This study introduced retrieval steps into automated related work generation, proposing a knowledge graph-enhanced three-stage framework(planning-retrieval-generation)to address topic drift and key reference omission in existing end-to-end approaches. The knowledge graph-augmented planning module captured multi-hop keyword relationships for comprehensive topic modeling. Experimental results demonstrated a fourfold improvement over direct generation methods and 89% over conventional RAG approaches. The overall low literature coverage indicated that planning-enhanced retrieval remains crucial for automated related work generation.

This paper addresses the issues of behavior sequence mixing and temporal changes in dominant preferences in the shared account TV recommendation scenario by proposing a mixture-of-experts shared account recommendation(MoE-SAR)algorithm. This method decomposes behavior sequences using a mixture of experts network and adaptively fuses expert outputs through a dynamic gating mechanism to accurately identify individual user characteristics. Additionally, the MoE-SAR method introduces a contrastive learning strategy to minimize the distance between samples from the same source expert and maximize the mutual information between outputs from different experts, effectively enhancing the discriminability and stability of representations. Furthermore, this approach integrates Transformers for temporal modeling to accurately capture the personalized preferences of individual users within shared accounts. The experimental results indicate that on the E-domain of the HVIDEO dataset, MoE-SAR improves MRR@20 by 24.0% and Recall@20 by 10.8% over the second-best baseline. On the V-domain, MRR@20 improves by 8.0% and Recall@20 improves by 4.7%.

Rapid physical modeling of human soft tissue is crucial for surgical simulation. The integration of deep learning with finite element analysis addresses the inefficiencies of traditional modeling approaches. However, existing research primarily focuses on deformation modeling under external loading, neglecting the importance of other physical fields, such as stress and reaction forces, which play a critical role in guiding surgical training. To address this gap, we propose a novel neural network-based approach for rapid multi-physics modeling of soft tissues. This approach efficiently predicts physical fields by compactly encoding the nonlinear relationships between the loading conditions and mechanical responses of the soft tissue. To resolve the issue of non-convergence during network training caused by the scale differences between physical fields, we introduce the convolutional block attention module(CBAM)to dynamically balance the weights between different physical fields, thus overcoming the dominance of large-scale physical fields in model training. Extensive experiments show that the proposed method outperforms similar approaches in terms of both accuracy and efficiency in predicting multi-physics fields of soft tissue under external loads. Compared to traditional numerical methods, it achieves a several-thousand-fold improvement in efficiency with only about a 5% loss in accuracy, making it a promising tool for computer-aided medical technologies such as surgical simulation.

Federated Learning enables multi-party data value sharing with privacy protection and has been widely applied in many fields such as healthcare and energy. However, the existence of abnormal clients can degrade the models performance and reduce system efficiency. Traditional abnormal clients detection algorithms rely on the assumption that the majority of clients are benign, which makes them ineffective against complex attacks and lacks interpretability. To address these issues, a visual analysis method for abnormal client detection in Federated Learning, named FACDVis, is proposed. The method first identifies suspicious clients and anomalous training rounds through the model performance evolution evaluation framework. Next, through the model behavior pattern analysis framework, it further locates the abnormal clients and their corresponding iterations. Finally, parameter heterogeneity diagnosis framework is employed to deeply analyze the attack methods and construct an interpretable multidimensional evidence chain detection framework. Experiments demonstrated that the proposed method effectively resolves data poisoning, model poisoning, and other attacks even when the number of abnormal clients exceeds 80%, the average recognition accuracy rate reaches 94%.

Color, as an important visual encoding channel, is often used in conjunction with visual marks such as points, lines, and areas to represent abstract data. Color difference has great impact on the discriminability and readability of visualizations. Previous research has shown that color difference perception is influenced by the shape and size of graphical elements, yet the extent of this influence in relation to interfering markings remains under-explored. To address this, this paper explored the perception of color differences among various graphical elements within complex visualization environments containing interference colors. We generated data with different interference colors and visual elements(such as scatterplots, bar charts, and line charts), designed and implemented an experimental system to validate color difference perception in complex visualization contexts, conducted a crowdsourcing experiment on Prolific, and performed a multi-faceted analysis of the experimental results to construct color difference models among different visualization settings. Our experimental results demonstrated that interference colors had a significant impact on users perception of color differences. Moreover, the perception of color differences between different interference colors was closely related to the shape and size of the graphical elements.

To address the performance degradation caused by intensity inhomogeneity in low-contrast image segmentation, this paper proposes a joint model for image segmentation and bias field correction, which achieves simultaneous estimationof the multiplicative bias field, additive bias field, and the true image. On this basis, an alternating minimization method(ADM)is designed to solve the variational functional minimization problem involving multiple unknown functions. Under given conditions, we prove the convergence of the proposed ADM. Experimental results demonstrate that the proposed image segmentation method has significant advantages in handling low-contrast, blurred boundary, and intensity inhomogeneous images.

As an important branch of information security, digital image encryption technology has become a major research hotspot. At present, most image encryption schemes adopt low-dimensional geometric transformations, which suffer from small periodicity and relatively fixed element values, making them difficult to resist exhaustive attacks. In contrast, high-dimensional geometric transformations are not easy to construct. Inspired by the two-dimensional Fibonacci transform and the three-dimensional Fibonacci-like transform, this paper proposes a quantum image scrambling algorithm based on high-dimensional generalized Fibonacci transforms. First, according to the principle of geometric sequences, two special high-dimensional integer matrices with determinant equal to 1 are constructed, and the high-dimensional generalized Fibonacci transform matrix is obtained via matrix operations. Second, due to the large periodicity of high-dimensional geometric transformations, decryption by directly using the periodicity is infeasible; therefore, the inverse transform of the high-dimensional generalized Fibonacci transform is constructed. Finally, based on the novel enhanced quantum representation(NEQR)model, the high-dimensional generalized Fibonacci transform and its inverse are applied to the quantum image encryption and decryption processes, respectively. The proposed algorithm features flexible and diverse transformation formulas and can generate high-dimensional encryption matrices. Taking 8-bit grayscale image encryption as an example, the effectiveness of the algorithm is verified. Simulation results show that the algorithm achieves satisfactory encryption and decryption effects, has a large key space and strong key randomness, and exhibits good resistance against attacks.

Portrait hairstyle removal aims to eliminate existing hairstyles from portrait images and generate high-fidelity bald images. It not only provides users with a flexible tool for virtual hairstyle editing, but also supplies unobstructed facial texture information for 3D face reconstruction, thereby improving the realism and geometric detail of reconstructed face models. However, achieving high-quality hairstyle removal remains challenging due to the complex and highly variable geometry of hairstyles, interference from occlusions such as hats and hair accessories, and the scarcity of paired training data. Existing methods often struggle to balance effective occlusion removal with faithful identity preservation. To address these issues, this paper proposes an identity-preserving portrait hairstyle removal framework for removing hairstyles and hat-related occlusions while generating natural and realistic bald portraits. First, the SegFace semantic segmentation model is employed to extract mask regions corresponding to hair and hats. A bald generator is then trained to focus on content synthesis within the masked regions, so that the generated content remains consistent with the original face and background in terms of skin tone, illumination, and semantic continuity. In addition, an identity loss is introduced to preserve facial identity during hairstyle removal. To further handle hair accessory occlusions with diverse shapes and spatial extents, facial landmarks are combined with Bézier curve fitting to refine the region below the eyebrows, thereby reducing interference with identity-related facial areas. Experimental results demonstrate that the proposed method effectively removes a wide range of hairstyles and hat-related occlusions while maintaining natural visual quality and identity consistency.

Traditional design and development methods for complex components face challenges including CAD-CAE model conversion, low levels of automation and intelligence, and insufficient integration of the design-simulation-optimization workflow. These issues reduce R&D efficiency and drive up costs. Therefore, this paper proposes intelligent parametric-optimization design method for complex components based on isogeometric analysis(IGA). The method achieves parametric modification and reuse of IGA models through constraint morphing, simulation verification of IGA morphed models under multi-physics fields, and optimization of mechanical property predictions. Using three representative models across diverse application scenarios, we validated the topological consistency and suitability for simulation of geometric models after parametric constraint morphing. In a case study on the optimization of a collision-beam IGA model, integrating the proposed methods and solvers within a visualization platform demonstrates the effectiveness of an intelligent optimization approach driven by the fusion of data and models. This method enhances optimization iteration efficiency for complex components and strengthens independent research and development capabilities.

With the rapid transmission of information flow in digital networks, images, as key information carriers, face severe security threats during transmission. This paper introduces a hyperbolic tangent memristor model into a four-dimensional chaotic system to construct a novel five-dimensional memristor chaotic system. Firstly, the dynamical characteristics of the system are systematically analyzed from aspects such as attractor phase diagrams, dissipation, stability of equilibrium points, bifurcation diagrams, and Lyapunov exponents. The research results show that the system exhibits rich hyperchaotic behaviors. Subsequently, the chaotic system is combined with an image encryption algorithm to design a new encryption scheme based on the scrambling-diffusion architecture. Finally, a comprehensive security analysis of the encryption algorithm is conducted. The experimental results demonstrate that the algorithm can effectively resist various attacks and has high security.

In the modeling of B-spline curves that interpolate specific data points, it is often necessary to inversely solve for their control points. Once the control points are determined, a curve shapes that satisfy engineering requirements can be generated. An inverse method for determining the control points of interpolating quadratic B-spline curves is proposed based on the minimization of internal energy. The method selects a control vertex with minimal internal energy and then determines all control vertices step by step using a recursive relationship. After the control points are established, the optimal interpolating curve is determined by minimizing internal energy and achieving optimal approximation, tailored to engineering needs. Finally, the effectiveness of the proposed method is demonstrated through numerical examples.

To address the problem of multiphysics coupling in combustion simulation, a numerical simulation method is proposed based on particles and polygonal surface meshes. The proposed method establishes a multiphysics coupling model for the combustion process by considering physical processes such as fluid flow, heat conduction, chemical reactions, turbulent disturbances, and wind-field effects, and analyzes the basic parameters and dynamic evolution characteristics of combustion. During the coupling modeling process, the complex interactions and coupling effects among flame propagation, heat transfer, airflow disturbance, and combustion reactions are specifically characterized. The proposed method can accurately simulate heat transfer, flame propagation, and their influence on the surrounding environment during combustion, thereby predicting fire-spread paths and potential hazards. Simulation results show that the proposed method can effectively predict the interactions among fluid flow, heat transfer, chemical reactions, and wind-field disturbances, as well as in capturing the dynamic characteristics of combustion.