Large-scale vision-language models enable powerful cross-modal understanding and generation, driving rapidly growing demand for online inference services. However, cloud-centric serving often suffers from high latency, rising costs, and network dependency, while purely on-device deployment is constrained by limited memory and reduced accuracy on complex tasks. To address this accuracy–latency–cost trilemma, we propose ShiftVL, a task-aware end–cloud serving framework that shifts suitable execution to the end device with a cloud fallback. ShiftVL serves high-frequency requests on an end-side small VLM enhanced with ViTexLoRA, a modality-disentangled parameter-efficient tuning method that preserves cross-modal alignment, while routing low-frequency or complex requests to a cloud-hosted large VLM for higher accuracy. Under tight device budgets, ShiftVL employs a predictive adapter scheduler that combines LRU-style caching with imitation learning to pre-load task-specific adapters. Experiments with InternVL models show that ShiftVL reduces cloud cost by up to 76.3% and latency by up to 42.9% while maintaining high multi-task accuracy, demonstrating its practicality for real-world vision-language model serving.

Unsupervised anomaly detection in cloud computing is crucial for system security and efficiency. However, the challenges posed by large data volumes, low anomaly rates, and diverse anomaly patterns in time series within cloud computing scenarios make it difficult for previous methods to obtain consistent and reliable representations for distinguishing anomalies. To avoid the degradation of model representation ability caused by abnormal sparsity, we propose the Density Transformer, a novel reconstruction-based explicit association modeling model that can amplify the non-trivial correlation of abnormal points with adjacent time points. Specifically, we express the density association by calculating the kernel density estimate at each time point, and the series association by calculating the self-attention at each time point. Then, the model uses an adversarial training strategy to produce a more significant difference in "association discrepancy" between normal points and abnormal points, thereby ensuring robust results in anomaly detection. Our model has been rigorously evaluated on a comprehensive collection of 6 publicly available real-world datasets, and the Density Transformer can achieve up to 46% improvement in F1-score compared to existing methods.

The guard channel scheme in wireless mobile networks has attracted and is still drawing research interest owing to easy implementation and flexible control. However guard channel schemes can not adapt to changing traffic loads because of static reserved guard channels. Therefore dynamic guard channel schemes have been proposed in the literature to adapt to varying traffic load. This paper presents a novel control-theoretic approach to dynamically reserve guard channels called PI-guard channel (PI-GC) controller. Experiments show that our proposed scheme can maintain the handoff blocking probability (HBP) to a predefined value while it still improves the channel resource utilization.

In the multi-service wireless networks, the bandwidth asymmetry is the distinctive characteristic. To avoid wasting uplink and downlink bandwidth in asymmetric network, admissible region for the nonreal-time (NRT) calls has been proposed. But in that scheme the static guard capacity for RT and NRT calls cannot adapt to traffic variation. In addition, the differentiated QoS services to users are not considered. In this paper, we propose an adaptive guard channel reservation (AGCR) scheme, which has the following novelties: i) the guard capacity for each class is dynamically adjusted so as to adapt to dynamic traffic, ii) the blocking probabilities of classes with higher priority are kept below certain targets and iii) differentiated QoS services to users are provided by maintaining the blocking probability ratios over different classes at the predefined targets. Simulation results have demonstrated the proposed AGCR scheme can keep the blocking probabilities of classes with higher priority below certain targets, track the ratios close to the targets well while maintaining a high bandwidth utilization.

This paper presents a fuzzy control approach that guarantees absolute delays in Web servers. Previous work has proposed the use of classical PI controllers for delay guarantees. However, a disadvantage of the classical PI controller is that the system model, which is obtained by system identification, mismatches the real system and inevitably degrades the performance of the Web system. In contrast with classical PI controllers, fuzzy controllers are nonlinear and therefore independent of the accurate model of the plant, i.e. the controlled system. Hence, fuzzy controllers seem to be very suitable for Web servers. Our experiments show that fuzzy controllers indeed perform better than PI controllers presented in earlier papers.

Handoff is one of the key elements in ensuring quality of service (QoS) in mobile wireless networks. Handoff connections generally have higher priority than new connections. Traditional reservation policies that reserve some channels for handoff connections are not adaptive to traffic load changes. This paper proposes a new dynamic guard channel scheme (DGCS), which 1) adapts to various traffic loads; 2) combines differentiated QoS service model and priority handoff mechanism; 3) provides fairness for differentiated QoS services; 4) does not need to exchange state information among different cells, so it is easy to be implemented and is simple enough to be used in real time environments; 5) and utilizes network resources efficiently and puts a bound on each service blocking probability. The simulation results show that the ratios among different QoS service probabilities are guaranteed to be predefined values and system utilization is improved greatly.