Bird's-eye-view (BEV) occupancy prediction estimates 3D occupied space from sequential sensor data, providing the environment model that underpins downstream planning and decision-making in autonomous driving. Existing methods often rely on dense fusion or naive feature stacking, inflating compute and memory, yielding poorly calibrated probabilities, and training brittleness under occlusion and long-tail categories. We propose PRISM-Occ, a dual-level sparse Mixture-of-Experts framework for multi-modal BEV occupancy. A path-routed hierarchical router (PRHR) with Sparse Top-K activates only a compact set of experts within and across modalities, reducing parameter count while sharpening specialization. A heteroscedastic occupancy head predicts a spatial temperature map to improve calibration, and a simple prior adjustment with a staged hard-sample schedule stabilizes training under occlusion and rare classes. On Occ3D-nuScenes and SurroundOcc, PRISM-Occ achieves state-of-the-art accuracy and better-calibrated probabilities using single-scale 256× 704 inputs and fixed, lower-resolution backbones, delivering a stronger accuracy–efficiency trade-off with reduced parameters and comparable runtime memory.

We present FGO-SLAM++, a real-time geometry-aware Gaussian SLAM system capable of processing video streams from arbitrary modalities. The system performs multi-view consistent map reconstruction by maintaining a continuous opacity field. Existing Gaussian SLAM systems face challenges in achieving efficient tracking and mapping while supporting multiple input modalities, which often necessitates a trade-off between rendering quality and geometric accuracy. This study demonstrates that it is possible to satisfy all these requirements simultaneously using video streams of arbitrary modalities. The core of the proposed method involves explicit geometric feature extraction to capture the underlying scene structure and estimate camera poses, followed by a Gaussian-based ray tracing strategy to construct and optimize an opacity field. Upon the detection of loop closures, the system performs global adjustment to enhance map consistency. Furthermore, the surface is directly extracted using the marching tetrahedra method and refined through a geometric constraint field. Extensive experiments demonstrate that the proposed method achieves superior performance in tracking accuracy, rendering quality, geometric reconstruction, and real-time efficiency.

The graph convolutional network (GCN) has demonstrated effectiveness well in the text classification task. However, inadequate handling of uncertainty in prediction results exists due to the under-utilization of text features extracted by a single deep-learning model. To mitigate the potential risk of text misclassification, we proposed an enhanced GCN model for text classification based on three-way decision, incorporating shadowed set theory (3WD-GCN). In this approach, we first employ GCN as a primary classifier to handle textual data, obtaining the initial predicted results and the membership matrix. Depending on the idea of processing in threes, these results were divided into acceptance, rejection, and subdivision regions, respectively. For the subdivision region, we introduce SVM as a secondary classifier to process objects with poor conformability and distinguishability, which can reduce the uncertainty of prediction results and improve the overall performance of text classification. A series of experiments based on several benchmark datasets extensively evaluated the proposed method. The results demonstrate the validity of the approach and show a significant improvement over popular baseline text classification models.

This study presents a task affinity scheduling framework that leverages multidimensional resource fingerprints to optimize resource management in cloud computing. The framework evaluates task affinity through complementarity and competitiveness analyses by constructing dynamic fingerprints from high-dimensional resource usage data. A novel algorithm captures temporal resource-usage patterns, enabling an adaptive scheduling strategy to improve resource utilization and performance. Experiments with real-world cloud traces demonstrated superior resource utilization, reduced task completion times, and improved quality of service under high-load conditions compared to existing methods. This approach offers a scalable solution for fine-grained resource management in a dynamic cloud environment.

Discovering valuable knowledge from massive time series data is challenging due to sophisticated temporal relationships and inherent uncertainties. This paper proposes a new multigranularity fuzzy association analysis of multigranularity incorporated with three-way decisions inspired by humans. In particular, different fuzzy association rules are first mined at multiple time granularities. A three-way decision model is then designed to evaluate the credibility of each rule as a “positive”, “negative”, or “unknown” correlation. Further, we propose a novel deep learning approach to integrate the three-way fuzzy decisions across different granularities. By integrating the three-way decisions, more comprehensive and reliable fuzzy association knowledge can be obtained from Big Data from time series at different granularities, achieving more comprehensive and reliable discoveries than existing techniques. Extensive experiments demonstrate significant performance gains in real-world data sets from various domains. The synergistic integration of multigranularity mining, three-way decisions, and deep learning underpins a new problem solving paradigm to advance temporal knowledge discovery.