Continuous sign language recognition (CSLR) aims to model the temporal evolution of visual gestures to recognize continuous semantic units, which is of great significance for applications in deaf communication assistance and intelligent human–computer interaction. While existing methods emphasize local segment modeling and long-range dependency capture, they often overlook the critical role of global semantic context in overall video comprehension—an oversight that contradicts the inherently context-dependent nature of sign language. Moreover, sign language videos frequently contain a large number of visually similar but semantically meaningless motions. These misleading segments are easily misperceived as valid glosses, thereby degrading recognition accuracy. To address these challenges, we propose GANet (Gloss-Aware Network), a novel CSLR framework with cross-modal input adaptability. Inspired by the hierarchical structure of "book–chapter–content", GANet explicitly models global context to guide local understanding while effectively suppressing irrelevant motion noise. Specifically, we introduce a Global Context Modeling Module to capture semantic patterns across frames and an auxiliary task to enhance the model's ability to learn high-level structural semantics. In addition, we propose a Gloss-Aware Module that leverages global semantics to model the spatiotemporal occurrence of glosses, thereby improving the recognition of meaningful gestures. Extensive experiments on multiple benchmark datasets demonstrate that GANet outperforms existing methods, validating its effectiveness, robustness, and broad adaptability to both RGB (red, green, and blue) and event-based data.

Systems biology has traditionally relied on network abstractions and mechanistic models to study complex biological systems. However, advances in model expressiveness and data availability have not translated into proportional improvements in mechanistic understanding. We suggest that a key limitation may arise from a structural mismatch: prevailing pairwise interaction models fail to capture the inherently higher-order organization of biological systems. Across molecular, cellular, and ecological scales, system behavior is governed by cooperative, conditional, and context-dependent multi-body interactions that cannot be faithfully represented by pairwise projections alone. To address this challenge, we advocate a paradigm shift toward explicit higher-order structural representations combined with data-driven, learnable dynamical models. Within this framework, artificial intelligence enables the inference of governing dynamical rules and the discovery of mechanisms operating on higher-order structures, while large language models can accelerate hypothesis generation and the integration of prior knowledge. Together, these advances point toward a unified, generative approach to systems biology that moves beyond descriptive networks toward an interpretable, mechanism-driven understanding of the processes underlying biological function.

The utilization of liquefied natural gas (LNG) cold energy represents an important approach to improving the efficiency of the LNG value chain. Existing research on LNG cold energy utilization has mainly focused on steady-state simulations and key parameter optimization, while studies on dynamic simulation remain relatively limited. To address this gap, this study develops a dynamic model of a boil-off gas (BOG) re-liquefaction system coupled with an Organic Rankine Cycle (ORC) power generation unit driven by LNG cold energy, with particular emphasis on system dynamic stability. The effects of disturbances in certain parameters, such as temperature and mass flow rate, on system stability and dynamic response are investigated. The results indicate that when the LNG mass flow rate increases by 5%, the BOG re-liquefaction rate rises from 2,006 kg/h to approximately 2,100 kg/h (about 5%), while the ORC power output even increases from 60 kW to around 65 kW (about 8%), demonstrating the importance of sufficient cold energy for the ORC system. In contrast, a ±1 °C variation in BOG temperature has a limited impact on ORC power output (generally less than 2%), but it significantly affects the BOG re-liquefaction rate, which can increase by up to 10%. This study provides valuable insights into the dynamic operational characteristics of the BOG-ORC system and highlights the potential of utilizing LNG cold energy for both BOG re-liquefaction and power generation.

Accurate drug-target interaction (DTI) prediction is essential for drug repositioning and accelerating drug discovery. Deep learning methods have made remarkable progress over traditional biological experiments, yet existing models often fail to capture local node topologies and multi-view semantic dependencies simultaneously. Moreover, most methods rely on basic loss functions that cannot filter out redundant noise, hindering the learning of compact and discriminative node representations. In this work, we propose a DTI prediction framework that integrates a hierarchical gated multi-head attention (HGMA) mechanism with an information bottleneck (IB) strategy. HGMA adopts a two-layer architecture: the first layer performs weighted aggregation over semantic meta-paths, and the second layer fuses attention heads via an adaptive gating mechanism, enhancing drug and target representations. The IB module compresses inputs by removing task-irrelevant redundancy while preserving predictive information, improving discriminability and generalization. Extensive experiments show that our model consistently outperforms state-of-the-art methods in both accuracy and robustness.

Accurate fault diagnosis of water injection pump is essential for ensuring operational safety and efficiency in oil and gas exploitation. However, traditional diagnostic methods often struggle with non-stationary vibration signals and severe category imbalance in complex industrial environments. To address these challenges, this paper proposes a multi-level Inception-long short-term memory (Inception-LSTM) network integrated with wavelet packet decomposition (WPD) and efficient channel attention (ECA), termed the multi-level Inception-LSTM network with WPD and ECA (MILN-WE). The proposed framework first employs WPD to decompose complex vibration signals into fine-grained frequency sub-bands, capturing subtle fault characteristics. Subsequently, a multi-scale Inception module is utilized to extract diverse spatial features, while an LSTM layer captures the long-term temporal dependencies of the signals. The integration of the ECA mechanism further enhances the model’s ability to focus on critical diagnostic information. The effectiveness of MILN-WE is validated using a private oilfield water injection pump dataset and a public rotating machinery dataset. Experimental results demonstrate that the proposed model achieves higher diagnostic accuracy and robustness compared to state-of-the-art methods, particularly under conditions of strong noise interference and data imbalance. Specifically, on the private oilfield water injection pump dataset, the model achieved an accuracy of 99.38%, improving upon traditional convolutional neural network (CNN) and class-balanced-CNN models by 6.05% and 3.24%, respectively. This study provides a high-precision and robust solution for the intelligent predictive maintenance of critical energy equipment, offering significant theoretical and practical value for industrial health monitoring systems.

With the increasing penetration of distributed renewable energy and the widespread integration of power electronics, the primary frequency response in islanded microgrids encounters significant challenges. To address this issue, a voltage-load-sensitivity-based auxiliary control method for Static Synchronous Compensator (STATCOM) is proposed, enabling their use in primary frequency regulation. During low-frequency events, the voltage at the controlled node is rapidly reduced, thereby effectively implementing fast load shedding for sensitive loads through Conservation Voltage Reduction (CVR) without direct active power injection. Furthermore, a closed-loop extension that incorporates an additional voltage outer loop is presented to compensate for the early power deficit. Simulation results demonstrate that, compared to traditional Automatic Voltage Regulation (AVR) and the combination of AVR with a STATCOM, the open-loop scheme rapidly lowers the local voltage before the frequency nadir occurs, which indicates a substantial improvement in frequency metrics via load shedding. Due to its earlier actuation timing, shorter response path, and broader control range, the open-loop feedforward control exhibits advantages over the closed-loop system in this scenario.