A key challenge in federated learning (FL) is the statistical heterogeneity that impairs generalization of global model on each client. To address this, we propose a method Federated with Adaptive Local Aggregation (FedALA) by capturing desired information for client models personalized FL. The component FedALA an (ALA) module, which can adaptively aggregate downloaded and local towards objective to initialize before training iteration. evaluate effectiveness FedALA, conduct extensive...

The cloud provides low-cost and flexible IT resources (hardware software) across the Internet. As more providers seek to drive greater business outcomes environments of become complicated, it is evident that era intelligent has arrived. faces several challenges, including optimizing economic service configuration adaptively allocating resources. In particular, there a growing trend toward using machine learning improve intelligence management. This article discusses an architecture resource...

Natural graphs with skewed distributions raise unique challenges to distributed graph computation and partitioning. Existing graph-parallel systems usually use a “one-size-fits-all” design that uniformly processes all vertices, which either suffer from notable load imbalance high contention for high-degree vertices (e.g., Pregel GraphLab) or incur communication cost memory consumption even low-degree PowerGraph GraphX). In this article, we argue in natural also necessitate differentiated...

Video object detection is more challenging than image because of the deteriorated frame quality. To enhance feature representation, state-of-the-art methods propagate temporal information into by aligning and aggregating entire maps from multiple nearby frames. However, restricted map's low storage-efficiency vulnerable content-address allocation, long-term not fully stressed these methods. In this work, we propose first guided external memory network for online video detection....

Large-scale graph-structured computation usually exhibits iterative and convergence-oriented computing nature, where input data is computed iteratively until a convergence condition reached. Such features have led to the development of two different modes for programs, namely synchronous (Sync) asynchronous (Async) modes. Unfortunately, there currently no in-depth study on their execution properties thus programmers manually choose mode, either requiring deep understanding underlying graph...

Inter-data center wide area networks (inter-DC WANs) carry a significant amount of data transfers that require to be completed within certain time periods, or deadlines. However, very little work has been done guarantee such The crux is the current inter-DC WAN lacks an interface for users specify their transfer deadlines and mechanism provider ensure completion while maintaining high utilization. In this paper, we address problem by introducing deadline-based network abstraction (DNA) WANs....

With the rapid growth of video data, summarization technique plays a key role in reducing people's efforts to explore content videos by generating concise but informative summaries. Though supervised approaches have been well studied and achieved state-of-the-art performance, unsupervised methods are still highly demanded due intrinsic difficulty obtaining high-quality annotations. In this paper, we propose novel yet simple method with attentive conditional Generative Adversarial Networks...

Recently, personalized federated learning (pFL) has attracted increasing attention in privacy protection, collaborative learning, and tackling statistical heterogeneity among clients, e.g., hospitals, mobile smartphones, etc. Most existing pFL methods focus on exploiting the global information client-level model parameters while neglecting that data is source of these two kinds information. To address this, we propose Federated Conditional Policy (FedCP) method, which generates a conditional...

Virtualization poses new challenges to I/O performance. The single-root virtualization (SR-IOV) standard allows an device be shared by multiple Virtual Machines (VMs), without losing runtime We propose a generic architecture for SR-IOV devices, which can implemented on Machine Monitors (VMMs). With the support of our architecture, driver is highly portable and agnostic underlying VMM. Based first implementation network driver, we applied several optimizations reduce overhead. Then, carried...

In-memory key/value store (KV-store) is a key building block for many systems like databases and large websites. Two requirements such are efficiency availability, which demand KV-store to continuously handle millions of requests per second. A common approach availability using replication, as primary-backup (PBR), which, however, requires M +1 times memory tolerate failures. This renders scarce unable useful user jobs. article makes the first case highly available in-memory by integrating...

Inter-datacenter wide area networks (inter-DC WAN) carry a significant amount of data transfers that require to be completed within certain time periods, or deadlines. However, very little work has been done guarantee such The crux is the current inter-DC WAN lacks an interface for users specify their transfer deadlines and mechanism provider ensure completion while maintaining high utilization.

Current control flow integrity (CFI) enforcement approaches either require instrumenting application executables and even shared libraries, or are unable to defend against sophisticated attacks due relaxed security policies, both; many of them also incur high runtime overhead. This paper observes that the main obstacle providing transparent strong defense adversaries is lack sufficient information. To this end, describes FlowGuard, a lightweight, CFI approach by novel reuse Intel Processor...

Vessel segmentation is critically essential for diagnosing a series of diseases, e.g., coronary artery disease and retinal disease. However, annotating vessel maps medical images notoriously challenging due to the tiny complex structures, leading insufficient available annotated datasets existing supervised methods domain adaptation methods. The subtle structures con-fusing background further suppress efficacy unsupervised In this paper, we propose self-supervised method via adversarial...

Federated Learning (FL) is popular for its privacy-preserving and collaborative learning capabilities. Recently, personalized FL (pFL) has received attention ability to address statistical heterogeneity achieve personalization in FL. However, from the perspective of feature extraction, most existing pFL methods only focus on extracting global or information during local training, which fails meet goals pFL. To this, we propose a new method, named GPFL, simultaneously learn each client. We...

Cyclops is a new vertex-oriented graph-parallel framework for writing distributed graph analytics. Unlike existing computation models, retains simplicity and computation-efficiency by synchronously computing over immutable view, which grants vertex with read-only access to all its neighboring vertices. The view provided via read- only replication of vertices edges spanning machines during cut. follows centralized model assigning master update propagate the value replicas unidirectionally in...

Differential privacy (DP) has been widely explored in academia recently but less so industry possibly due to its strong guarantee. This paper makes the first attempt implement three basic DP architectures deployed telecommunication (telco) big data platform for mining applications. We find that all have than 5% loss of prediction accuracy when weak guarantee is adopted (e.g., budget parameter ε ≥ 3). However, assumed ≤ 0:1), lead 15% ~ 30% loss, which implies real-word industrial systems...

This paper describes a map matching program submitted to the ACM SIGSPATIAL Cup 2012. We first summarize existing algorithms into three categories, and compare their performance thoroughly. In general, global max-weight methods using Viterbi dynamic programming algorithm are most accurate but accuracy varies at different sampling intervals weight functions. Our submission selects hybrid that improves upon best two functions such its is better than both robust against varying rates. addition,...

To achieve efficient resource management on a graphics processing unit (GPU), there is demand to develop framework for scheduling virtualized resources in cloud gaming. In this article, we propose VGRIS, GPU isolation and A set of application programming interfaces (APIs) provided so that variety algorithms can be implemented within the without modifying itself. Three are by APIs VGRIS. Experimental results show VGRIS effectively schedule among various workloads.

The hybrid wireless and power line communication (HWPLC) networks address the problem that mobile sensors (PLC) cannot communicate with each other within an Internet of Things (IoT) network. In this paper, we design a relay equipped dual PLC interface, which connects both into IoT Furthermore, dual-interface forwards messages by adaptively selecting interface according to channel state. A general mathematical probability model relaying system is presented. density function output...

As the virtualization technology for GPUs matures, cloud gaming has become an emerging application among services. In addition to poor default mechanisms of GPU resource sharing, performance games is inevitably undermined by various runtime uncertainties such as rendering complex game scenarios. The question how handle sharing remains unanswered. To address this challenge, we propose vGASA, a virtualized adaptive scheduling algorithm in gaming. vGASA interposes algorithms graphics API...