To accelerate the inference of machine-learning (ML) model serving, clusters of machines require the use of expensive hardware accelerators (e.g., GPUs) to reduce execution time. Advanced inference serving systems are needed to satisfy latency service-level objectives (SLOs) in a cost-effective manner. Novel autoscaling mechanisms that greedily minimize the number of service instances while ensuring SLO compliance are helpful. However, we find that it is not adequate to guarantee cost effectiveness across heterogeneous GPU hardware, and this does not maximize resource utilization. In this paper, we propose HetSev to address these challenges by incorporating heterogeneity-aware autoscaling and resource-efficient scheduling to achieve cost effectiveness. We develop an autoscaling mechanism which accounts for SLO compliance and GPU heterogeneity, thus provisioning the appropriate type and number of instances to guarantee cost effectiveness. We leverage multi-tenant inference to improve GPU resource utilization, while alleviating inter-tenant interference by avoiding the co-location of identical ML instances on the same GPU during placement decisions. HetSev is integrated into Kubernetes and deployed onto a heterogeneous GPU cluster. We evaluated the performance of HetSev using several representative ML models. Compared with default Kubernetes, HetSev reduces resource cost by up to 2.15× while meeting SLO requirements.

Cross-modal retrieval aims to elucidate information fusion, imitate human learning, and advance the field. Although previous reviews have primarily focused on binary and real-value coding methods, there is a scarcity of techniques grounded in deep representation learning. In this paper, we concentrated on harmonizing cross-modal representation learning and the full-cycle modeling of high-level semantic associations between vision and language, diverging from traditional statistical methods. We systematically categorized and summarized the challenges and open issues in implementing current technologies and investigated the pipeline of cross-modal retrieval, including pre-processing, feature engineering, pre-training tasks, encoding, cross-modal interaction, decoding, model optimization, and a unified architecture. Furthermore, we propose benchmark datasets and evaluation metrics to assist researchers in keeping pace with cross-modal retrieval advancements. By incorporating recent innovative works, we offer a perspective on potential advancements in cross-modal retrieval.

In large-scale distributed storage simulations, disk simulation plays a critical role in evaluating system reliability, scalability, and performance. However, the existing virtual disk technologies face challenges in supporting ultra-large capacities and high-concurrency workloads under constrained physical resources. To address this limitation, we propose an overcapacity mapping (OCM) virtual disk technology that substantially reduces simulation costs while preserving functionality similar to real physical disks. OCM integrates thin provisioning and data deduplication at the Linux Device Mapper layer to construct virtual disks whose logical capacities greatly exceed their physical capacities. We further introduce an SSD-based tiered asynchronous I/O strategy to mitigate performance bottlenecks under high-concurrency random read/write workloads. Our experimental results show that OCM achieves substantial space savings in scenarios with data duplication. In high-concurrency workloads involving small-block random I/O, cache acceleration yields up to 7.8× write speedup and 248.2× read speedup. Moreover, we deploy OCM in a Kubernetes environment to construct a Ceph system with 3 PB logical capacity using only 8.8 TB of physical resources, achieving 98.36% disk cost savings.

Crisis events generate massive information flows from diverse sources, which need to be consolidated into public opinion reports to enable timely response by governments and enterprises. Current LLMs, despite strong generation capabilities, fail to achieve perspective diversity, maintain factual consistency, and perform coherent high-level planning. To address these gaps, we propose MPSR: a multi-perspective self-reflection framework. Our framework first assigns diverse stakeholder personas to agents who independently generate initial writing plans from complementary viewpoints. Subsequently, a three-stage debate mechanism refines these plans by identifying conflicts, formulates resolution strategies, and produces a consensus plan, thereby enhancing factual consistency. Finally, we introduce a Report Fusion mechanism to synthesize reports across temporal batches, ensuring comprehensive event coverage. Extensive experiments demonstrate that MPSR significantly outperforms baselines, achieving Date F1 of 0.67, G-Eval of 4.54, and MiniCheck score of 79.43, which represent improvements of 17.5%, 70.0%, and 25.8% over the strongest baseline, respectively.

Public Opinion Reports are essential tools for crisis management, yet their evaluation remains a critical bottleneck that often delays response actions. Recently, dominant Large Language Model (LLM)-based evaluators often overlook a critical challenge: highly open-ended dimensions such as “innovation” and “feasibility” require synthesizing diverse stakeholder perspectives, as different groups judge these qualities from fundamentally different perspectives. Motivated by this, we propose the Role-based Adaptive Evaluation (RAE) framework. This framework employs an adaptive mechanism leveraging multi-perspective evaluation insights through role-based analysis, and further introduces dynamically generated roles tailored to specific contexts for these dimensions. RAE further incorporates multi-role reasoning aggregation to minimize individual biases and enhance evaluation robustness. Extensive experiments demonstrate that RAE significantly improves alignment with human expert judgments, especially on challenging highly open-ended dimensions.

With the development of information technology, the speed and scope of dissemination have been greatly improved, the content of information directly affects the social order, and the mining of communication laws has become a hot issue. At present, there are three common analytical models, which are based on the popularity curve fitting, interpersonal networks simulation, and machine learning models. Although these models play a major role in diffusion effects analysis, there are also problems such as the data analysis results are too dependent on the data and the versatility is weak. Based on the classical Schramm's eight elements theory and previous research results, in this study, we identified evolution curve usually has multiple peaks, and assumed a composite decomposition function to establish a mathematical abstraction. Draws on the denoising autoencoder neural network model, a multi-scale sliding-window model is proposed for fitting the energy variety curve with the number of microblogs as the influence index. The theoretical basis of the model makes it interpretable, and because of the use of a large number of known news dissemination data for training, so the versatility is better, and the denoising characteristics of the autoencoder model weaken the intensity of data cleaning. Finally, compared with the SIR function and SEIR function, the experiment shows that the fitting error of the function obtained from the model is significantly reduced. However, the model’s algorithm needs to be improved in parallel processing, and the model itself needs to be further optimized.