Honeypots are widely used as cyber-deception tools to study adversarial behaviour, yet their effectiveness is limited by a trade-off between realism and security risk. Low-interaction honeypots are easily detected, while high-interaction honeypots provide realistic data at the cost of network exposure. Recent advances suggest that Large Language Models (LLMs) can mitigate this trade-off by dynamically generating convincing outputs without requiring a vulnerable backend. In this paper, we present GenPot, a fine-tuned LLM-powered honeypot that integrates command-line, API, and dynamic web interaction to deliver realistic yet non-compromisable environments. Our approach combines supervised fine-tuning with prompt engineering, cybersecurity safeguards, and state management to simulate consistent and realistic system responses. As a proof-of-concept use case, we implement the system on top of an OpenCanary baseline simulating a Synology NAS device. To rigorously assess the framework, we conducted an exhaustive multi-dimensional evaluation encompassing deep semantic fidelity, inference latency, high-concurrency scalability, and operational energy efficiency (RPS/W). Technical results demonstrate that the optimized architecture sustains over 1,000 requests per second with near-perfect structural validity, while token-level guardrails ensure high resilience against prompt injection attacks. Crucially, we validate the system’s practical deception efficacy through a 14-day in-the-wild deployment and a human-in-the-loop credibility assessment, where experts were unable to distinguish the honeypot from a physical device (45
Anomaly detection plays a critical role in ensuring the security of industrial control systems. As a classical anomaly detection method, One-Class Support Vector Machine (OCSVM) has been widely applied across various domains, but its performance heavily depends on the selection of kernel function parameters. Moreover, as the dimensionality of industrial data increases, the data distribution becomes sparser, making it difficult for OCSVM to capture the distribution characteristics of normal samples, thereby compromising detection performance. To address these two challenges, this paper proposes a novel industrial anomaly detection method named Information Clustering One-Class Support Vector Machine (IC-OCSVM). First, each feature dimension is treated as a separate system, and its information entropy is calculated to eliminate redundant dimensions. Subsequently, information content is introduced to characterize the differences between samples, and an explicit mapping function based on information content is designed to replace the conventional Gaussian kernel mapping, enabling OCSVM-based anomaly detection. By constructing this mapping function, the proposed method effectively avoids the dependence of traditional OCSVM on kernel parameter selection. In addition, by using information content to measure the distance between samples, the constructed OCSVM effectively alleviates the problem of data sparsity in high-dimensional data. Experiments were conducted on two industrial control system security datasets (SWaT and WADI) and compared with several state-of-the-art anomaly detection algorithms. Experimental results demonstrate that the proposed method achieves superior accuracy, attaining F1-scores of 85.91
RGB-T object tracking aims to achieve robust, all-weather tracking in complex environments by fusing complementary information from visible (RGB) and thermal infrared (T) images. However, existing methods often suffer from insufficient feature representation and redundant cross-modal invalid information due to the lack of targeted fusion strategies for different modal features. This leads to tracking drift and reduced accuracy in complex scenarios. To address these issues, we propose the BFA-HARF Tracker. Our framework is built upon two synergistic core modules: a Bidirectional Feature Adapter (BFA) and a Hybrid Attention with Receptive Fields (HARF) module. First, the BFA module facilitates a continuous bidirectional information flow between RGB and thermal modalities throughout the backbone network, ensuring features from both sources are progressively aligned and mutually enhanced. Subsequently, the HARF module is strategically inserted to process these aligned features, where it collaboratively captures global dependencies via self-attention and fine-grained local patterns via convolution. This two-stage process of “align-then-fuse” ensures a comprehensive feature representation for the final tracking task. Extensive experiments on four public RGB-T benchmark datasets demonstrate that the proposed algorithm effectively alleviates bounding box ambiguity and tracking drift, achieving performance comparable to state-of-the-art methods.
High-dimensional features are widely used in fields such as gene expression analysis and finance due to their ability to capture complex system information. However, high-dimensional data often exhibit redundancy, sparsity, and nonlinear relationships, which hinder effective feature construction and pose challenges for machine learning models. Dimensionality reduction is therefore essential to transform high-dimensional data into compact, discriminative representations. Although Genetic Programming (GP) has shown potential in feature subset construction, it still faces challenges in handling high-dimensional data: (1) random terminal node selection produces low-quality individuals; (2) fixed evolutionary strategies limit the discovery of informative features; and (3) the vast search space necessitates effective dimensionality reduction. To address these issues, a multi-tree and multi-strategy iterative genetic programming (MMSGP) algorithm is proposed for feature subset construction and classification. (1) A feature-importance-based terminal node selection improves individual quality. (2) A triple iterative strategy effectively guides evolution. (3) A dynamic dual-subset feature selection reduces the search space by adaptively focusing on valuable features. Experimental results across twelve high-dimensional datasets demonstrate that MMSGP achieves substantial performance improvements compared with five baseline methods. Specifically, MMSGP attains higher average balanced classification accuracy than all baselines on nine datasets and surpasses them in macro F1-score on ten datasets. Ablation studies further confirm the effectiveness of the proposed components.
As critical components of port logistics systems, container terminals rely heavily on yard cranes (YCs) to handle large volumes of containers, where operational efficiency and energy consumption are tightly coupled and difficult to balance. While existing studies primarily emphasize efficiency-oriented YC scheduling, the substantial energy consumption induced by gantry startups and crane movements has received relatively limited attention. Moreover, YC scheduling and yard vehicle positioning constitute a highly interdependent and combinatorial decision problem, which is challenging to solve using conventional optimization approaches. To address this complex real-world scheduling problem, this study proposes an intelligent bi-objective optimization framework that simultaneously determines YC operation sequences and vehicle parking positions, with the objectives of minimizing makespan and gantry energy consumption. A mixed-integer linear programming model is formulated to explicitly characterize energy consumption arising from gantry startups, empty travel, and loaded travel. To efficiently solve the resulting large-scale and conflicting optimization problem, an improved non-dominated sorting genetic algorithm II (INSGA-II) is developed. The proposed algorithm integrates multiple intelligent strategies, including a parking-slot allocation heuristic for high-quality population initialization, problem-specific neighborhood search operators to reduce unnecessary crane movements and startup events, and a destruction-reconstruction local search mechanism to enhance solution intensification and avoid local optima. Numerical experiments based on realistic terminal operation scenarios demonstrate that the proposed intelligent optimization approach achieves high-quality trade-off solutions, significantly reducing energy consumption while maintaining operational efficiency. The results confirm the effectiveness of integrating evolutionary computing and heuristic search for intelligent decision support in complex industrial scheduling systems.
