Personalized Machine Learning (PML) is a branch of ML where models are built to make predictions for an individual by analyzing the behavior of people similar to that individual. In healthcare applications, the data used for training PML models are often given as a large number (e.g., corresponding to thousands of patients) of very long, unsynchronized time series, e.g., of vital signs of patients. Capturing the biological similarities between humans (in terms of body vital signs) is much more intricate than capturing similarities between their preferences towards things such as books, airlines, grocery items, cars, sports teams, etc. Therefore, a good accuracy of PML models may sometimes be achieved only if models are trained with the data of each individual separately. This, however, translates into a prohibitive computational cost in the case of applications dealing with large populations. In this paper, we evaluate the tradeoffs between accuracy and computational cost of PML models when used in the context of healthcare. Specifically, we consider the case of predicting glycemia levels using a data set of time series of glucose values of 16 people. We use dynamic time warping (DTW) to generate three clusters of the given 16 time series. For each individual, we train two LSTM models, one using the original time series and one using the centroid of the cluster containing the original time series of that individual. Our study reveals that DTW clustering is a good alternative to generate clusters of time series that can be used to train PML models with an accuracy comparable to individually trained PML models.

People may want to monitor their health conditions while they conduct their daily activities. For this, they need solutions that perform continuous health monitoring. Today, most existing solutions in the market (e.g., Apple watch, SpO2 ring) only report current vital sign measurements; they do not predict future values of those vital signs. We developed two predictive models that give people (patients or non-patients) predictions regarding the measurements of two critical vital signs, namely, heart rate and Oxygen saturation level. Our experimental results indicate that both models have a good accuracy.

The proliferation of private clouds that are often underutilized and the tremendous computational potential of these clouds when combined has recently brought forth the idea of volunteer cloud computing (VCC), a computing model where cloud owners contribute underutilized computing and/or storage resources on their clouds to support the execution of applications of other members in the community. This model is particularly suitable to solve big data scientific problems. Scientists in data-intensive scientific fields increasingly recognize that sharing volunteered resources from several clouds is a cost-effective alternative to solve many complex, data- and/or compute-intensive science problems. Despite the promise of the idea of VCC, it still remains at the vision stage at best. Challenges include the heterogeneity and autonomy of member clouds, access control and security, complex inter-cloud virtual machine scheduling, etc. In this paper, we present CloudFinder, a system that supports the efficient execution of big data workloads on volunteered federated clouds (VFCs). Our evaluation of the system indicates that VFCs are a promising cost-effective approach to enable big data science.

Software as a service is a well accepted software deployment and distribution model that has grown exponentially in the last few years. One of the biggest benefits of SaaS is the automated composition of different services in a composite system. It allows users to automatically find and bind these services (to maximize the productivity), meeting both functional and non-functional requirements. In this paper, we present a framework for modeling the dependency relationships among different Quality of Service parameters of the component services. Our proposed approach considers the different invocation patterns of component services, and presents a service composition framework to model the dependencies and uses the global QoS for service selection. We evaluate the efficiency of our proposed technique on the WSDream-QoS Dataset [1] .

Hardware failures in cloud data centers may cause substantial losses to cloud providers and cloud users. Therefore, the ability to accurately predict when failures occur is of paramount importance. In this paper, we present FailureSim, a simulator based on CloudSim that supports failure prediction. FailureSim obtains performance related information from the cloud and classifies the status of the hardware using a neural network. Performance information is read from hosting hardware and stored in a variable length windowing vector. At specified stages, various aggregation methods are applied to the windowing vector to obtain a single vector that is fed as input to the trained classification algorithm. Using conservative host failure behavior models, FailureSim was able to successfully predict host failure in the cloud with roughly 89% accuracy.

We provided an XACML-based implementation of a semantic-based privacy management framework that incorporates context into dynamic rule evaluation and decision enforcement. Our evaluation results are promising, and we believe that future enhancements on the current implementation can provide a foundation for modern health records infrastructures and inspire collaborative data sharing.

The rise of the Internet of Things, social networking, and embedded connectivity has led to an explosion of available data. In order to better analyze this big data, many different tools have been created that can process the data efficiently. However, the increase in the amount of tools available makes it more difficult to determine which one will provide the most efficient solution to a given big data problem. In this paper, we present a delegation system that takes various frameworks and problem parameters as input and computes the best framework to use for a specific big data problem. To evaluate our system, we used two big data processing frameworks, namely, Hadoop MapReduce and AJIRA, with problem size as an input parameter. Preliminary results show that the system is able to select the most optimal big data processing framework for a given problem 90% of the time. Moreover, the proposed delegation system introduces only an additional 1% overhead when compared to the individual framework in terms of execution time.

In this article, we propose a Statistical Cloud-Assisted Reputation Estimation (sCARE) approach for service-oriented environments in uncertain situations. sCARE uses the ratings from cooperating service consumers to uniformly describe the randomness and fuzziness of the different submitted ratings and their associated relationships in quantitative terms. We also define discriminant functions to model the honesty (or lack thereof) of the service raters. Experiment results show that our proposed model performs in a fairly accurate manner for a number of real-world scenarios. A comparison study with similar existing works is also provided to asses sCARE’s performance.

In recent years, significant progress has been made to improve the power efficiency of mobile devices. In particular, new GPU architectures have made it possible to run compute-intensive applications directly on battery-powered mobile devices. In parallel, research is also being conducted in the area of application offloading, the process of running compute-intensive tasks on cloud servers and delivering the results of these computations to mobile devices through their wireless interfaces. It is important to understand the power consumption implications of each of these two options. In this paper, we use mobile cloud gaming as an example to evaluate and compare these two alternatives (running games on the cloud or on mobile devices.) Based on this comparison, we introduce the concept of Visualization as a Service (VaaS) as a new model to design and implement graphics-intensive applications for mobile devices. In this model, advanced visualization capabilities (e. g., interactive visualization of high resolution videos/images) would be provided to mobile users as a service via the Internet. We show through actual hardware specifications that, despite the recent introduction of ultra low power GPUs for mobile devices, it remains far more power efficient to offload graphics-intensive tasks to the cloud. The associated latency can still be tolerated in most applications.

Some fairly recent research has focused on providing XACML-based solutions for dynamic privacy policy management. In this regard, a number of works have provided enhancements to the performance of XACML policy enforcement point (PEP) component, but very few have focused on enhancing the accuracy of that component. This paper improves the accuracy of an XACML PEP by filling some gaps in the existing works. In particular, dynamically incorporating user access context into the privacy policy decision, and its enforcement. We provide an XACML-based implementation of a dynamic privacy policy management framework and an evaluation of the applicability of our system in comparison to some of the existing approaches.

In volunteer cloud federations (VCFs), volunteers join and leave without restrictions and may collectively contribute a large number of heterogeneous virtual machine instances. A challenge is to efficiently allocate this dynamic, heterogeneous capacity to a flow of incoming virtual machine (VM) instantiation requests, i.e., maximize the number of virtual machines that may be placed on the VCF. Cloud federations may allocate VMs far more efficiently if they can accurately predict the demand in terms of VM instantiation requests. In this paper, we present a stochastic technique that forecasts future demand to efficiently allocate VMs to VM instantiation requests. Our approach uses a Markov Chain Monte Carlo (MCMC) simulation known as the Poisson-Gamma Gibbs (PGG) sampler. The PGG sampler is used to determine the arrival rate of each type of VM instantiation requests. This arrival rate is then used to determine an optimal VM placement for the incoming VM instantiation requests. We compared our approach to a solution that adopts a static smallest-fit approach. The experimental results showed that our solution reacts quickly to abrupt changes in the frequency of VM instantiation requests and performs 10% better than the static smallest-fit approach in terms of the total number of satisfied requests.

The exponential growth of information has prompted the introduction of new technologies such as Semantic Web and Common Sense knowledge bases. To connect the different knowledge presentations together is a primary requirement, and ontologies are central we need for this transformation. In this paper we introduce ConceptOnto which is an ontology based on the ConceptNet knowledge base with extension of some of the other properties in some of the more acclaimed upper ontologies. Our goal in the creation of ConceptOnto is readability for humans, and maximizing the functionality while saving the generality of the ontology.

Extracting Ontop mapping rules and OWL ontology manually from a relational schema is a tedious task. We present an automatic approach for extracting Ontop mappings and OWL ontology from an existing database schema. The end users can access the underlying data source through SPARQL queries. A SPARQL

Volunteer cloud federations (VCFs) are cloud federations where clouds may join and leave a federation without restrictions and may contribute resources to the federation without long term commitment. This makes it difficult to predict the long term availability of resources. Also, in IaaS VCFs, volunteers may collectively contribute a large number of heterogeneous virtual machine instances. In this paper, we focus on the problem of efficiently allocating this dynamic, heterogeneous capacity to a flow of incoming VM instantiation requests. We propose an approach, called stochastic least differential capacity (SLDC),that allows over-provisioning only when necessary. The approach uses historical information about recent instantiation requests to derive stochastic predictions regarding future demand. We implemented VCFSim, a VCF simulator that uses the proposed resource allocation solution. The results of the experimental evaluation show that the proposed approach is able to improve the success rate of VM instantiation requests by up to 38%compared to an approach that uses exact matching with no demand forecasting.

A modern organization's day to day operations rely heavily on the proper functioning of its IT systems. Increasingly, most of these systems are starting to be based on the Service Oriented Architectures (SOAs). SOAs enable the automatic creation of business applications from independently developed and deployable Web services. When a SOA component suffers failure it results in sub-par operational performance and ultimately hampers the revenue stream of the organization. There are two approaches to minimize the failure risk in SOA: (a) Fault Intolerance tries to eliminate the faults before it occurs in a system. (b) Fault-Tolerance techniques deal with faults that occur on run time and try averting that faults from turning into system failures. Most of the current fault management techniques focus on the latter. However, in our study we develop a hybrid model. The method proposed in this paper (FT-SOA) predicts the failure based on HSMM and clustering techniques for SOAs and Web services. We focus on failure prediction because it plays a major role in having a highly available system. The objective of the online fault prediction is to predict the occurrence of faults in the system, based on the current information such as: service weight, previous history and reputation. A new prediction method is created to assess the fault likelihood in the near future execution of a specific service in a composite solution. We compare our approach with similar techniques to show the applicability of our approach.

Due to the many advantages offered by the cloud computing paradigm, it is fast becoming an enabling technology for many organizations, and even individual users. Flexibility and availability are two of the most important features that promote the wide spread adoption of this technology. In cloud-based data storage scenarios, where the data is controlled by a third party (i.e. the cloud service provider), the data owner usually does not have full control of its data at all stages. Consequently, this poses a prime security threat, and a major challenge is the development of a secure protocol for data storage, sharing, and retrieval. In recent years, a number of research works have targeted this problem. In this paper, we discuss some of the major approaches for secure data sharing in the cloud computing environment. The goal is to provide a concise survey of existing solutions, discuss their benefits, and point out any shortcomings for future research. Specifically, we focus on the use of encryption schemes, and provide a comparative study of the major schemes, through implementation of some representative frameworks.

Forecasting dust storms for large geographical areas with high resolution poses great challenges for scientific and computational research. Limitations of computing power and the scalability of parallel systems preclude an immediate solution to such challenges. This article reports our research on using adaptively coupled models to resolve the computational challenges and enable the computability of dust storm forecasting by dividing the large geographical domain into multiple subdomains based on spatiotemporal distributions of the dust storm. A dust storm model (Eta-8bin) performs a quick forecasting with low resolution (22 km) to identify potential hotspots with high dust concentration. A finer model, non-hydrostatic mesoscale model (NMM-dust) performs high-resolution (3 km) forecasting over the much smaller hotspots in parallel to reduce computational requirements and computing time. We also adopted spatiotemporal principles among computing resources and subdomains to optimize parallel systems and improve the performance of high-resolution NMM-dust model. This research enabled the computability of high-resolution, large-area dust storm forecasting using the adaptively coupled execution of the two models Eta-8bin and NMM-dust.

We propose a novel technique called dimming to improve the energy efficiency of cellular networks by reducing the capacity, services, and energy consumption of cells without turning off the cells. We define three basic methods to dim the network: coverage, frequency, and service dimming. We construct a multi-time period optimization problem to implement frequency dimming and extend it to implement both frequency and service dimming together. We illustrate the ability of dimming techniques to adapt the capacity and network services in proportion to the dynamic spatial and temporal load resulting in significant energy savings through numerical results for a sample network.

In most existing formulations of the problem of routing delay-constrained traffic, two important aspects have been largely overlooked: (i) delays caused by the MAC layer, and (ii) delays caused by packet queuing at the different nodes in a path. While the former aspect has been considered in some recent studies, the latter has yet to be investigated. In this paper, we present CARP, a context-aware protocol for routing delay-constrained traffic in sensor networks. CARP dynamically adapts each node's routing behavior to the current context of that node, i.e., its own routing load as well as the routing load of other nodes in its neighborhhod. We implemented CARP on top of TinyOS and conducted several experiments that confirmed CARP's efficiency in terms of routing delay-constrained traffic.

We present reliable adaptive service-driven efficient routing (muRACER), a routing protocol suite based on a novel service-oriented design for sensor-actuator networks where nodes expose their capabilities to applications as a service profile. A node's service profile consists of a set of services (i.e., sensing and actuation capabilities) that it provides and the quality-of-service (QoS) parameters associated with those services (delay, accuracy, freshness, etc.). muRACER uses an efficient service-aware routing approach that aggressively reduces downstream traffic (from the sink to the network's nodes) by translating service profiles into efficient paths for queries. To support QoS, muRACER dynamically adapts each node's routing behavior and service profile according to the current context of that node, i.e., number of pending queries and number and type of messages to be routed. Finally, muRACER improves end-to-end reliability through a scalable reputation-based approach in which each node is able to locally estimate the next hop of the most reliable path to the sink. Service- and context-aware reliable routing enhances the network's efficiency and effectiveness (satisfaction of applications' QoS requirements). We implemented muRACER on top of TinyOS and conducted several experiments that confirmed muRACER's ability with regard to each of its design objectives.

We present DIA, a Web services-based infrastructure for the Discovery, Integration, and Analysis of geoscience data, tools, and services. DIA provides a collaborative environment where scientists can share their resources (e.g., geochemical data, filtering services, etc.) by registering them through well-defined ontologies. We have developed a planetary materials ontology in OWL for this purpose. The ontology is used by different geoscientists (using Web services) to explore, extract, and integrate information from different heterogeneous data sets. The DIA system is now in its final pre-release phase. It is currently made accessible to a few geoscientists for conducting usability analyses, and it will eventually be made available to the community at large through the geoscience portal (GEON) at the San Diego Supercomputer Center.

Most routing protocols for sensor-actuator networks (SANETs) are built under the assumption that nodes normally cooperate in forwarding each other's messages. In practice, this assumption is not realistic; SANETs are environments where nodes may or may not cooperate. For several reasons, a node may fail to operate as planned at deployment time. As a result, when actually deployed, protocols and applications may not be as efficient as expected. In this paper, we present TARP (Trust-Aware Routing Protocol), a routing protocol for sensor-actuator networks that exploits past nodes' routing behavior and links' quality to determine efficient paths. We implemented TARP in a TinyOS-based SANET and conducted several experiments to evaluate its performance. The obtained results confirmed that TARP achieves substantial improvements in terms of energy consumption and scalability.

Query processing in sensor networks entails three phases: (i) query routing, (ii) query evaluation, and (iii) aggregating results and routing them back to the query source. Most existing query routing approaches are "overreaching", i.e., they use broadcasting schemes where nodes use some default transmission power to broadcast queries to all of their neighbors. Moreover, current routing schemes generally do not exploit the semantics of queries. As a result, a query may be routed through several hops to end up at nodes that do not provide the sensing capabilities that it requires or that do not have paths to nodes that provide the required capabilities. This obviously incurs excessive communication and processing overhead that is not strictly required to deliver the query to its recipients. To address these deficiencies, we introduce the notion of service-driven query routing (SQR) which is a routing scheme based on the idea of modeling nodes' capabilities as services. In SQR, nodes maintain local information about their neighbors' services. They combine this information with the queries' semantics to determine efficient routing paths and minimal radio transmission power levels when forwarding queries to their neighbors

The emergence of Web databases has introduced new challenges related to their organization, access, integration, and interoperability. New approaches and techniques are needed to provide across-the-board transparency for accessing and manipulating Web databases irrespective of their data models, platforms, locations, or systems. In meeting these needs, it is necessary to build a middleware infrastructure to support flexible tools for information space organization, communication facilities, information discovery, content description, and assembly of data from heterogeneous sources. In this paper, we describe a scalable middleware for efficient data and application access that we have built using the available technologies. The resulting system is called WebFINDIT. It is a scalable and uniform infrastructure for locating and accessing heterogeneous and autonomous databases and applications.

In our project, we are adopting a new approach for handling video data. We view the video as a well-defined data type with its own description, parameters and applicable methods. The system is based on PREDATOR, an open-source object-relational DBMS. PREDATOR uses Shore as the underlying storage manager. Supporting video operations (storing, searching-by-content and streaming) and new query types (query-by-example and multi-feature similarity searching) requires major changes in many of the traditional system components. More specifically, the storage and buffer manager has to deal with huge volumes of data with real-time constraints. Query processing has to consider the video methods and operators in generating, optimizing and executing the query plans.