Edge clouds must increasingly co-serve privacy-critical streams (e.g., per-user telemetry and industrial control loops) and best-effort utility services (e.g., large language model inference and augmented-reality rendering) on the same constrained nodes to meet strict latency targets and sustain resource utilization. Operating them on disjoint server pools satisfies privacy requirements but leaves capacity restricted because private demand is substantial. Naive colocation improves utilization but cannot offer hard service-level agreements (SLAs) or data-residency guarantees. Hence, we propose DynaHyEdge, a hybrid scheduler that 1) continuously partitions capacity between private and public domains, 2) enforces per-core time isolation with microsecond domain flips, and 3) uses event-driven admission to utilize idle computational resources without preemption. This joint design maximizes on-time completion while provably meeting private-task SLAs, keeping sensitive data local, and reclaiming otherwise idle computational resources. Experiments on real-world and synthetic traces show that DynaHyEdge increases the deadline success rates, decreases the latency, and increases CPU utilization over greedy, fixed-partition, and earliest-deadline-first baselines without compromising privacy.

Catastrophic forgetting is currently the greatest challenge faced in Exemplar-Free Class Incremental Learning (EFCIL), which does not allow the replay of old data from previous tasks because of factors such as user privacy and device capacity limitations. In this paper, we propose a Comprehensive Ensemble Framework for exemplar-free Class Incremental Learning (CEFCIL), which includes an ensemble Nearest Class Mean (NCM) classifier based on the Mahalanobis metric with a given number of diversified base networks, a cached root model consisting of initialized base networks for root knowledge distillation, a dual knowledge distillation strategy, and a dimensional collapse prevention strategy. Across diverse experimental conditions, CEFCIL exhibits superior performance in EFCIL and possesses robust cross-domain capabilities.

The amount and scale of worldwide data centers grow rapidly in the era of big data, leading to massive energy consumption and formidable carbon emission. To achieve the efficient and sustainable development of information technology (IT) industry, researchers propose to schedule data or data analytics jobs to data centers with low electricity prices and carbon emission rates. However, due to the highly heterogeneous and dynamic nature of geo-distributed data centers in terms of resource capacity, electricity price, and the rate of carbon emissions, it is quite difficult to optimize the electricity cost and carbon emission of data centers over a long period. In this paper, we propose an energy-aware data backup and job scheduling method with minimal cost (EDJC) to minimize the electricity cost of geo-distributed data analytics jobs, and simultaneously ensure the long-term carbon emission budget of each data center. Specifically, we firstly design a cost-effective data backup algorithm to generate a data backup strategy that minimizes cost based on historical job requirements. After that, based on the data backup strategy, we utilize an online carbon-aware job scheduling algorithm to calculate the job scheduling strategy in each time slot. In this algorithm, we use the Lyapunov optimization to decompose the long-term job scheduling optimization problem into a series of real-time job scheduling optimization subproblems, and thereby minimize the electricity cost and satisfy the budget of carbon emission. The experimental results show that the EDJC method can significantly reduce the total electricity cost of the data center and meet the carbon emission constraints of the data center at the same time.

Decentralized Online Learning (DOL) extends online learning to the domain of distributed networks. However, limitations of local data in decentralized settings lead to a decrease in the accuracy of decisions or models compared to centralized methods. Considering the increasing requirement to achieve a high-precision model or decision with distributed data resources in a network, applying ensemble methods is attempted to achieve a superior model or decision with only transferring gradients or models. A new boosting method, namely Boosting for Distributed Online Convex Optimization (BD-OCO), is designed to realize the application of boosting in distributed scenarios. BD-OCO achieves the regret upper bound $𝒪\mathsf{}\mathrm{\left(}\frac{M\mathrm{+}N}{M\mathsf{}N}\mathsf{}T\mathrm{\right)}$, where $M$ measures the size of the distributed network and $N$ is the number of Weak Learners (WLs) in each node. The core idea of BD-OCO is to apply the local model to train a strong global one. BD-OCO is evaluated on the basis of eight different real-world datasets. Numerical results show that BD-OCO achieves excellent performance in accuracy and convergence, and is robust to the size of the distributed network.