RonPub -- Research Online Publishing
Not a member yet
199 research outputs found
Sort by
Combining Process Guidance and Industrial Feedback for Successfully Deploying Big Data Projects
Full text linkCompanies are faced with the challenge of handling increasing amounts of digital data to run or improve their business. Although a large set of technical solutions are available to manage such Big Data, many companies lack the maturity to manage that kind of projects, which results in a high failure rate. This paper aims at providing better process guidance for a successful deployment of Big Data projects. Our approach is based on the combination of a set of methodological bricks documented in the literature from early data mining projects to nowadays. It is complemented by learned lessons from pilots conducted in different areas (IT, health, space, food industry) with a focus on two pilots giving a concrete vision of how to drive the implementation with emphasis on the identification of values, the definition of a relevant strategy, the use of an Agile follow-up and a progressive rise in maturity
Differentially Private Linear Models for Gossip Learning through Data Perturbation
Full text linkPrivacy is a key concern in many distributed systems that are rich in personal data such as networks of smart meters or smartphones. Decentralizing the processing of personal data in such systems is a promising first step towards achieving privacy through avoiding the collection of data altogether. However, decentralization in itself is not enough: Additional guarantees such as differential privacy are highly desirable. Here, we focus on stochastic gradient descent (SGD), a popular approach to implement distributed learning. Our goal is to design differentially private variants of SGD to be applied in gossip learning, a decentralized learning framework. Known approaches that are suitable for our scenario focus on protecting the gradient that is being computed in each iteration of SGD. This has the drawback that each data point can be accessed only a small number of times. We propose a solution in which we effectively publish the entire database in a differentially private way so that linear learners could be run that are allowed to access any (perturbed) data point any number of times. This flexibility is very useful when using the method in combination with distributed learning environments. We show empirically that the performance of the obtained model is comparable to that of previous gradient-based approaches and it is even superior in certain scenarios
Technology Selection for Big Data and Analytical Applications
Full text linkThe term Big Data has become pervasive in recent years, as smart phones, televisions, washing machines, refrigerators, smart meters, diverse sensors, eyeglasses, and even clothes connect to the Internet. However, their generated data is essentially worthless without appropriate data analytics that utilizes information retrieval, statistics, as well as various other techniques. As Big Data is commonly too big for a single person or institution to investigate, appropriate tools are being used that go way beyond a traditional data warehouse and that have been developed in recent years. Unfortunately, there is no single solution but a large variety of different tools, each of which with distinct functionalities, properties and characteristics. Especially small and medium-sized companies have a hard time to keep track, as this requires time, skills, money, and specific knowledge that, in combination, result in high entrance barriers for Big Data utilization. This paper aims to reduce these barriers by explaining and structuring different classes of technologies and the basic criteria for proper technology selection. It proposes a framework that guides especially small and mid-sized companies through a suitable selection process that can serve as a basis for further advances
Semantic Blockchain to Improve Scalability in the Internet of Things
Full text linkGenerally scarce computational and memory resource availability is a well known problem for the IoT, whose intrinsic volatility makes complex applications unfeasible. Noteworthy efforts in overcoming unpredictability (particularly in case of large dimensions) are the ones integrating Knowledge Representation technologies to build the so-called Semantic Web of Things (SWoT). In spite of allowed advanced discovery features, transactions in the SWoT still suffer from not viable trust management strategies. Given its intrinsic characteristics, blockchain technology appears as interesting from this perspective: a semantic resource/service discovery layer built upon a basic blockchain infrastructure gains a consensus validation. This paper proposes a novel Service-Oriented Architecture (SOA) based on a semantic blockchain for registration, discovery, selection and payment. Such operations are implemented as smart contracts, allowing distributed execution and trust. Reported experiments early assess the sustainability of the proposal
Controlled Components for Internet of Things As-A-Service
Full text linkIn order to facilitate developers willing to create future Internet of Things (IoT) services incorporating the nonfunctional aspects, we introduce an approach and an environment based on controlled components. Our approach allows developers to design an IoT "as-a-service", to build the service composition and to manage it. This is important, because the IoT allows us to observe and understand the real world in order to have decision-making information to act on reality. It is important to make sure that all these components work according to their mission, i.e. their Quality of Service (QoS) contract. Our environment provides the modeling, generates Architecture Description Language (ADL) formats, and uses them in the implementation phase on an open-source platform
OnGIS: Semantic Query Broker for Heterogeneous Geospatial Data Sources
Full text linkQuerying geospatial data from multiple heterogeneous sources backed by different management technologies poses an interesting problem in the data integration and in the subsequent result interpretation. This paper proposes broker techniques for answering a user's complex spatial query: finding relevant data sources (from a catalogue of data sources) capable of answering the query, eventually splitting the query and finding relevant data sources for the query parts, when no single source suffices. For the purpose, we describe each source with a set of prototypical queries that are algorithmically arranged into a lattice, which makes searching efficient. The proposed algorithms leverage GeoSPARQL query containment enhanced with OWL 2 QL semantics. A prototype is implemented in a system called OnGIS
Definition and Categorization of Dew Computing
Full text linkDew computing is an emerging new research area and has great potentials in applications. In this paper, we propose a revised definition of dew computing. The new definition is: Dew computing is an on-premises computer software-hardware organization paradigm in the cloud computing environment where the on-premises computer provides functionality that is independent of cloud services and is also collaborative with cloud services. The goal of dew computing is to fully realize the potentials of on-premises computers and cloud services. This definition emphasizes two key features of dew computing: independence and collaboration. Furthermore, we propose a group of dew computing categories. These categories may inspire new applications
Doing More with the Dew: A New Approach to Cloud-Dew Architecture
Full text linkWhile the popularity of cloud computing is exploding, a new network computing paradigm is just beginning. In this paper, we examine this exciting area of research known as dew computing and propose a new design of cloud-dew architecture. Instead of hosting only one dew server on a user's PC - as adopted in the current dewsite application - our design promotes the hosting of multiple dew servers instead, one for each installed domain. Our design intends to improve upon existing cloud-dew architecture by providing significantly increased freedom in dewsite development, while also automating the chore of managing dewsite content based on the user's interests and browsing habits. Other noteworthy benefits, all at no added cost to dewsite users, are briefly explored as well
Hierarchical Multi-Label Classification Using Web Reasoning for Large Datasets
Full text linkExtracting valuable data among large volumes of data is one of the main challenges in Big Data. In this paper, a Hierarchical Multi-Label Classification process called Semantic HMC is presented. This process aims to extract valuable data from very large data sources, by automatically learning a label hierarchy and classifying data items.The Semantic HMC process is composed of five scalable steps, namely Indexation, Vectorization, Hierarchization, Resolution and Realization. The first three steps construct automatically a label hierarchy from statistical analysis of data. This paper focuses on the last two steps which perform item classification according to the label hierarchy. The process is implemented as a scalable and distributed application, and deployed on a Big Data platform. A quality evaluation is described, which compares the approach with multi-label classification algorithms from the state of the art dedicated to the same goal. The Semantic HMC approach outperforms state of the art approaches in some areas
Runtime Adaptive Hybrid Query Engine based on FPGAs
Full text linkThis paper presents the fully integrated hardware-accelerated query engine for large-scale datasets in the context of Semantic Web databases. As queries are typically unknown at design time, a static approach is not feasible and not flexible to cover a wide range of queries at system runtime. Therefore, we introduce a runtime reconfigurable accelerator based on a Field Programmable Gate Array (FPGA), which transparently incorporates with the freely available Semantic Web database LUPOSDATE. At system runtime, the proposed approach dynamically generates an optimized hardware accelerator in terms of an FPGA configuration for each individual query and transparently retrieves the query result to be displayed to the user. During hardware-accelerated execution the host supplies triple data to the FPGA and retrieves the results from the FPGA via PCIe interface. The benefits and limitations are evaluated on large-scale synthetic datasets with up to 260 million triples as well as the widely known Billion Triples Challenge