The dark web’s anonymity infrastructure facilitates serious crimes, including drug trafficking, arms dealing, and the distribution of child sexual abuse material. These illicit ecosystems are organizationally structured; the same actors disperse activities across multiple domains to evade investigation. Consequently, approaches that analyze individual dark web sites independently have limited ability to identify criminal organizations, derive investigative leads, or reveal cross-site relationships. To address this limitation, we propose a dark web external identifier clustering methodology that detects common organizations by leveraging shared external identifiers (email, Telegram, and Twitter) observed across distributed sites. From approximately 1.49 million crawled dark web URLs, our recursive crawler observed approximately 100,000 subdomains returning HTTP 200. Among these, 24,403 subdomains contained at least one external identifier, from which we selected 10,274 criminal subdomains for analysis. We organized the subdomains under seven scenarios spanning three strategies: Single Identifier Clustering, which groups domains that share a single chosen identifier; Sequential Clustering, which traces inter-organizational ties by varying the combinations and initial order of the three identifiers; and Global Clustering, which organizes the full connectivity induced by all collected identifiers. For each discovered organization, we verified organizational sameness through HTML similarity, domain prefix matching, and manual verification via direct access. Our results show that Global Clustering identifies the largest number of organizations, whereas email anchored Sequential Clustering yields the most effective clues for understanding and tracking organizational operating structures.

Vibrations can reduce the controllability of vehicles and can be uncomfortable for passengers. In particular, nonlinear vibrations can lead to larger amplitudes and complex behaviors. Among these vibrations, chaotic vibrations are inherently unpredictable and may lead to instability, which is highly undesirable in practical applications. Although several control methods have been developed to eliminate chaotic vibrations in active suspension systems, they require a priori knowledge about vehicle systems. Thus, designing chaos controllers for practical scenarios can be very challenging. In this study, we propose a method based on machine learning to overcome this limitation. The deep deterministic policy gradient algorithm is used to learn optimal control input to suppress chaotic vibrations. As a typical model of nonlinear motion, we considered a quarter-car model with jumping nonlinearity and trained a reinforcement learning model to suppress chaotic vibrations. The numerical results show that learned policies were able to suppress chaotic vibration and reduce the overall level of vibration. The obtained policies were evaluated across a wide range of forcing frequencies and simulations with stochastic excitation noises. The results demonstrate that chaos control based on deep reinforcement learning was able to reduce vibrations and stabilize chaotic behavior effectively in this application.

The development of agentic software engineering (ASE) is shifting the software development process, specifically by adopting autonomous agents into the fundamental software development process, where autonomous agents using large language models (LLMs) and artificial intelligence (AI) make decisions, write code, and collaborate across the entire software development lifecycle. The systematic review is a summary of literature on the agentic software engineering methodologies, frameworks, and techniques published in 2022-2025. We identify and compare methods based on autonomous coding, multi-agent systems, iterative refinement, and human-agent collaboration, drawing on 61 high-quality studies. We find that LLM-based agents show substantial potential in automated software engineering and that recent work is exploring new trends in agent cooperation, reasoning systems, and tool integration. We recognize such key challenges as the coordination procedures, reliability, explainability, and the ethical issues. This review characterizes the emerging role of agentic software engineering in improving the productivity of developers, the quality of their code, and the automation of their processes, and it offers a structured roadmap for future research. With the rapid evolution of the field, agentic approaches will tend to transform the conception, creation, and maintenance of software and provide more autonomous, adaptive, and intelligent systems.

This work aims to evaluate the crucial need for reliable, long-term electrical-load forecasting in hospitals to prevent power outages, which can cause the patient-fatalities, with reference to real-life case studies in India and other locations. Conventional approaches, such as ARIMA, are designed for short-term forecasting and do not deal with the nonlinear and continuous-dynamics of a hospital caused by the HVAC system, medical equipment, and emergencies. Experimental load data from one-year of hospital was used to evaluate seven ML/DL models: MLP, XGBoost, RF, SVR, LSTM, CNN-LSTM, and Transformer. The evaluation-criteria were RMSE, MAE, MAPE, and R². RF, XG Boost, CNN-LSTM, and SVR beat Transformer with RMSE=0.0434 ± 0.0003, MAE=0.0349 ± 0.0002, MAPE=0.0983 ± 0.0009, R²=0.8799 ± 0.0004. Both LSTM and MLP models showed similar results, with an RMSE of 0.0436 and an R² value of 0.8792. The novelty and superiority of this proposed work lie in the Transformer-based energy forecasting model, which uses self-attention to capture long-range dependencies, making it pioneering. Since convergence analysis shows steady loss curves, this variation is supported. The authentication loss plateaus between 25 and 35 epochs, indicating no overfitting. Over 20 separate trials, convergence stability is consistent with low variance. A Friedman test (χ² > 45, p < 0.0001) and Wilcoxon signed-rank tests with Holm correction (adjusted p-values < 0.0005) displayed that the Transformer outperformed other models. In contrast, LSTM/MLP models showed no substantial difference (p > 0.05). It reduces energy consumption by 10–15%, improves energy management in healthcare, and reduces patient mortality worldwide.

The maritime industry’s accelerating digital transformation, driven by the widespread adoption of Internet of Things (IoT) devices, integrated vessel systems, and automated port operations, has significantly expanded its cyber-attack surface, rendering it increasingly vulnerable to sophisticated threats. This paper investigates the application of Open-Source Intelligence (OSINT) as a proactive, scalable, and cost-effective methodology to enhance maritime cybersecurity. We present a specialized maritime OSINT framework, comprising a systematic, five-phase workflow (Identification, Collection, Processing, Analysis, Dissemination) tailored to the sector’s unique operational and threat landscape. The framework is supported by a comprehensive taxonomy of maritime cyber threats and a curated toolkit of over 100 OSINT resources spanning vessel tracking, corporate intelligence, digital reconnaissance, and analytical platforms. Through applied case studies, we demonstrate how this approach can identify critical vulnerabilities, such as exposed operational technology, weak software supply chain links, and human-factor risks derived from publicly accessible data. Our findings confirm that maritime OSINT provides an indispensable intelligence layer, enabling stakeholders to transition from reactive security postures to proactive risk management through early threat detection, informed vulnerability assessment, and enhanced situational awareness while adhering to proposed ethical and legal guidelines for responsible intelligence collection.

As floating offshore wind farms move further offshore into more challenging environments, robust and scalable monitoring strategies become critical for maintaining operational safety and performance. This study investigates the use of Sentinel-2 satellite imagery to remotely track the positional offset of semi-submersible floating wind platforms and evaluate their dynamic response to varying metocean conditions. A convolutional neural network (U-Net) was developed to segment turbine platforms in optical imagery, enabling automated extraction of positional data. These positions were then correlated with wind and wave parameters using both linear and multivariate regression models. The results indicate a statistically significant but moderate global correlation between wind speed and platform horizontal displacement, which improves substantially when accounting for wind directionality, reaching <inline-formula> <tex-math notation="LaTeX">$R^{2}$ </tex-math></inline-formula> values of 0.73 for westerly winds. Horizontal displacement was found to be anisotropic and strongly dependent on the direction of environmental forcing, with both wind and wave contributions playing key roles. Clustering analysis further revealed four distinct offset regimes, each associated with characteristic wind patterns, confirming the potential for satellite-driven classification of operational states. Additionally, the method demonstrated value for anomaly detection, with deviations in turbine position highlighting possible mooring line failures. Independent in-situ position measurements (e.g., GNSS/INS, SCADA position channels, or mooring telemetry) were not available for the study period; therefore, we report uncertainty bounds and surrogate consistency evidence rather than a direct validation of absolute offsets. We demonstrate an end-to-end workflow for 1) extracting apparent platform offsets from Sentinel-2 imagery and 2) reporting descriptive associations between these offsets and farm-scale metocean forcing; any asset-health or degradation inference is presented only as an exploratory condition-indicator example and is not claimed as a validated diagnostic. These findings suggest that, despite limitations in spatial resolution, Sentinel-2 imagery can complement traditional in-situ monitoring systems by offering a cost-effective, wide-area surveillance layer. The approach supports not only real-time tracking but also the development of predictive maintenance tools, contributing to safer and more resilient offshore wind operations.

Underwater visual perception plays a crucial role in marine resource exploration, engineering inspection, and ecological monitoring. However, detecting small underwater objects remains challenging due to degraded image quality, complex backgrounds, and computational constraints in embedded systems. This paper presents a lightweight and efficient underwater small object detection framework that achieves a balance between accuracy, inference speed, and deployability. The proposed network employs a partial convolution-based lightweight backbone to reduce redundant computation, an enhanced attention mechanism integrating statistical pooling and multi-scale convolution for refined texture perception, and a multi-branch auxiliary fusion network to preserve spatial and semantic information across scales. Evaluations on the URPC2021 dataset show that the framework attains 83.3% mAP@0.5, 86.4% recall, and 103 FPS with only 3.1M parameters and 7.5 GFLOPs, outperforming existing state-of-the-art lightweight detectors. The results confirm its strong potential for real-time deployment in underwater robotic and embedded applications.

To advance the efficiency and dynamic responsiveness of electric vehicles (EVs), this study introduces a comprehensive intelligent control and energy-management framework that couples a superconducting magnetic energy storage (SMES) assisted hybrid energy storage system (HESS) with a synchronous reluctance motor (SynRM) drive. The proposed configuration exploits the high power density and instantaneous response of SMES to enhance performance during transient load changes. To mitigate nonlinear behaviors, parameter deviations, and external disturbances within both HESS and SynRM subsystems, two resilient nonlinear controllers are designed: an Adaptive Barrier Function based Super-Twisting Sliding Mode Controller (ABC-STSMC) and a Synergetic Controller. The adaptive gains of the ABC-STSMC are tuned automatically through the Whale Optimization Algorithm (WOA) to ensure optimal robustness. Moreover, a deep reinforcement learning–driven energy management system (RL-EMS) governs real-time power allocation among the SMES, supercapacitor, and fuel cell, thereby maximizing overall efficiency and component longevity across variable driving conditions. The framework is validated through MATLAB/Simulink simulations and Hardware-in-the-Loop (HIL) experimentation. Quantitative comparisons with conventional Sliding Mode Control (SMC) and Integral Sliding Mode Control (ISMC) demonstrate that theWOA-optimized ABC-STSMC improves transient response by 28%, enhances tracking accuracy by 35%, and strengthens disturbance rejection. Simultaneously, the RL-EMS lowers switching losses and ensures balanced energy utilization. Collectively, the integrated control and management scheme offers higher reliability, robustness, and adaptability, positioning it as a promising solution for future high-performance electric mobility systems.