Search Results (807)

The Urban Heat Island (UHI) phenomenon constitutes one of the most significant climate-related challenges for contemporary cities, intensifying thermal stress, energy demand, and social vulnerability. This study proposes a methodological framework that integrates multi-source data with explainable machine learning techniques in order to both analyse and support the refinement of climate adaptation policies. The approach combines satellite-derived land surface temperature from Sentinel-3, meteorological and air quality indicators, and biophysical and anthropogenic variables. After a preprocessing stage, clustering and classification models (Logistic Regression, Support Vector Classifier) were trained for the city of Madrid, with inference applied to Athens as a reference case. The evaluation of model performance was complemented by explainability techniques (Feature Importance and SHAP), which highlighted temporality, soil moisture, and urban morphology as the most decisive factors for UHI intensity, while atmospheric pollutants were found to play a secondary role. These insights were systematically compared with existing international, European, and national policy frameworks, including the Sustainable Development Goals, the European Green Deal, and Spain’s National Energy and Climate Plan. The findings demonstrate how interpretable, data-driven analysis can bridge the gap between predictive modelling and governance, providing a transparent basis for targeted and evidence-based urban climate adaptation strategies. Full article

As conventional computing architectures face fundamental physical limitations and the von Neumann bottleneck constrains computational efficiency, neuromorphic systems have emerged as a promising paradigm for next-generation information processing. Memristive neurons, particularly third-order circuits operating near the edge of chaos, exhibit rich neuromorphic dynamics that closely mimic biological neural activities but present significant prediction challenges due to their complex nonlinear behavior. Current approaches typically require complete system state measurements, which is often impractical in real-world neuromorphic hardware implementations where only partial state information is accessible. This paper addresses this critical limitation by proposing an innovative hybrid machine learning framework that integrates a Modified Next-Generation Reservoir Computing (MNGRC) with XGBoost regression. The core novelty lies in its dual-path prediction architecture designed specifically for partial state observability scenarios. The primary path employs NGRC to capture and forecast the system’s temporal dynamics using available state variables and input stimuli, while the secondary path leverages XGBoost as an efficient state estimator to infer unobserved state variables from minimal measurements. This strategic combination enables accurate prediction of diverse neuromorphic patterns with significantly reduced sensor requirements. Experimentally, the framework demonstrates its capability to identify and predict the complex spectrum of neuromorphic behaviors exhibited by the third-order memristive neuron. This includes accurately capturing all 18 distinct neuronal patterns, which are theoretically grounded in Hopf bifurcation analysis near the edge of chaos. Additionally, the framework successfully addresses the inverse problem of input stimulus reconstruction. By achieving accurate prediction of complex dynamics from limited states, our approach represents a key breakthrough, where full state access is often impossible, thereby addressing a critical challenge in edge AI and brain-inspired computing. Full article

This work addresses the issue of estimating air pollution maps for urban areas. Spatially dense maps of air pollution can be calculated using physical models, such as ADMS-Urban; however, due to the high computational cost of such models, maps are verified with low temporal resolution (such as monthly or yearly averages). We investigate the feasibility of using machine learning models to predict air pollution maps based on historical data and current measurements from a limited number of monitoring stations. The models are trained on spatially dense pollution maps generated by physical models, along with corresponding measurements from monitoring stations and selected meteorological data. We evaluate the performance of the models using real-world data from a central district in Wrocław, Poland, considering various pollutants such as $P{M}_{2.5}$, $P{M}_{10}$, CO, VOC, and NOx, presented on spatially dense pollution maps with ca. $2\times {10}^5$ points with a 10 × 10 m grid. The results demonstrate that the proposed method can effectively predict air pollution maps with high spatial resolution and a fast inference time, making it suitable for generating pollution maps with significantly higher temporal resolution (e.g., hourly) compared to physical models. We also experimentally demonstrated that PM10, CO, and VOC pollution models can be built based on measurements from $P{M}_{2.5}$ monitoring stations only with similar, and in the case of CO, higher, accuracy than using measurements from $P{M}_{10}$, CO, and VOC monitoring stations, respectively. Full article

Graph data mining has emerged as a prominent area of research in both academic and industrial domains. Dynamic link prediction, a critical subfield within graph data mining, offers a more realistic representation of real-world networks compared to static link prediction, making dynamic link prediction attacks particularly threatening to privacy. While privacy protection in dynamic networks can be achieved by removing certain sensitive links, attackers can still infer hidden sensitive connections from observable network data. Moreover, existing studies seldom address target-level defense against dynamic link prediction attacks. To address these challenges, this paper proposes a Target-Level Privacy protection method against Dynamic Link Prediction attacks (TP-DLP). The method leverages temporal information in dynamic networks to implement targeted protection based on link gradient information, operating within a perturbation range that preserves the inherent characteristics of dynamic networks. Using dynamic social networks as a case study, the approach distinguishes the privacy levels of dynamic links to achieve target-level privacy protection. Extensive experimental results demonstrate that TP-DLP significantly enhances privacy protection while preserving network utility, making it well-suited for targeted defense against dynamic network link prediction. It can be concluded that our method achieves a balanced trade-off between privacy protection effectiveness and network structural fidelity. Full article

This study examines the determinants of environmental, social, and governance (ESG) performance among Chinese financial institutions, with particular emphasis on community engagement, firm size, and ownership structure as drivers of ESG performance and their contribution to the Sustainable Development Goals (SDGs). Utilizing ESG ratings from CSRHub and annual reports from 107 financial companies spanning 2022–2024, hierarchical regression analyses demonstrate that community engagement significantly predicts ESG performance (β = 0.816, p < 0.001), explaining 67.7% of the variance in ESG ratings. Conversely, the firm (β = 5.687 × 10⁻⁶, p > 0.05) and the ownership structure (β = 1.35, p > 0.05) exhibit no statistically significant effect. Robustness evaluations, concerning bootstrapping methodologies and calculations of heteroscedasticity-consistent standard errors, check these findings. The cross-sectional design limits causal inference. Longitudinal studies would allow deeper exploration of temporal dynamics. The results specify that community engagement acts as the primary factor affecting ESG performance within Chinese financial institutions, whereas firm size and ownership structure exercise insignificant influence. Financial institutions should prioritize substantive, sustained community initiatives rather than relying on organizational scale or state affiliation. For policymakers, the findings suggest that incentive mechanisms (e.g., tax credits or green-finance subsidies) should reward verifiable community-impact outcomes rather than firm size or state ownership, which do not reliably predict superior ESG performance. Full article

We propose a Hybrid Parallel Temporal–Spatial CNN-LSTM (HPTS-CL) architecture for optimized indoor environment modeling in sports halls, addressing the computational and scalability challenges of high-resolution spatiotemporal data processing. The sports hall is partitioned into distinct zones, each processed by dedicated CNN branches to extract localized spatial features, while hierarchical LSTMs capture both short-term zone-specific dynamics and long-term inter-zone dependencies. The system integrates model and data parallelism to distribute workloads across specialized hardware, dynamically balanced to minimize computational bottlenecks. A gated fusion mechanism combines spatial and temporal features adaptively, enabling robust predictions of environmental parameters such as temperature and humidity. The proposed method replaces monolithic CNN-LSTM pipelines with a distributed framework, significantly improving efficiency without sacrificing accuracy. Furthermore, the architecture interfaces seamlessly with existing sensor networks and control systems, prioritizing critical zones through a latency-aware scheduler. Implemented on NVIDIA Jetson AGX Orin edge devices and Google Cloud TPU v4 pods, HPTS-CL demonstrates superior performance in real-time scenarios, leveraging lightweight EfficientNetV2-S for CNNs and IndRNN cells for LSTMs to mitigate gradient vanishing. Experimental results validate the system’s ability to handle large-scale, high-frequency sensor data while maintaining low inference latency, making it a practical solution for intelligent indoor environment optimization. The novelty lies in the hybrid parallelism strategy and hierarchical temporal modeling, which collectively advance the state of the art in distributed spatiotemporal deep learning. Full article

In modern high-demand applications, such as real-time video communication, cloud gaming, and high-definition live streaming, achieving both superior transmission speed and high visual fidelity is paramount. However, unstable networks and packet loss remain major bottlenecks, making accurate and low-latency video error concealment a critical challenge. Traditional error control strategies, such as Forward Error Correction (FEC) and Automatic Repeat Request (ARQ), often introduce excessive latency or bandwidth overhead. Meanwhile, receiver-side concealment methods struggle under high motion or significant packet loss, motivating the exploration of predictive models. SimVP-v2, with its efficient convolutional architecture and Gated Spatiotemporal Attention (GSTA) mechanism, provides a strong baseline by reducing complexity and achieving competitive prediction performance. Despite its merits, SimVP-v2’s reliance on 2D convolutions for implicit temporal aggregation limits its capacity to capture complex motion trajectories and long-term dependencies. This often results in artifacts such as motion blur, detail loss, and accumulated errors. Furthermore, its single-modality design ignores the complementary contextual cues embedded in the audio stream. To overcome these issues, we propose A3DSimVP (Audio- and 3D-Enhanced SimVP-v2), which integrates explicit spatio-temporal modeling with multimodal feature fusion. Architecturally, we replace the 2D depthwise separable convolutions within the GSTA module with their 3D counterparts, introducing a redesigned GSTA-3D module that significantly improves motion coherence across frames. Additionally, an efficient audio–visual fusion strategy supplements visual features with contextual audio guidance, thereby enhancing the model’s robustness and perceptual realism. We validate the effectiveness of A3DSimVP’s improvements through extensive experiments on the KTH dataset. Our model achieves a PSNR of 27.35 dB, surpassing the 27.04 of the SimVP-v2 baseline. Concurrently, our improved A3DSimVP model reduces the loss metrics on the KTH dataset, achieving an MSE of 43.82 and an MAE of 385.73, both lower than the baseline. Crucially, our LPIPS metric is substantially lowered to 0.22. These data tangibly confirm that A3DSimVP significantly enhances both structural fidelity and perceptual quality while maintaining high predictive accuracy. Notably, A3DSimVP attains faster inference speeds than the baseline with only a marginal increase in computational overhead. These results establish A3DSimVP as an efficient and robust solution for latency-critical video applications. Full article

This paper introduces AIDE, an active inference-driven evaluation framework designed to provide a unified and theoretically grounded approach for analyzing sequential textual data. AIDE formulates the evaluation problem as variational inference in a latent dynamical system, enabling joint treatment of representation, temporal structure, and predictive reasoning. The framework integrates (i) a representation and augmentation module based on variational learning and contrastive semantic encoding, (ii) a parametric state–space model that captures the evolution of latent states and supports probabilistic forecasting, and (iii) a policy-selection mechanism that minimizes the expected free energy, guiding a latent diffusion generator to produce coherent and interpretable evaluation outputs. This formulation yields a principled pipeline linking evidence accumulation, latent-state inference, and policy-driven generative reporting. Experimental studies demonstrate that AIDE provides stable inference, coherent predictions, and consistent evaluation behavior across heterogeneous textual sequences. The proposed framework offers a general probabilistic foundation for dynamic evaluation tasks and contributes a structured methodology for integrating representation learning, dynamical modeling, and generative mechanisms within a single variational paradigm. Full article

Despite advancements in deep learning-based Monocular Depth Estimation (MDE), applying these models to video sequences remains challenging due to geometric ambiguities in texture-less regions and temporal instability caused by independent per-frame inference. To address these limitations, we propose STF-Depth, a novel post-processing framework that enhances depth quality by logically fusing heterogeneous information—geometric, semantic, and panoptic—without requiring additional retraining. Our approach introduces a robust RANSAC-based Vanishing Point Estimation to guide Dynamic Depth Gradient Correction for background separation, alongside Adaptive Instance Re-ordering to clarify occlusion relationships. Experimental results on the KITTI, NYU Depth V2, and TartanAir datasets demonstrate that STF-Depth functions as a universal plug-and-play module. Notably, it achieved a 25.7% reduction in Absolute Relative error (AbsRel) and significantly enhanced temporal consistency compared to state-of-the-art backbone models. These findings confirm the framework’s practicality for real-world applications requiring geometric precision and video stability, such as autonomous driving, robotics, and augmented reality (AR). Full article

This paper introduces an advanced framework for modeling and scheduling construction processes using causal inference techniques, with particular emphasis on capturing complex technological and organizational interdependencies. By integrating causal calculus and counterfactual reasoning, the study demonstrates how construction schedules can be analyzed and optimized not only through temporal relationships but also through explicit cause–effect structures. A matrix-based scheduling methodology is presented, incorporating diagonal and reverse-diagonal time couplings consistent with the Time Coupling Method (TCM). The computational procedure is detailed, including the determination of earliest and latest event times, identification of the critical path, and computation of activity floats. Based on an in-depth examination of technological and organizational constraints, eight theorems are formulated and proven, establishing the fundamental properties of a scheduling approach that embeds causal mechanisms. The findings indicate that the integration of causal inference into construction planning enables more accurate identification of factors influencing project duration, enhances synchronization of dependent activities, and minimizes conflicts and idle times. This causally informed framework strengthens decision-making by allowing practitioners to predict the consequences of modifications in project execution strategies. The developed models constitute a robust foundation for future research on leveraging causal inference algorithms and artificial intelligence to advance construction process management. Full article

Antarctic ice shelves regulate ice-sheet discharge and global sea-level rise, yet their rapid retreat underscores the need for new, low-cost monitoring tools. We analyze ambient seismic noise recorded by seismometers on the Pine Island Glacier ice shelf to characterize wind-induced signals and detect persistent structural resonances. Power spectral analysis shows that wind sensitivity is strongly damped compared with bedrock sites: noise increases only 5–7 dB from 0 to 25 m s⁻¹ winds, versus a 42 dB increase at an inland bedrock station, reflecting the contrasted coupling environments of floating and grounded substrates. The horizontal-to-vertical spectral ratio (HVSR) spectrograms reveal two temporally stable peaks at ~2.2 Hz and ~4.3 Hz that persist across stations and remain independent of environmental forcing. Forward modeling indicates that these peaks correspond to S-wave resonances within the ice shelf. The inferred ice-water interface depth (~440 m) agrees with the Bedmap2 thickness estimate (466 m). This work demonstrates that HVSR provides an effective passive, single-station method for measuring ice shelf thickness. Full article

This study proposes a collision-risk assessment framework that combines an interval type-2 fuzzy inference system with Dempster–Shafer evidence theory to more reliably evaluate vessel collision risk under the uncertainty inherent in AIS-based marine navigation data. The fuzzy system models membership-function uncertainty through a footprint of uncertainty and produces time-indexed basic probability assignments that are subsequently combined through a Dempster–Shafer–based temporal integration process. Robust combination rules are incorporated to mitigate the counterintuitive results often produced by classical evidence combination. Furthermore, Lenart’s time-based criterion and Fujii’s spatial safety domain are unified to construct a three-level risk labeling scheme, overcoming the limitations of conventional binary risk classification. Case studies using real AIS data demonstrate improved predictive accuracy and significantly reduced uncertainty, particularly when using the robust symmetric combination rule. Overall, the proposed framework provides a systematic approach for handling structural uncertainty in maritime environments and supports more reliable collision-risk prediction and safer navigational decision-making. Full article

Surface crack detection and temporal evolution analysis are fundamental tasks in remote sensing and photogrammetry, providing critical information for slope stability assessment, infrastructure safety inspection, and long-term geohazard monitoring. However, current unmanned aerial vehicle (UAV)-based crack detection pipelines typically treat spatial detection and temporal change analysis as separate processes, leading to weak geometric consistency across time and limiting the interpretability of crack evolution patterns. To overcome these limitations, we propose the Longitudinal Crack Fitting Network (LCFNet), a unified and physically interpretable framework that achieves, for the first time, integrated time-series crack detection and evolution analysis from UAV remote sensing imagery. At its core, the Longitudinal Crack Fitting Convolution (LCFConv) integrates Fourier-series decomposition with affine Lie group convolution, enabling anisotropic feature representation that preserves equivariance to translation, rotation, and scale. This design effectively captures the elongated and oscillatory morphology of surface cracks while suppressing background interference under complex aerial viewpoints. Beyond detection, a Lie-group-based Temporal Crack Change Detection (LTCCD) module is introduced to perform geometrically consistent matching between bi-temporal UAV images, guided by a partial differential equation (PDE) formulation that models the continuous propagation of surface fractures, providing a bridge between discrete perception and physical dynamics. Extensive experiments on the constructed UAV-Filiform Crack Dataset (10,588 remote sensing images) demonstrate that LCFNet surpasses advanced detection frameworks such as You only look once v12 (YOLOv12), RT-DETR, and RS-Mamba, achieving superior performance (mAP_50:95 = 75.3%, F1 = 85.5%, and CDR = 85.6%) while maintaining real-time inference speed (88.9 FPS). Field deployment on a UAV–IoT monitoring platform further confirms the robustness of LCFNet in multi-temporal remote sensing applications, accurately identifying newly formed and extended cracks under varying illumination and terrain conditions. This work establishes the first end-to-end paradigm that unifies spatial crack detection and temporal evolution modeling in UAV remote sensing, bridging discrete deep learning inference with continuous physical dynamics. The proposed LCFNet provides both algorithmic robustness and physical interpretability, offering a new foundation for intelligent remote sensing-based structural health assessment and high-precision photogrammetric monitoring. Full article

Passenger-gathering behavior often triggers safety incidents such as stampedes due to overcrowding, posing significant challenges to public order maintenance and passenger safety. Traditional early warning algorithms for passenger-gathering behavior typically perform only global modeling of image appearance, neglecting the analysis of individual passenger actions in practical 3D physical space, leading to high false-alarm and missed-alarm rates. To address this issue, we decompose the modeling process into two stages: human pose estimation and gathering behavior recognition. Specifically, the pose of each individual in 3D space is first estimated from images, and then fused with global features to complete the early warning. This work focuses on the former stage and aims to develop an accurate and efficient human pose estimation model capable of real-time inference on resource-constrained devices. To this end, we propose a 3D human pose estimation framework that integrates a hybrid spatio-temporal Transformer with three collaborative agents. First, a reinforcement learning-based architecture search agent is designed to adaptively select among Global Self-Attention, Window Attention, and External Attention for each block to optimize the model structure. Second, a feedback optimization agent is developed to dynamically adjust the search process, balancing exploration and convergence. Third, a quantization agent is employed that leverages quantization-aware training (QAT) to generate an INT8 deployment-ready model with minimal loss in accuracy. Experiments conducted on the Human3.6M dataset demonstrate that the proposed method achieves a mean per joint position error (MPJPE) of 42.15 mm with only 4.38 M parameters and 19.39 GFLOPs under FP32 precision, indicating substantial potential for subsequent gathering behavior recognition tasks. Full article

The rapid evolution of modern vehicles into intelligent and interconnected systems presents new complexities in both functional safety and cybersecurity. In this context, ensuring the reliability and integrity of critical sensor data, such as wheel speed inputs for anti-lock brake systems (ABS), is essential. Effective detection of wheel speed sensor faults not only improves functional safety, but also plays a vital role in keeping system resilience against potential cyber–physical threats. Although data-driven approaches have gained popularity for system development due to their ability to extract meaningful patterns from historical data, a major limitation is the lack of diverse and representative faulty datasets. This study proposes a novel dual learning model, based on Temporal Convolutional Networks (TCN), designed to accurately distinguish between normal and faulty wheel speed sensor behavior within a hardware-in-the-loop (HIL) simulation platform implemented on an FPGA. To address dataset limitations, a TruckSim–MATLAB/Simulink co-simulation environment is used to generate realistic datasets under normal operation and eight representative fault scenarios, yielding up to 5000 labeled sequences (balanced between normal and faulty behaviors) at a sampling rate of 60 Hz. Two TCN models are trained independently to learn normal and faulty dynamics, and fault decisions are made by comparing the reconstruction errors (MSE and MAE) of both models, thus avoiding manually tuned thresholds. On a test set of 1000 sequences (500 normal and 500 faulty) from the 5000 sample configuration, the proposed dual TCN framework achieves a detection accuracy of 97.8%, a precision of 96.5%, a recall of 98.2%, and an F1-score of 97.3%, outperforming a single TCN baseline, which achieves 91.4% accuracy and an 88.9% F1-score. The complete dual TCN architecture is implemented on a Xilinx ZCU102 FPGA evaluation kit (AMD, Santa Clara, CA, USA), while supporting real-time inference in the HIL loop. These results demonstrate that the proposed approach provides accurate, low-latency fault detection suitable for safety-critical ABS applications and contributes to improving both functional safety and cyber-resilience of braking systems. Full article