Inspired by advances in LLMs, reasoning-enhanced sequential recommendation performs multi-step deliberation before making final predictions, unlocking greater potential for capturing user preferences. However, current methods are constrained by static reasoning trajectories that are ill-suited for the diverse complexity of user behaviors. They suffer from two key limitations: (1) a static reasoning direction, which uses flat supervision signals misaligned with human-like hierarchical reasoning, and (2) a fixed reasoning depth, which inefficiently applies the same computational effort to all users, regardless of pattern complexity. These rigidity lead to suboptimal performance and significant computational waste. To overcome these challenges, we propose DTRec, a novel and effective framework that explores the Dynamic reasoning Trajectory for Sequential Recommendation along both direction and depth. To guide the direction, we develop Hierarchical Process Supervision (HPS), which provides coarse-to-fine supervisory signals to emulate the natural, progressive refinement of human cognitive processes. To optimize the depth, we introduce the Adaptive Reasoning Halting (ARH) mechanism that dynamically adjusts the number of reasoning steps by jointly monitoring three indicators. Extensive experiments on three real-world datasets demonstrate the superiority of our approach, achieving up to a 24.5% performance improvement over strong baselines while simultaneously reducing computational cost by up to 41.6%.

The existence of inner uncertainty and external perturbation usually becomes a hindrance for the effective time-variant control of robot manipulators. Both the robustness and convergence property are regarded as two significant issues to be addressed for preferred solutions to robot manipulators. To handle the time-variant motion control of robot manipulators in the presence of both uncertainty and perturbation, a robust recurrent neural network (RRNN) model with definable convergence time (DCT) property is proposed in this paper. Theoretical analysis based on Lyapunov theory rigorously proves that the proposed RRNN model inherently possesses the global stability, robustness and time efficiency. The solution synthesized via the proposed model with uncertainty and perturbation shows desirable time-variant control performance, i.e., faster convergence and higher accurate. In addition, detailed path-tracking examples, performance comparisons, visual-assisted depth maintenance tracking control demonstrations, and extensive tests by applying both PUMA 560 and INNFOS are presented to validate the effectiveness and superiority of the proposed RRNN model for time-variant control of robot manipulators. Note to Practitioners—This article addresses the issue of uncertainty in robot information, a common occurrence in real-time robot learning and control. This paper presents a precise, efficient, and stable solution that leverages real-time feedback information to resolve real-time control problems for robotic manipulators at the velocity level. Additionally, the paper provides a comprehensive overview of the algorithmic steps and theoretical foundations of the RRNN model to facilitate understanding. To validate the effectiveness and superiority of the proposed approach, the study conducts computer simulations and comparisons using actual parameters and models. Finally, an application to the depth maintenance trecking control of robot mainpulators provides an applicative demo of the porposed neural dynamics for practitioners.

The immediate feedback tracking control system design of heterogeneous robots with uncertainty is considered to be a significant issue in robotic research. Note that when the robot information is uncertain, the scale of computation would become increasingly large and the accuracy of tracking control would become exceptionally low. The realization of the immediate feedback control system of heterogeneous robots with uncertainty remains to be a challenging problem. Many conventional zeroing neural network (CZNN) models have been developed accordingly. However, most of them are supported by the hypothesis that the robot parameters are complete and accurate, and the associated models possess the exponential convergence property. To handle the robot uncertainty as well as to improve the convergence performance, a new zeroing neural network (ZNN) with super-exponential convergence (SEC) rate is put forward in this paper termed SEC-ZNN, to resolve the robust control issue of uncertain heterogeneous robots. The proposed SEC-ZNN takes full advantage of effector real-time information, with robust controlling and super-exponential convergence performance so far as to the robot information is uncertain. Theoretically, the super-exponential convergence properties including lower error bound and faster convergence rate are rigorously proved. Moreover, circular path-tracking example, comparisons and tests via MATLAB, Coppeliasim and experiment via robot INNFOS substantiate the efficaciousness and preponderance of the SEC-ZNN for the immediate feedback control system for heterogeneous robots with uncertainty.Note to Practitioners—This paper is motivated by the problem that most robots which need real-time tracking control in real applications come with uncertainty. It is important to note that traditional robot tracking control algorithms mostly require complete robot information or assume information complete, which does not correspond to the actual situation of robot control. Moreover, for practical applications in robotics, the real-time tracking control problem is very attractive. Therefore, an accurate, efficient and stable solution is of great significance to practitioners in this area. In this paper, the SEC-ZNN algorithm is proposed to solve the problem of real-time control of heterogeneous robots with uncertainty in real applications for practitioners. The proposed methos makes full use of the real-time feedback infromation to solve the real-time tracking control problem of heterogeneous robots with uncertainty at the velocity level. The algorithmic steps and principle explanation of the SEC-ZNN scheme are also presented for better understanding. Simulation studies and comparisons are performed on a Stewart robot to confirm the effectiveness and superiority of the proposed scheme. Furthermore, the simulation experiment in Coppeliasim platform is performed to confirm the possibility of portability of the SEC-ZNN to real robot operations. Finally, applications on a real-world robot INNFOS verify the physical relizability of the proposed SEC-ZNN for the engineering practice via heterogeneous robots.

Several deep learning-based methods have been proposed to extract vulnerable plaques of a single class from intravascular optical coherence tomography (OCT) images. However, further research is limited by the lack of publicly available large-scale intravascular OCT datasets with multi-class vulnerable plaque annotations. Additionally, multi-class vulnerable plaque segmentation is extremely challenging due to the irregular distribution of plaques, their unique geometric shapes, and fuzzy boundaries. Existing methods have not adequately addressed the geometric features and spatial prior information of vulnerable plaques. To address these issues, we collected a dataset containing 70 pullback data and developed a multi-class vulnerable plaque segmentation model, called PolarFormer, that incorporates the prior knowledge of vulnerable plaques in spatial distribution. The key module of our proposed model is Polar Attention, which models the spatial relationship of vulnerable plaques in the radial direction. Extensive experiments conducted on the new dataset demonstrate that our proposed method outperforms other baseline methods. Code and data can be accessed via this link: https://github.com/sunjingyi0415/IVOCT-segementaion.

In recent years, there has been widespread attention drawn to convolutional neural network (CNN) based blind image quality assessment (IQA). A large number of works start by extracting deep features from CNN. Then, those features are processed through spatial average pooling (SAP) and fully connected layers to predict quality. Inspired by full reference IQA and texture features, in this paper, we extend SAP (1stmoment) into spatial moment pooling (SMP) by incorporating higher order moments (such as variance, skewness). Moreover, we provide learning friendly normalization to circumvent numerical issue when computing gradients of higher moments. Experimental results suggest that simply upgrading SAP to SMP significantly enhances CNN-based blind IQA methods and achieves state of the art performance.

Biometric identification has verified its effectiveness in personal identity verification because of the uniqueness and noninvasion. In this research, we tend to apply the detection of biometric information to a remote sensing system for the purpose of security area monitoring. Our system is established by collecting signals from the coming individuals via the remote measurement in the specific condition where both kinds of data are detected to determine the identity. Specifically, the measuring of gait signals and facial images is integrated to provide a way of improving the detection accuracy and the robustness. In addition, the fuzzy association rule (FAR) is employed for data analysis in line with the outcomes of different methods. As such, the signals are integrated and transmitted for further processing and remote identification. Experiments are conducted to demonstrate the capability of the proposed system. With the training data increases, a high detection accuracy of 95.2% is obtained, which makes it a promising basis for the realization of remote identity verification.