国家预印本平台

The progressive development of particle detectors toward four-dimensional (4D) tracking imposes increasing demands on both timing and spatial resolution. While resistive AC-coupled Low Gain Avalanche Diode (AC-LGAD) sensors exhibit excellent timing performance, high-precision position reconstruction remains challenging due to complex charge sharing signals across multiple readout pads. In this work, we present an end-to-end deep learning framework for high-precision two-dimensional (2D) hit reconstruction in AC-LGADs, consisting of a one-dimensional convolutional neural network (1D-CNN) backbone and a fully connected (FC) regression head, that directly reconstructs 2D hit coordinates from multi-channel waveforms, without relying on hand-crafted features or explicit physical models. With data acquired from high-resolution transient current technique (TCT) scans, the proposed method achieves position resolutions of about 5.5 micron near the center of the square region bounded by four neighboring electrodes, and of about 9.0 micron near the edges where charge sharing is reduced, representing a substantial improvement over the conventional amplitude-based reconstruction method used as a baseline. Furthermore, the model demonstrates robustness under reduced sampling rates compatible with realistic front-end electronics, maintaining stable performance at 5 GS/s or above, with only slight deterioration observed at lower sampling rates. These results establish the 1D-CNN-based method as a powerful and flexible tool for high-precision 2D hit reconstruction in AC-LGADs and highlight the promise of deep learning approaches in advancing silicon detector technology.

The high density fuels, including uranium monocarbide UC, UN, and the uranium carbonitride U(CN) that results from carbon impurities following UN fabrication, have recently attracted attention for light-water reactor (LWR) applications because of their facilitation on increased 235 U loading at a fixed enrichment, high thermal conductivity and high melting temperatures. Despite these favorable properties, numerous performance aspects must be evaluated before any lesser-studied uranium compounds become viable LWR fuel forms. This study is first to perform thermodynamic calculation to evaluate thermal stability pure UC and UN in a closed system mimicking Pressure-water reactor (PWR) coolant, based on which UN or UC is not thermodynamically stable in any aqueous electrochemical system during a cladding breach. Then the potential interactions between UN and UCN with Zr or SiC cladding were systematically evaluated using a thermodynamic database of U-Zr-Si-C-N was developed in this work using the CALPHAD approach and validated with available literature data. The interfacial stability of UN/Zr, UC/Zr, UN/SiC and UC/SiC were assessed by calculating the isothermal sections U-Zr-N, U-Zr-C, U-Si-C, U-Si-C-N at 1500,1000 and 500 C, as well as the isopleth sections of U(C 0.3 N 0.7 )-Zr. The results predict that a complex reaction pathway between UN and Zr will produce multiple layers including phases of bcc(U,Zr), UZr2, hcp(Zr,N) and fcc(U,Zr)N. This response is more complicated than that between UC and Zr where a single ZrC layer is predicted. Improved thermodynamic stability is predicted when compatability with SiC cladding is considered. Both UN and UC are in equilibrium with SiC when modeled under the same temperature conditions. Potential carbon impurities present in UN as a result of the fabrication process were not found to contribute detrimentally to fuel-cladding contact for either Zr or SiC cladding under conditions evaluated here.

心理预算是约束支出决策的关键因素，区分其作用范围是评估约束效果的前提，但现有文献尚未对其进行系统分析。为明确心理预算对支出决策的约束空间，研究构建了包含认知标签构建、支出识别和预算参考阶段的约束过程模型。研究指出，心理预算系统的约束范围有限、约束条件可塑：那些无法归类到预算账户、未在决策前完成识别，以及效用高于超支边界的支出难以被心理预算约束；决策者可能通过调整支出的归类改变预算约束条件。而且，该系统还具备提升决策效率和收益的双重适应价值。研究完善了心理预算的理论体系，为预算约束下的支出决策动机分析提供了新框架，对个体预算管理、企业营销及公共政策制定具有参考价值。

Clustering is widely used for unsupervised structure discovery, yet it offers limited insight into how reliable each individual assignment is. Diagnostics, such as convergence behavior or objective values, may reflect global quality, but they do not indicate whether particular instances are assigned confidently, especially for initialization-sensitive algorithms like k-means. This assignment-level instability can undermine both accuracy and robustness. Ensemble approaches improve global consistency by aggregating multiple runs, but they typically lack tools for quantifying pointwise confidence in a way that combines cross-run agreement with geometric support from the learned cluster structure. We introduce CAKE (Confidence in Assignments via K-partition Ensembles), a framework that evaluates each point using two complementary statistics computed over a clustering ensemble: assignment stability and consistency of local geometric fit. These are combined into a single, interpretable score in [0,1]. Our theoretical analysis shows that CAKE remains effective under noise and separates stable from unstable points. Experiments on synthetic and real-world datasets indicate that CAKE effectively highlights ambiguous points and stable core members, providing a confidence ranking that can guide filtering or prioritization to improve clustering quality.

Streaming video understanding requires models to robustly encode, store, and retrieve information from a continuous video stream to support accurate video question answering (VQA). Existing state-of-the-art approaches rely on key-value caching to accumulate frame-level information over time, but use a limited number of tokens per frame, leading to the loss of fine-grained visual details. In this work, we propose scaling the token budget to enable more granular spatiotemporal understanding and reasoning. First, we find that current methods are ill-equipped to handle dense streams: their feature encoding causes query-frame similarity scores to increase over time, biasing retrieval toward later frames. To address this, we introduce an adaptive selection strategy that reduces token redundancy while preserving local spatiotemporal information. We further propose a training-free retrieval mixture-of-experts that leverages external models to better identify relevant frames. Our method, MemStream, achieves +8.0% on CG-Bench, +8.5% on LVBench, and +2.4% on VideoMME (Long) over ReKV with Qwen2.5-VL-7B.