Performance Study of Serverless Workloads in Confidential Virtual Machines
研究探讨了保密虚拟机中无服务器工作负载的性能问题,通过实证分析内存效率与运行时开销,并提出优化方法以减少启动延迟和提高资源利用效率。
研究探讨了保密虚拟机中无服务器工作负载的性能问题,通过实证分析内存效率与运行时开销,并提出优化方法以减少启动延迟和提高资源利用效率。
研究通过自改进RAG框架,利用三个专门代理和反馈驱动的自我纠正机制解决金融问答中的数据验证和合规问题,提高了答案准确性和可解释性。
This study addresses the challenge of planning public transit services in “transit deserts”—areas lacking reliable ridership data—by proposing a risk-sensitive, incremental planning approach. The method formulates the problem as a partially observable Markov decision process (POMDP) with a conditional value-at-risk (CVaR) constraint to explicitly manage demand uncertainty. It leverages multi-source data on population, land use, and employment to construct a prior distribution over latent travel demand and employs Bayesian updating to dynamically refine belief states. Drawing inspiration from financial risk modeling, the framework innovatively incorporates tail-risk constraints and features a belief-aware myopic planner that remains robust under budget limitations and biased priors. Evaluated across 25 cities over five years, the approach reduced transit deserts by an average of 53.6%, outperforming static optimization by 5.0 percentage points, with 16 cities showing significant improvements and sustained efficacy even under 50% prior error.
This work addresses the vulnerability of vision-language and generative image models to geometric transformations such as input rotation, which can degrade robustness and exacerbate demographic biases. The study presents the first systematic investigation into how rotation perturbations jointly impact fairness and robustness in multimodal models. To mitigate these issues, the authors propose an integrated framework combining data augmentation, representation alignment, and model regularization. Evaluated across multiple benchmark datasets, the approach substantially enhances robustness to rotation, effectively curbs bias amplification, and maintains or even improves overall task performance, thereby achieving a synergistic optimization of robustness, fairness, and accuracy.
To address the excessive memory overhead of the D-basis algorithm in attribute implication analysis, this paper proposes Small Space, a space-efficient hypergraph dualization method. The approach abandons full rule storage and instead adopts a backward-search-based hypergraph dualization framework, dynamically and incrementally counting the frequency of antecedents for each consequent attribute while retaining only essential intermediate states. By tightly integrating the structural properties of the D-basis with hypergraph dualization theory—and focusing specifically on antecedent frequency as the key statistic—it achieves substantial memory reduction. Experimental results demonstrate that Small Space reduces peak memory consumption by up to several orders of magnitude, while maintaining nearly unchanged runtime performance. This enables scalable attribute implication analysis on large-scale datasets.
研究探讨了保密虚拟机中无服务器工作负载的性能问题,通过实证分析内存效率与运行时开销,并提出优化方法以减少启动延迟和提高资源利用效率。
研究通过自改进RAG框架,利用三个专门代理和反馈驱动的自我纠正机制解决金融问答中的数据验证和合规问题,提高了答案准确性和可解释性。
This study addresses the challenge of planning public transit services in “transit deserts”—areas lacking reliable ridership data—by proposing a risk-sensitive, incremental planning approach. The method formulates the problem as a partially observable Markov decision process (POMDP) with a conditional value-at-risk (CVaR) constraint to explicitly manage demand uncertainty. It leverages multi-source data on population, land use, and employment to construct a prior distribution over latent travel demand and employs Bayesian updating to dynamically refine belief states. Drawing inspiration from financial risk modeling, the framework innovatively incorporates tail-risk constraints and features a belief-aware myopic planner that remains robust under budget limitations and biased priors. Evaluated across 25 cities over five years, the approach reduced transit deserts by an average of 53.6%, outperforming static optimization by 5.0 percentage points, with 16 cities showing significant improvements and sustained efficacy even under 50% prior error.
This work addresses the vulnerability of vision-language and generative image models to geometric transformations such as input rotation, which can degrade robustness and exacerbate demographic biases. The study presents the first systematic investigation into how rotation perturbations jointly impact fairness and robustness in multimodal models. To mitigate these issues, the authors propose an integrated framework combining data augmentation, representation alignment, and model regularization. Evaluated across multiple benchmark datasets, the approach substantially enhances robustness to rotation, effectively curbs bias amplification, and maintains or even improves overall task performance, thereby achieving a synergistic optimization of robustness, fairness, and accuracy.
To address the excessive memory overhead of the D-basis algorithm in attribute implication analysis, this paper proposes Small Space, a space-efficient hypergraph dualization method. The approach abandons full rule storage and instead adopts a backward-search-based hypergraph dualization framework, dynamically and incrementally counting the frequency of antecedents for each consequent attribute while retaining only essential intermediate states. By tightly integrating the structural properties of the D-basis with hypergraph dualization theory—and focusing specifically on antecedent frequency as the key statistic—it achieves substantial memory reduction. Experimental results demonstrate that Small Space reduces peak memory consumption by up to several orders of magnitude, while maintaining nearly unchanged runtime performance. This enables scalable attribute implication analysis on large-scale datasets.