Applying federated learning (FL) on Internet of Things (IoT) devices is necessitated by the large volumes data they produce and growing concerns privacy. However, there are three challenges that need to be addressed make FL efficient: 1) execution with limited computational capabilities; 2) accounting for stragglers due heterogeneity devices; 3) adaptation changing network bandwidths. This article presents <monospace xmlns:mml="http://www.w3.org/1998/Math/MathML"...

Increasingly high performance computing (HPC) application developers are opting to use cloud resources due higher availability. Virtualized GPUs would be an obvious and attractive option for HPC using hosting services. Unfortunately, existing GPU virtualization software is not ready address fairness, utilization, limitations associated with consolidating mixed workloads. This paper presents FairGV, a radically redesigned system that achieves system-wide weighted fair sharing strong isolation...

Efficiently running federated learning (FL) on resource-constrained devices is challenging since they are required to train computationally intensive deep neural networks (DNN) independently. DNN partitioning-based FL (DPFL) has been proposed as one mechanism accelerate training where the layers of a (or computation) offloaded from device server. However, this creates significant communication overheads intermediate activation and gradient need be transferred between server during training....

Deep neural networks (DNNs) underpin many machine learning applications. Production quality DNN models achieve high inference accuracy by training millions of parameters which has a significant resource footprint. This presents challenge for resources operating at the extreme edge network, such as mobile and embedded devices that have limited computational memory resources. To address this, are pruned to create lightweight, more suitable variants these devices. Existing pruning methods...

The density-based clustering algorithm is a fundamental data technique with many real-world applications. However, when the database frequently changed, how to effectively update results rather than reclustering from scratch remains challenging task. In this work, we introduce IncAnyDBC, unique parallel incremental approach deal problem. First, IncAnyDBC can process changes in <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">bulks</i>...

Software product line (SPL) engineering has emerged to become a mature domain for maximizing reuse within the context of family related software products. Within process SPL, variability and commonality among different products scope is captured modeled into system's `feature model'. Currently, there are no architecture description languages (ADLs) that support relationship between feature model system domain, leaving gap which significantly increases complexity analyzing insuring it...

The satisfiability problem is known to be NP-Complete; therefore, there should relatively small instances that take a very long time solve. However, most of the smaller benchmarks were once thought challenging, especially satisfiable ones, can processed quickly by modern SAT-solvers. We describe and make available generator produces both unsatisfiable and, more significantly, formulae longer solve than any others known. At two recent international SAT Competitions, smallest unsolved created...

Product line software engineering depends on capturing the commonality and variability within a family of products, typically using feature modeling, this information to evolve generic reference architecture for family. For embedded systems, possible in hardware operating system platforms is an added complication. The design process can be facilitated by first exploring behavior associated with features. In paper we outline bidirectional modeling scheme that supports capture platform...

Fog computing has emerged as a paradigm aimed at addressing the issues of latency, bandwidth and privacy when mobile devices are communicating with remote cloud services. The concept is to offload compute services closer data. However many challenges exist in realisation this approach. During offloading, (part of) application underpinned by may be unavailable, which user will experience down time. This paper describes work building models allow prediction such time based on metrics...

Federated learning (FL) is a privacy-preserving distributed machine technique that trains models while keeping all the original data generated on devices locally. Since may be resource-constrained, offloading can used to improve FL performance by transferring computational workload from edge servers. However, due mobility, participating in leave network during training and need connect different server. This challenging because offloaded computations an server migrated. In line with this...

Fog computing offloads latency critical services of a Cloud application onto resources located at the edge network that are in close proximity to end-user devices. The research this paper is motivated towards characterising and estimating time taken offload service using containers, which investigated context 'Save Load' container migration technique. To end, addresses questions such as whether fog offloading can be accurately modelled system related parameters influence offloading. These...

Summary We present a rigorous methodology and new metrics for fair comparison of server microserver platforms. Deploying our metrics, we compare with ARM cores against two servers ×86 running the same real‐time financial analytics workload. define workload‐specific but platform‐independent performance platform comparison, targeting both datacenter operators end users. Our establishes that based on Xeon Phi co‐processor delivers highest energy efficiency. However, by scaling out...

For some time, the satisfiability formulae that have been most difficult to solve for their size crafted be unsatisfiable by use of cardinality constraints. Recent solvers introduced explicit checking such constraints, rendering previously trivial solve. A family is described derived from sgen4 but cannot solved using constraints detection and reasoning alone. These were found during SAT2014 competition a significant margin include shortest unsolved benchmark in competition, sgen6-1200-5-1.cnf.

Deep Neural Networks (DNNs) are an application class that benefit from being distributed across the edge and cloud. A DNN is partitioned such specific layers of deployed onto cloud to meet performance privacy objectives. However, there limited understanding of: whether how evolving operational conditions (increased CPU memory utilization at or reduced data transfer rates between cloud) affect already DNNs, a new partition configuration required maximize performance. adapts changing referred...