Task-driven Processing with Coarse-to-Fine Glimpse-based Active Perception
本文针对高分辨率图像中特定目标检测问题,提出了一种从粗到细基于瞥视的任务驱动感知方法(CF-GAP),通过迭代聚焦相关区域提高检测精度。
本文针对高分辨率图像中特定目标检测问题,提出了一种从粗到细基于瞥视的任务驱动感知方法(CF-GAP),通过迭代聚焦相关区域提高检测精度。
This study addresses the challenge of quantifying uncertainty arising from internal climate variability, which is hindered by the prohibitive computational cost of running high-resolution, large-ensemble climate simulations. For the first time, a conditional variational autoencoder (cVAE) is applied to CMIP6 CanESM5 monthly-scale data, enhanced with an output noise injection mechanism to better capture multiscale climate variability. The proposed method generates physically consistent synthetic ensembles of arbitrary size from limited samples, accurately reproducing realistic teleconnection patterns and both low- and high-order statistical features—including extremes—even under climate conditions not present in the training data. This approach significantly improves the reliability of uncertainty assessments for both historical and future climate scenarios, while maintaining computational efficiency and mathematical interpretability.
Social media platforms are often criticized for amplifying antisocial behaviors and lacking effective mechanisms to foster prosocial tendencies such as curiosity. This study addresses this gap by constructing an independent experimental platform and conducting a randomized controlled trial with 2,282 U.S. adults in a highly controlled environment. Using AI-driven virtual users to simulate authentic social interactions, the research systematically manipulated platform social norms and interface design. Findings demonstrate that curiosity-inducing interventions significantly increased users’ question-asking frequency and textual markers of curiosity while reducing toxic language. Although these interventions decreased generalized engagement metrics—such as likes and comments—they did not adversely affect subjective user experience or time spent creating content. The study thus provides causal evidence and a practical design pathway for promoting prosocial behavior on digital platforms.
Early identification of thermal hotspots and excessive safety margins remain key challenges in induction motor thermal design. Method: This paper proposes a two-dimensional thermal model calibration approach based on inverse field problems, jointly estimating material thermal properties and equivalent parameters for three-dimensional (3D) thermal effects using only measured temperature data—without requiring prior knowledge of detailed 3D geometry. Contribution/Results: The method is the first systematic application of inverse modeling to induction motor thermal analysis. Integrated with parametric sensitivity analysis and validated against both synthetic and experimental data, it significantly reduces thermal prediction error in both academic benchmarks and real-world motors. The approach enables accurate localization of thermal weak points at early design stages, thereby facilitating reduction of unnecessary safety margins, enhancement of power density, and improvement of overall reliability.
Existing open-source multiphysics simulation tools (e.g., openCFS) lack efficient and flexible Python-based data processing capabilities for coupled-field problems such as aeroacoustics, resulting in fragmented pre- and post-processing ecosystems. To address this, we propose the first lightweight, extensible, Python-native data framework tailored for openCFS. It unifies parsing of openCFS’s native XML configuration and finite-element data formats, and integrates HDF5 I/O, NumPy/Pandas-backed computation, and object-oriented design principles. The framework bridges the cross-language gap between openCFS’s C++ core and user-facing data analysis, enabling modular, pipeline-driven preprocessing and postprocessing. It supports batched mesh analysis, real-time visualization, and machine-learning-ready data export. Evaluated across multiple aeroacoustic case studies, the framework demonstrates robustness, usability, and significant efficiency gains in end-to-end workflows.
为解决电喉语音编码问题,提出多教师知识蒸馏框架训练轻量级流式内容编码器,通过自监督学习模型和微调识别模型提供目标,降低电喉语音错误率。
证明了线性方程模n的伪心灵感应问题,通过结合希尔伯特空间子空间上的群值测度结果与量子伪心灵感应的多项式-仆从特征方法。
本文探讨了社交网络中的不平等问题,通过识别十种网络效应及其在决策过程中的系统性偏见,呼吁采用整体、动态的方法来实现公平。
研究通过在主动学习的每次迭代中整合权重剪枝方法,解决非稳定数据环境下模型稀疏化问题,从而提高计算效率。
本文针对高分辨率图像中特定目标检测问题,提出了一种从粗到细基于瞥视的任务驱动感知方法(CF-GAP),通过迭代聚焦相关区域提高检测精度。