Industrial Anomaly Detection (IAD) plays a critical role in quality control and manufacturing efficiency across various industries. Recent advancements in Large Vision-Language Models (LVLMs) have introduced new paradigms for addressing the challenges in IAD, overcoming limitations of traditional approaches. This survey provides a comprehensive review of the integration of LVLMs with IAD, analyzing their evolution, methodologies, and applications. We systematically categorize existing approaches into three main frameworks: traditional anomaly detection, zero-shot methods, and LVLM-based solutions. We examine how these models leverage multimodal capabilities to enhance anomaly detection, reasoning, and explanation in industrial settings. Furthermore, we compare the performance of various methods across standard benchmarks, discuss current challenges, and highlight promising future research directions. Our findings indicate that LVLM-based approaches offer significant advantages in terms of flexibility, interpretability, and generalization capabilities, particularly in scenarios with limited anomaly samples and complex industrial environments.

Magnetic Resonance Imaging(MRI) has a wide range of applications in the medical field, such as diagnosis, treatment, pathological research,etc.However,due to hardware limitations, obtaining high-quality MR images is often challeng ing in clinical practice. Therefore, reconstructing high-quality MRimages through partial images has important significance in medical research. Existing super-resolution methods usually use multi-contrast MRI to reconstruct MR images. However, existing methods usually use single-scale MR images for reconstruction and do not combine with the specificity of MR images. To address this issue, we propose a multi-scale feature transfer network(SFSR) based on spatial and frequency domains, which comprises four components, including the shallow feature extrac tor, and the Multi-Scale Frequency Attention Block(MFAB), and the Multi-Scale Spatial Attention Block(MSAB), and the Multi Scale Fusion Block(MSFB).Firstly, we utilize the shallow feature extractor to extract features at three scales from both the target and reference images. These features are then separately fed into the Multi-Scale Frequency Attention Block and the Multi-Scale Spatial Attention Block to align the features. Finally, the Multi Scale Fusion Block are employed to fuse the aligned features across different scales.Extensive experiments on IXI and FastMRI datasets show that SFSR achieves the most competitive results over state-of-the-art approaches.

The Hybrid Flow-Shop Scheduling Problem (HFSP) is a pivotal challenge in the intelligent transformation of manufacturing, where efficient solutions are critical for enhancing production efficiency and resource utilization. However, traditional swarm intelligence algorithms for solving HFSP often exhibit weaknesses in global search capabilities, susceptibility to local optima, and insufficient convergence efficiency to meet practical scheduling demands. Furthermore, complex constraints in real industrial scenarios—such as dynamic variations in manual operation times and split orders—have not been adequately modeled, resulting in a significant gap between theoretical research and engineering applications. To address these limitations, this study conducts research from both theoretical optimization and engineering adaptation perspectives.For the classic HFSP with the objective of minimizing makespan, an Improved Subtraction-Average-Based Optimizer (ISABO) is proposed. This algorithm integrates a simulated annealing perturbation mechanism and a subgroup learning strategy. The simulated annealing perturbation enhances global search capabilities by probabilistically accepting inferior solutions during iterations, thereby avoiding local optima. Concurrently, the subgroup learning strategy guides partial populations toward optimal solutions, improving solution quality. The research follows a progression from "basic problem optimization" to "real-world scenario adaptation." Experiments are conducted using production data from a paint roller manufacturing plant, with three widely-used HFSP algorithms—Genetic Algorithm (GA), Whale Optimization Algorithm (WOA), and Grey Wolf Optimizer (GWO)—selected as benchmarks. For the classic HFSP, ISABO achieves shorter makespan than all baseline algorithms in 100% of test instances and demonstrates the highest stability (measured by standard deviation) in 61.11% of cases .

Searchable encryption (SE) enables secure retrieval over encrypted data, balancing privacy and usability in cloud storage. This paper reviews SE's evolution, contrasting symmetric (SSE) and asymmetric (ASE/PEKS) models, and summarizes breakthroughs in dynamic data support, multi-keyword search, and leakage resilience. Current challenges include efficiency in complex queries, dynamic scenario adaptability, and quantum threats. Future directions focus on lightweight protocols, post-quantum designs (e.g., lattice-based cryptography), and integration with blockchain/homomorphic encryption to advance privacy-preserving frameworks. This work provides critical insights for SE's development in the post-quantum era.

Although the Zebra Optimization Algorithm (ZOA) has the advantages of fewer parameters and easy implementation, its premature convergence and tendency to fall into local optima limit its practical applications. To address these issues, this paper proposes a multi-strategy improved Zebra Optimization Algorithm (IZOA). Firstly, a longitudinal-transverse crossover strategy is integrated to enable multi-dimensional information interaction within the population, enhancing global search capabilities in complex solution spaces. Secondly, a nonlinear dynamic scaling factor is designed to balance exploration and exploitation through an adaptive step-size adjustment mechanism. Additionally, a lens imaging reverse learning strategy is introduced to expand the search scope by generating reverse solutions based on optical imaging principles. Finally, 13 standard benchmark functions are utilized to comprehensively evaluate the improved IZOA, the original ZOA, the Whale Optimization Algorithm (WOA), and the Golden Jackal Optimization (GJO) algorithm. Experimental results demonstrate that IZOA outperforms the other three algorithms in terms of convergence speed and global search capability.

The rise of cloud computing has completely transformed the architecture and operation mode of traditional IT infrastructure, and task scheduling algorithms play a crucial role in this process. An excellent task scheduling algorithm can ensure the efficient and stable operation of cloud computing systems, enhance user experience, and maximize the economic benefits of service providers. This paper addresses the issues of uneven heterogeneous resource load and insufficient resource utilization in cloud task scheduling. By taking task completion time, load balance degree, and resource waste as optimization objectives, this paper improves the recently proposed Nutcracker Optimization Algorithm (NOA) and creatively applies it to the cloud task scheduling scenario to improve the scheduling efficiency of cloud computing systems.

This paper proposes an Improved Nutcracker Optimization Algorithm with Adaptive Chaotic Strategy (INOACS). By introducing random reverse learning, elite pool selection, and non-exclusive local search strategies, the quality of solutions is improved, and the local exploitation and global search capabilities are enhanced, avoiding the algorithm falling into local optima. To address the problems of uneven initial population distribution and the difficulty in balancing global search and local exploitation, chaotic mapping initialization and an adaptive α parameter based on population diversity are introduced, further improving the convergence speed and optimization ability of the algorithm.

A scheduling model with task completion time, load balance degree, and resource waste as optimization objectives is constructed based on INOACS. Simulation experiments verify the good performance of the INOACS algorithm under different task scalesand numbers of virtual machines.