
Dr Francois Carrez
Publications
The requirements of analyzing heterogeneous data streams and detecting complex patterns in near real-time have raised the prospect of Complex Event Processing (CEP) for many internet of things (IoT) applications. Although CEP provides a scalable and distributed solution for analyzing complex data streams on the fly, it is designed for reactive applications as CEP acts on near real-time data and does not exploit historical data. In this regard, we propose a proactive architecture which exploits historical data using machine learning (ML) for prediction in conjunction with CEP. We propose an adaptive prediction algorithm called Adaptive Moving Window Regression (AMWR) for dynamic IoT data and evaluated it using a real-world use case with an accuracy of over 96%. It can perform accurate predictions in near real-time due to reduced complexity and can work along CEP in our architecture. We implemented our proposed architecture using open source components which are optimized for big data applications and validated it on a use-case from Intelligent Transportation Systems (ITS). Our proposed architecture is reliable and can be used across different fields in order to predict complex events.
As sensors are adopted in almost all fields of life, the Internet of Things (IoT) is triggering a massive influx of data. We need efficient and scalable methods to process this data to gain valuable insight and take timely action. Existing approaches which support both batch processing (suitable for analysis of large historical data sets) and event processing (suitable for realtime analysis) are complex. We propose the hut architecture, a simple but scalable architecture for ingesting and analyzing IoT data, which uses historical data analysis to provide context for real-time analysis. We implement our architecture using open source components optimized for big data applications and extend them where needed. We demonstrate our solution on two real-world smart
The gathering of real-world data is facilitated by many pervasive data sources such as sensor devices and smartphones. The abundance of the sensory data raises the need to make the data easily available and understandable for the potential users and applications. Using semantic enhancements is one approach to structure and organize the data and to make it processable and interoperable by machines. In particular, ontologies are used to represent information and their relations in machine interpretable forms. In this context, a significant amount of work has been done to create real-world data description ontologies and data description models; however, little effort has been done in creating and constructing meaningful topical ontologies from a vast amount of sensory data by automated processes. Topical ontologies represent the knowledge from a certain domain providing a basic understanding of the concepts that serve as building blocks for further processing. There is a lack of solution that construct the structure and relations of ontologies based on real-world data. To address this challenge, we introduce a knowledge acquisition method that processes real-world data to automatically create and evolve topical ontologies based on rules that are automatically extracted from external sources. We use an extended k-means clustering method and apply a statistic model to extract and link relevant concepts from the raw sensor data and represent them in the form of a topical ontology. We use a rule-based system to label the concepts and make them understandable for the human user or semantic analysis and reasoning tools and software. The evaluation of our work shows that the construction of a topological ontology from raw sensor data is achievable with only small construction errors.
IoT data analytics is underpinning numerous applications, however the task is still challenging predominantly due to heterogeneous IoT data streams, unreliable networks and ever increasing size of the data. In this context, we propose a two layer architecture for analyzing IoT data. The first layer provides a generic interface using a service oriented gateway to ingest data from multiple interfaces and IoT systems, store it in a scalable manner and analyze it in real-time to extract high-level events whereas second layer is responsible for probabilistic fusion of these high-level events. In the second layer, we extend state-ofthe- art event processing using Bayesian networks (BNs) in order to take uncertainty into account while detecting complex events. We implement our proposed solution using open source components optimized for large-scale applications. We demonstrate our solution on real-world use-case in the domain of intelligent transportation system (ITS) where we analysed traffic, weather and social media data streams from Madrid city in order to predict probability of congestion in real-time. The performance of the system is evaluated qualitatively using a web-interface where traffic administrators can provide the feedback about the quality of predictions and quantitatively using F-measure with an accuracy of over 80%.
The ability to manage the distributed functionality of large multi-vendor networks will be an important step towards ultra-dense 5G networks. Managing distributed scheduling functionality is particularly important, due to its influence over inter-cell interference and the lack of standardization for schedulers. In this paper, we formulate a method of managing distributed scheduling methods across a small cluster of cells by dynamically selecting schedulers to be implemented at each cell. We use deep reinforcement learning methods to identify suitable joint scheduling policies, based on the current state of the network observed from data already available in the RAN. Additionally, we also explore three methods of training the deep reinforcement learning based dynamic scheduler selection system. We compare the performance of these training methods in a simulated environment against each other, as well as homogeneous scheduler deployment scenarios, where each cell in the network uses the same type of scheduler. We show that, by using deep reinforcement learning, the dynamic scheduler selection system is able to identify scheduler distributions that increase the number of users that achieve their quality of service requirements in up to 77% of the simulated scenarios when compared to homogeneous scheduler deployment scenarios.
providing adaptive resource intensive Web services from mobile hosts needs to be done in a rather light-weight manner to allow continuous service provisioning. Processing and communication will drain the battery rapidly; hence both should be kept at a minimum. This paper describes the outcomes of an investigation into offloading and migration mechanisms that facilitate provision of adaptive and distributed mobile Web services. The investigation goes through three phases. The first phase integrates these mechanisms with the Simple Object Access Protocol (SOAP) and Representational State Transfer (REST) architectures producing extended mobile Web service frameworks. This phase is achieved by the implementation of a prototype that allows performance evaluation of both extended frameworks. The evaluation of the load and performance of the distributed services is taking place using resource intensive applications. The results presented show that basing distributed mobilehosted services on REST is more suitable than using SOAP as underlying Web service infrastructure. The second phase relies on the outperforming REST-based framework to examine four distinct strategies for mobile Web service distribution mechanisms. In the last phase, evaluation results of the second phase are interpreted as Fuzzy Logic rules. These rule sets are used to trigger and control offloading schemes. © 2012 ACADEMY PUBLISHER.
Purpose Mobile computing enables end-users to create small services on their mobiles and share valuable and context-aware information with others. The purpose of this paper is to introduce a platform for end-user generated mobile services – so-called microservices. Design/methodology/approach As a key component the authors present a microservice description language for user-driven mobile service creation and platform-independent service execution and rendering. The paper also gives insight into the authors' visual authoring tool. The chosen design approach is evaluated in two phases: an intermediate evaluation with a small hands-on trial and an online survey; and a final laboratory test with 24 test users in total. Findings The paper provides empirical insights about the methods and motivations of end-users creating small mobile services. The main purposes of service creation would be mostly to exchange information, stay in contact, and just for fun (on the basis of non-commercial use). The evaluations also indicate the visual drag and drop approach of putting service blocks together as being the most favored in terms of user satisfaction. Originality/value The concepts and findings introduced in this paper will help in designing mobile service authoring environments, which is appealing to software communities/vendors and mobile network operators. The presented platform is, to the authors' knowledge, the first designed and implemented infrastructure enabling end-user mobile service creation.
: The Internet-of-Things (IoT) is unanimously identified as one of the main pillars of future smart scenarios. However, despite the growing number of IoT deployments, the majority of IoT applications tend to be self-contained, thereby forming vertical silos. Indeed, the ability to combine and synthesize data streams and services from diverse IoT platforms and testbeds, holds the promise to increase the potential of smart applications in terms of size, scope and targeted business context. This paper describes the system architecture for the FIESTA-IoT platform, whose main aim is to federate a large number of testbeds across the planet, in order to offer experimenters the unique experience of dealing with a large number of semantically interoperable data sources. This system architecture was developed by following the Architectural Reference Model (ARM) methodology promoted by the IoT-A project (FP7 “light house” project on Architecture for the Internet of Things). Through this process, the FIESTAIoT architecture is composed of a set of Views that deals with a “logical” functional decomposition (Functional View, FV) and data structuring and annotation, data flows and inter-functional component interactions (Information View, IV).
Gateways in sensor networks are used to relay, aggregate and communicate information from capillary networks to more capable (e.g. IP-based) networks. However Gateway-to-Gateway (G2G) communication to exchange and update information among the gateways in large-scale sensor networks for query processing, data fusion and other similar tasks has been less discussed in recent works. The requirements for large-scale sensor networks such as dynamic topology and update strategies to reduce the overall network load makes G2G communications an important aspect in the network design. In this paper, we introduce a mediated gossip-based G2G communication mechanism. The proposed solution leverages the publish/subscribe approach and uses high-level context assigned to publish/subscribe channels to enable the information discovery and G2G communications. Gateways store/aggregate sensor observation and measurement data according to specific context which is defined based on features such as spatial and temporal attributes, observed phenomena (i.e. feature of interest) and sensor device features. The gateways communicate with each other to exchange data and also to forward related queries for data aggregation in cases that the data should be aggregated from two different sources. The proposed solution also facilitates reliable sensor service provisioning by enabling gateways to communicate and/or forward requests to other gateways when a resource fails or a sensor node becomes unavailable. We compare our results to probabilistic gossiping algorithms and run benchmarks on different dynamic network topologies based on indicators such as number of sent messages and dissemination delay.
The wide field of wireless sensor networks requires that hun- dreds or even thousands of sensor nodes have to be main- tained and configured. With the upcoming initatives such as Smart Home and Internet of Things, we need new mecha- nism to discover and manage this amount of sensors. In this paper, we describe a middleware architecture that uses con- text information of sensors to supply a plug-and-play gate- way and resource management framework for heterogeneous sensor networks. Our main goals are to minimise the effort for network engineers to configure and maintain the network and supply a unified interface to access the underlying het- erogeneous network. Based on the context information such as battery status, routing information, location and radio signal strength the gateway will configure and maintain the sensor network. The sensors are associated to nearby base stations using an approach that is adapted from the 802.11 WLAN association and negotiation mechanism to provide registration and connectivity services for the underlying sen- sor devices. This abstracted connection layer can be used to integrate the underlying sensor networks into high-level ser- vices and applications such as IP-based networks and Web services.
Collaboration towards a goal involves groups of entities collectively possessing characteristics required to accomplish the goal. Facilitating collaborations in pervasive environments requires the automated formation of such groups. The group formation process is especially challenging in decentralised environments where there is no single central entity that can coordinate the formation process. It is also important that the group formation mechanisms are generic in nature so that they can be utilised in heterogeneous target environments regardless of their domain and requirements. This paper proposes a generic approach for automating group formation in decentralised environments. © 2011 IEEE.
Most of the IoT applications are distributed in nature generating large data streams which have to be analyzed in near real-time. Solutions based on Complex Event Processing (CEP) have the potential to extract high-level knowledge from these data streams but the use of CEP for distributed IoT applications is still in early phase and involves many drawbacks. The manual setting of rules for CEP is one of the major drawback. These rules are based on threshold values and currently there are no automatic methods to find the optimized threshold values. In real-time dynamic IoT environments, the context of the application is always changing and the performance of current CEP solutions are not reliable for such scenarios. In this regard, we propose an automatic and context aware method based on clustering for finding optimized threshold values for CEP rules. We have developed a lightweight CEP called CEP to run on low processing hardware which can update the rules on the run. We have demonstrated our approach using a real-world use case of Intelligent Transportation System (ITS) to detect congestion in near real-time.
The widespread use of IoT devices has opened the possibilities for many innovative applications. Almost all of these applications involve analyzing complex data streams with low latency requirements. In this regard, pattern recognition methods based on CEP have the potential to provide solutions for analyzing and correlating these complex data streams in order to detect complex events. Most of these solutions are reactive in nature as CEP acts on real-time data and does not exploit historical data. In our work, we have explored a proactive approach by exploiting historical data using machine learning methods for prediction with CEP. We propose an adaptive prediction algorithm called Adaptive Moving Window Regression (AMWR) for dynamic IoT data and evaluated it using a realworld use case. Our proposed architecture is generic and can be used across different fields for predicting complex events.
Even mobile Web Services are still provided using servers that usually reside in the core networks. Main reason for not providing large and complex Web Services from resource limited mobile devices is not only the volatility of wireless connections and mobility of mobile hosts, but also, the often limited processing power. Offloading of some of the processing tasks is one step towards achieving optimal mobile Web Service provision. This paper presents two frameworks for providing distributed mobile Web Services: One mobile service provision framework is built on Simple Object Access Protocol (SOAP), while the other implements Representational State Transfer (REST) architecture. Both frameworks have been extended with offloading functionality and different types of resource intensive operations, i.e., process intensive and bandwidth intensive services, have been tested. The results show that using a REST-based framework leads of a better performing offloading behaviour, compared to SOAP-based mobile services. Distributed mobile services based on REST consume fewer resources and achieve better performance compared to SOAP based mobile services. The paper describes the approach, evaluation method and findings.
One of the goals that can be achieved by providing adaptive web services from mobile hosts is to allow continuous service provisioning. However, there are limitations in terms of complexity and size of the services that may be executed on mobile hosts. In this paper, two steps are taken towards providing adaptive web services from resource limited mobile devices. The first step is to investigate mechanisms that facilitate distributing the execution of mobile web services; the main mechanisms are offloading and migration. The second step is to integrate these mechanisms with available web service architectures to produce an extended mobile web service framework. In this case we integrated them with both SOAP as well as REST. The paper describes the offloading and migration mechanisms as well as the implementation of a prototype that allows performance evaluation of both extended frameworks. To investigate the load and performance of the distributed services, the prototype implements resource intensive applications. The results presented show that basing distributed mobile-hosted services on REST is more suitable than using SOAP as underlying web service infrastructure. © 2011 IEEE.
Providing adaptive web services from mobile hosts is a new approach in mobile web services to cope resource scarcity of mobile network environment. This approach is explored through investigating some mechanisms to allow continuous and reliable service provisioning. However, there is a clear limitation in terms complexity and size of the services that may be executed on mobile hosts. In this paper, Simple Partial Offloading mechanism is studied to facilitate mobile web service adaptation through distributing the execution of mobile web services and modeling the transfer of required location-based information. The distribution can be classified into Forward or Bounce offloading while the transfer modeling is based on either Frontend or Backend scheme. Hence, four distinct types of mobile web service frameworks have been implemented; each of these architectures represents a different strategy for achieving adaptive and distributed web services. The paper describes the four prototypes that allow performance evaluation using resource intensive applications. The results presented show that basing distributed mobile hosted services on Backend Bounce Offload strategy is more suitable for mobile network environment.
The information generated from the Internet of Things (IoT) potentially enables a better understanding of the physical world for humans and supports creation of ambient intelligence for a wide range of applications in different domains. A semantics-enabled service layer is a promising approach to facilitate seamless access and management of the information from the large, distributed and heterogeneous sources. This paper presents the efforts of the IoT.est project towards developing a framework for service creation and testing in an IoT environment. The architecture design extends the existing IoT reference architecture and enables a test-driven, semantics-based management of the entire service lifecycle. The validation of the architecture is shown though a dynamic test case generation and execution scenario.
We present an emerging approach of Mobile Social Spaces (MOSS) that intends to improve the ways in which people communicate in the modern world. Pervasive content and service creation and provisioning, in particular for dynamically changing social groups, is a complex task and subject to varying locations of individuals, of the complete group and its context. MOSS tries to remove some of the obstacles in this area and defines a range of functionalities that will support dynamic ubiquitous creation and instantiation of community content and services.