3S-DTDZNN) and 2-step discrete time-dependent ZNN (2S-DTDZNN) models are also developed for solving the discrete TDLNIS. In addition, theoretical analyses and results indicate the effectiveness and superiority of the proposed 4S-DTDZNN model. Finally, numerical experimental results further substantiate the effectiveness and superiority of the proposed 4S-DTDZNN model.

In this paper, a six-point discretization (6PD) formula is presented to discretize continuous-time models. Then, by using the 6PD formula and introducing the adaptivity/variability of parameter, a new-type discrete-time gradient (DTG) model is proposed to solve a discrete dynamic nonlinear equation system (DDNES) with occasionally-singular Jacobian matrix. For comparative purposes, based on the 6PD formula, a 6PD-type discrete-time zeroing (DTZ) model and an old-type (i.e., conventional) DTG model are also presented to handle the same problem. Finally, comparative numerical experiments, including an application to the discrete-time motion control of a robot manipulator, are conducted to substantiate the validity and superiority of the new-type DTG model for solving the DDNES with occasionally-singular Jacobian matrix. That is, when the Jacobian matrix of DDNES occasionally becomes singular as time evolves, the new-type DTG model can provide a feasible and effective solution to the singular Jacobian problem, whereas the other presented models fail to achieve such a solution.

In this article, a future different-layer nonlinear and linear equation system (DLNLES) is investigated. First, based on a zeroing neural network (ZNN) method, a zeroing equivalency theorem is proposed. Then, a continuous ZNN (CZNN) model is developed for continuous DLNLES solving. Next, a new eight-node Zhang et al. discretization formula is proposed to discretize the CZNN model, and thus, an eight-node discrete ZNN (DZNN) model is proposed for the future DLNLES solving. Five-node and four-node DZNN models are also developed for the same problem solving. Besides, numerical experiments are executed to substantiate the validity and superiority of the proposed eight-node DZNN model. Finally, the path-tracking control problem of a four-link redundant robot arm is formulated as a specific future DLNLES problem and can, thus, be solved by the three DZNN models. Comparative numerical results further indicate that the proposed eight-node DZNN model is much superior to the other two DZNN models.

The following topics are dealt with: power grids; invertors; voltage control; electric current control; control system synthesis; distributed power generation; power convertors; machine control; power generation control; switching convertors.

In this paper, we propose a novel online event discovery model DP-density to capture various events from the social media data. The proposed model can flexibly accommodate the incremental arriving of the social documents in an online manner by leveraging Dirichlet Process, and a density based technique is exploited to deduce the temporal dynamics of events. The spatial patterns of events are also incorporated in the model by a mixture of Gaussians. To remove the bias caused by the streaming process of the documents, Sequential Monte Carlo is used for the parameter inference. Our extensive experiments over two different real datasets show that the proposed model is capable to extract interpretable events effectively in terms of perplexity and coherence.

The availability of geographical and temporal tagged documents enables many location and time based mining tasks. Event discovery is one of such tasks, which is to identify interesting happenings in the geographical and temporal space. In recent years, several techniques have been proposed. However, no existing work has provided a nonparametric algorithm for detecting events in the joint space crossing geographical and temporal dimensions. Furthermore, though some prior works proposed to capture the periodicities of topics in their solutions, some restrictions on the temporal patterns are often placed and they usually ignore the spatial patterns of the topics. To break through such limitations, in this paper we propose a novel nonparametric model to identify events in the geographical and temporal space, where any recurrent patterns of events can be automatically captured. In our approach, parameters are automatically determined by exploiting a Dirichlet Process. To reduce the influence from noisy terms in the detection, we distinguish its event role from its background role using a Bernoulli model in the solution. Experimental results on three real world datasets show the proposed algorithm outperforms previous state-of-the-art approaches.