The acquisition of precise building contour and height information is essential for such tasks as the assessment of photovoltaic resources in urban planning. Current deep learning-based methods still face challenges in effectively integrating features from heterogeneous data sources (e.g., optical imagery and digital surface models (DSM) and in handling irregular building boundaries produced through segmentation. To address these limitations, this paper proposes SE-Build, a multi-task network that jointly performs building segmentation, edge extraction, and height estimation. SE-Build utilizes a dual-branch encoder to independently extract features from optical imagery and DSM, and integrates a bidirectional attention feature fusion block (BAFB) to facilitate deep feature interaction and fusion. In the decoding stage, this study consisted of designing a multi-scale convolution (MConv) to balance global structure with local detail modeling, and of implementing a three-branch decoder to enable the prediction of building masks, edge lines, and height maps. Geometric constraints were applied to adjust the edge lines and enhance polygon regularity, and these adjusted contours were then combined with height maps to generate LOD1 building models. The experimental results show that SE-Build outperforms state-of-the-art methods in building extraction and 3D modeling on the Urban3D dataset, with a building extraction IoU of 78.7% and a 3D quality score of 58.8%. This demonstrates the framework's effectiveness and provides a robust technical solution for various applications, such as assessing the photovoltaic resources of buildings.
The evolution of vehicle-to-everything (V2X) communication has driven the emergence of advanced services that impose heterogeneous quality-of-service requirements. Satisfying these requirements in existing network infrastructures is challenging, owing to the highly dynamic nature of vehicular environments. To address this, the open radio access network (O-RAN) architecture has emerged as a transformative solution specifically designed to facilitate artificial intelligence (AI)-driven and programmable network operations via a RAN intelligent controller. This study presents a comprehensive survey of the applicability of O-RAN-enabled V2X communications. We systematically analyze the specific roles of non-real-time and near-real-time RAN intelligent controllers and investigated optimal AI/ML workflow deployment options. Furthermore, we examined representative V2X use cases derived from recent standards and provided a detailed analysis of mapping these use cases to O-RAN control mechanisms to address the core performance challenges.
The growing number of unverified pieces of information on social media platforms and online news sources has contributed to increasing levels of misinformation, disinformation, fake news, and propaganda. As such, understanding the characteristics of propaganda in online news sources is critical in offsetting its effects on society. In this article, we propose a framework for the automatic retrieval and analysis of propaganda pieces in online news articles through text data and computational intelligence. Given the effectiveness of the RoBERTa pre-trained language model, we set about developing a deep learning model for propaganda piece identification in online news articles through a transformer learning model combined with LSTM and CRF modules together with multi-label tagging. The model is tested on a number of different data sources including the SemEval-2020 dataset, a ProText dataset with more than 750 online news sources, and an information retrieval dataset that has 180 fact-checked articles. The results of the testing conducted on these data sources indicate the effectiveness of the model by means of F1 testing, with a result of 0.4672, 0.5164, and 0.5309 on the respective data sources. The findings of this research work are significant since they add to the development of a deep learning model on propaganda span detection with significant applications in web-based fact-checking.
The widespread adoption of intelligent voice assistants makes understanding the ethical implications of their interaction design-particularly for older adults-essential. This population is especially sensitive to conversational design decisions due to asymmetries in digital literacy, trust management, and social vulnerability. We present a controlled empirical study wherein older adults evaluated two approaches to Alexa's conversational design: one emphasizing neutrality and transparency, and another relying on emotional manipulation and opacity. We first identify eight key ethical dimensions relevant to conversational voice assistant design and develop a questionnaire to assess users' perceived ethicality in this context. Our findings provide evidence of how manipulative design strategies are associated with changes in participants' perceptions of the ethical behavior of the voice assistant. Based on these results, we propose a set of ethical design recommendations for conversational assistants aimed at promoting the ethical well-being of older adults.
Cloud-native systems increasingly rely on containerized microservices for scalable and flexible service deployment. In such environments, proactive autoscaling can mitigate latency associated with reactive scaling by allocating resources in advance. However, point-forecast-based autoscaling is vulnerable to nonstationary workloads because forecast reliability varies over time, and prediction errors affect scaling decisions. To address this problem, this work proposes prediction uncertainty-aware proactive autoscaling (PUPA), an uncertainty-aware autoscaling framework for containerized microservices. The PUPA framework integrates reversible instance normalization and segment recurrent neural networks for workload forecasting and employs adaptive signed-error conformal prediction based on adaptive conformal inference to construct time-varying prediction intervals. The critical contribution of this work is an uncertainty-aware decision mechanism that adjusts the representative workload and target utilization threshold based on predictive uncertainty. The experiments on three Alibaba trace datasets in a Kubernetes-based offline simulation revealed that PUPA improves the stability-efficiency trade-off. Compared with Kubernetes horizontal pod autoscaling, PUPA reduces the overload count by 22 to 198 occurrences while achieving a mean resource utilization of 75.97% to 81.20%.
