Automated machine learning (autoML) methods often require multiple passes over data and are computationally intensive, rendering them unsuitable for streaming scenarios where data is continuously generated and distributions evolve over time. The few existing autoML solutions for stream learning mainly rely on random search or genetic algorithms, which struggle to maintain high performance in dynamic environments. By contrast, leading methods in batch learning such as the Sequential Model-based Algorithm Configuration (SMAC) leverage modelbased approaches, suggesting opportunities for improvement in stream settings. To address these challenges and meet the requirements of stream scenarios, we introduce OnlineSMAC, a model-based optimizer for data streams. OnlineSMAC combines Bayesian optimization with an extension of the SMAC optimizer to dynamically select optimal processing pipelines and hyperparameters. Our results show that this approach is highly competitive, achieving performance on par with state-of-the-art stream autoML methods. This highlights the promising potential of using Bayesian optimization for data streams.

Internet of Things (IoT) applications consist of diverse resource-constrained/rich devices with a considerable portion being mobile. Such devices demand lightweight, loosely coupled interactions in terms of time, space, and synchronization. IoT protocols at the middleware layer support several interaction types (e.g., asynchronous messaging, streaming, etc.) ensuring successful interactions between devices that use the same protocol. Additionally, they introduce different Quality of Service (QoS) delivery modes for data exchange with respect to available device and network resources. On the other hand, interconnecting heterogeneous IoT devices requires mapping both their functional and QoS properties. This calls for advanced interoperability solutions integrated with QoS modeling and analysis techniques. This paper introduces an automated synthesis of QoS-aware mediating artifacts. Such mediators enable the interconnection between IoT devices employing heterogeneous middleware protocols. Additionally, representative QoS models are synthesized. Leveraging these models, system designers can evaluate the effectiveness of the interconnection in terms of end-to-end QoS. We evaluate the usefulness of our approach through experimentation with a case study employing heterogeneous middleware protocols. In particular, we statistically analyze through simulations the effect of varying system parameters on the end-to-end QoS.

Automated Machine Learning (autoML) is a novel topic that aims to tackle the parameter configuration issue using automatic monitoring models and comprises different machine learning tasks, such as feature selection, model selection, and hyper-parameter tuning. It makes easier use of algorithms for non-ML experts as well as ML experts by automating tasks that rely on expert domain knowledge. Nevertheless, autoML is in its infancy stage and not well explored yet in the offline and stream settings. In this paper, we propose automated Classification (auto-Class) method for automated algorithm selection and configuration for data stream classification. AutoClass consists of training an ensemble of different tuned configurations and selecting the best-performing configuration to do the prediction. We present experiments performed on a diverse set of real and artificial datasets and show how our proposed approach can outperform the performance of competitive state-of-the-art ensemble and single-based methods.

Data stream processing and analytics (DSPA) engines are used to extract in (near) real-time valuable information from multiple IoT data streams. Deploying DSPA applications at the IoT network edge through Edge/Fog architectures is currently one of the core challenges for reducing both network delays and network bandwidth usage to reach the Cloud. In this paper, we address the problem of scheduling continuous DSPA operators to Fog-Cloud nodes featuring both computational and network resources. We are paying particular attention to the dynamic workload of these nodes due to variability of IoT data stream rates and the sharing of nodes' resources by multiple DSPA applications. In this respect, we propose TSOO, a resource-aware and time-efficient heuristic algorithm that takes into account the limited Fog computational resources, the real-time response constraints of DSPA applications, as well as, congestion and delay issues on Fog-to-Cloud network resources. Via extensive simulation experiments, we show that TSOO approximates an optimal operators' placement with a low execution cost.

Internet of Things (IoT) aim to leverage data from multiple sensors, actuators and devices for improving peoples' daily life and safety. Multiple data sources must be integrated, analyzed from the corresponding application and notify interested stakeholders. To support the data exchange between data sources and stakeholders, the publish/subscribe (pub/sub) middleware is often employed. Pub/sub provides additional mechanisms such as reliable messaging, event dropping, prioritization, etc. The event dropping mechanism is often used to satisfy Quality of Service (Q0S) requirements and ensure system stability. To enable event dropping, basic approaches apply finite buffers or data validity periods and more sophisticated ones are informationaware. In this paper, we introduce a pub/sub mechanism for probabilistic event dropping by considering the stakeholders' intermittent connectivity and QoS requirements. We model the pub/sub middleware as a network of queues which includes a novel ON/OFF queueing model that enables the definition of join probabilities. We validate our analytical model via simulation and compare our mechanism with existing ones. Experimental results can be used as insights for developing hybrid dropping mechanisms.

Civic engagement refers to any collective action towards the identification and solving of public issues. Current civic technologies are traditional Web-or mobile-based platforms that make difficult, or just impossible, the participation of citizens via different communication technologies. Moreover, connected objects sensing physical-world data can nourish participatory processes by providing physical evidence to citizens; however, leveraging these data is not direct and still a time-consuming process for civic technologies developers. This paper introduces the concept of social middleware for civic engagement. Social middleware allows citizens to engage in participatory processes - supported by civic technologies-via their favorite communication tools, and to interact not only with other citizens but also with relevant connected objects and software platforms. The mission of social middleware goes beyond the connection of all these heterogeneous entities. It aims at easing the implementation of distributed applications oriented toward civic engagement by featuring dedicated built-in services.

Understanding the sensing context of raw data is crucial for assessing the quality of large crowdsourced spatio-temporal datasets. Detecting sensing contexts in the wild is a challenging task and requires features from smartphone sensors that are not always available. In this paper, we propose three heuristic algorithms for detecting sensing contexts such as in/out-pocket, under/over-ground, and in/out-door for crowdsourced datasets that are destined for human mobility mining. These are unsupervised binary classifiers with a small memory footprint and execution time. Using a segment of the Ambiciti real dataset - a feature-limited crowdsourced dataset - we report that our algorithms perform equally well in terms of balanced accuracy (within 4.3%) when compared to machine learning (ML) models reported by an AutoML tool.

Online Social Network Services (OSNSs) are changing the fabric of our society, impacting almost every aspect of it. Over the past few decades, an aggressive market rivalry has led to the emergence of multiple competing, “closed” OSNSs. As a result, users are trapped in the walled gardens of their OSNS, encountering restrictions about what they can do with their personal data, the people they can interact with, and the information they get access to. As an alternative to the platform lock-in, “open” OSNSs promote the adoption of open, standardized APIs. However, users still massively adopt closed OSNSs to benefit from the services’ advanced functionalities and/or follow their “friends,” although the users’ virtual social sphere is ultimately limited by the OSNSs they join. Our work aims at overcoming such a limitation by enabling users to meet and interact beyond the boundary of their OSNSs, including reaching out to “friends” of distinct closed OSNSs. We specifically introduceUniversal Social Network Bus (USNB), which revisits the “service bus” paradigm that enables interoperability across computing systems to address the requirements of “social interoperability.” USNB featuressynthetic profilesandpersonaefor interaction across the boundaries of closed and open and profile- and non-profile-based OSNSs through areference social interaction service. We ran a 1-day workshop with a panel of users who experimented with the USNB prototype to assess the potential benefits of social interoperability for social network users. Results show the positive evaluation of users for USNB, especially as an enabler of applications for civic participation. This further opens up new perspectives for future work, among which includes enforcing security and privacy guarantees.

Mobile IoT applications are typically deployed on resource-constrained devices with intermittent network connectivity. To support the deployment of such applications, the Publish/Subscribe (pub/sub) interaction paradigm is often employed, as it decouples mobile peers in time and space. Pub/sub middleware protocols and APIs consider the Things' hardware limitations and support the development of effective applications by providing QoS features. These features aim to enable developers to tune an application by switching different levels of response times and success rates. However, the profusion of pub/sub protocols coupled with intermittent connectivity result in non-trivial application tuning. In this paper, we model the performance of middleware protocols found in IoT, which are classified within the pub/sub interaction paradigm. We model reliable and unreliable protocols, by considering QoS semantics for data validity, buffer capacities as well as the intermittent availability of peers. Finally, we perform statistical analysis by varying these QoS semantics, demonstrating their significant effect on the rate of successful interactions. We showcase the application of our analysis in concrete scenarios relating to Traffic Information Management systems, that integrate both reliable and unreliable participants. The consequent PerfMP performance modeling pattern may be tailored for a variety of deployments, in order to control fine-grained QoS policies.

Internet of Things (IoT) applications consist of diverse Things (sensors and devices) in terms of hardware resources. Furthermore, such applications are characterized by the Things' mobility and multiple interaction types, such as synchronous, asynchronous, and streaming. Middleware IoT protocols consider the above limitations and support the development of effective applications by providing several Quality of Service (QoS) features. These features aim to enable application developers to tune an application by switching different levels of response times and delivery success rates. However, the profusion of the developed IoT protocols and the intermittent connectivity of mobile Things, result to a non-trivial application tuning. In this paper, we model the performance of the middleware overlay infrastructure using Queueing Network Models (QNMs). To represent the mobile Thing's connections/disconnections, we model and solve analytically an ON/OFF queueing center. We apply our approach to Streaming interactions with mobile peers. Finally, we validate our model using simulations. The deviations between the performance results foreseen by the analytical model and the ones provided by the simulator are shown to be less than 5%.

Systems deployed in mobile environments are typically characterized by intermittent connectivity and asynchronous sending/reception of data. To create effective mobile systems for such environments, it is essential to guarantee acceptable levels of timeliness between sending and receiving mobile users. In order to provide QoS guarantees in different application scenarios and contexts, it is necessary to model the system performance by incorporating the intermittent connectivity. Queueing Network Models (QNMs) offer a simple modeling environment, which can be used to represent various application scenarios, and provide accurate analytical solutions for performance metrics, such as system response time. In this paper, we provide an analytical solution regarding the end-to-end response time between users sending and receiving data by modeling the intermittent connectivity of mobile users with QNMs. We utilize the publish/subscribe (pub/sub) middleware as the underlying communication infrastructure for mobile users. To represent the user's connections/disconnections, we model and solve analytically an ON/OFF queueing system by applying a mean value approach. Finally, we validate our model using simulations with real-world workload traces. The deviations between the performance results foreseen by the analytical model and the ones provided by the simulator are shown to be less than 5% for a variety of scenarios.

Service composition is a widely used method in ubiquitous computing that enables accomplishing complex tasks required by users based on elementary (hardware and software) services available in ubiquitous environments. To ensure that users experience the best Quality of Service (QoS) with respect to their quality needs, service composition has to be QoS-aware. Establishing QoS-aware service compositions entails efficient service selection taking into account the QoS requirements of users. A challenging issue towards this purpose is to consider service selection under global QoS requirements (i.e., Requirements imposed by the user on the whole task), which is of high computational cost. This challenge is even more relevant when we consider the dynamics, limited computational resources and timeliness constraints of ubiquitous environments. To cope with the above challenge, in this paper we present QASSA, an efficient service selection algorithm that provides the appropriate ground for QoS-aware service composition in ubiquitous environments. The contribution of QASSA is three-fold. First, it formulates service selection under global QoS requirements as a set-based optimisation problem, benefiting from recent proposals in the domain of multi-objective optimisation. Second, QASSA resolves this problem in an efficient way using clustering techniques, namely the K-Means algorithm. Third, QASSA is devised in two versions, viz., centralised and distributed versions, so that it can be executed on top of centralised and decentralised infrastructures in ubiquitous environments. Results of experimental studies are presented to illustrate the timeliness and optimality of QASSA.

Large-scale mobile environments are characterized by, among others, a large number of mobile users, intermittent connectivity and non-homogeneous arrival rate of data to the users, depending on the region's context. Multiple application scenarios in major cities need to address the above situation for the creation of robust mobile systems. Towards this, it is fundamental to enable system designers to tune a communication infrastructure using various parameters depending on the specific context. In this paper, we take a first step towards enabling an application platform for large-scale information management relying on `mobile social crowd-sourcing'. To inform the stakeholders of expected loads and costs, we model a large-scale mobile pub/sub system as a queueing network. We introduce additional timing constraints such as (i) mobile user's intermittent connectivity period; and (ii) data validity lifetime period (e.g. that of sensor data). Using our MobileJINQS simulator, we parameterize our model with realistic input loads derived from the D4D dataset (CDR) and varied lifetime periods in order to analyze the effect on response time. This work provides system designers with coarse grain design time information when setting realistic loads and time constraints.

Quality of Service (QoS) in orchestrated web services compositions have been well studied with probabilistic and multi-dimensional models. Choreographies that involve message passing among services, on the other hand, require further analysis. In this paper, we begin with the set of QoS domains that may be studied in case of choreographies and the algebraic rules for their composition. As choreographies manage QoS composition in a distributed fashion, techniques to enrich functional specifications with QoS are examined using the model proposed in the CHOReOS project. These are further analyzed with choreographies that may reconfigure due to functional or QoS requirements. Studies on the effects of such reconfiguration on multiple QoS domains can lead to better understanding of optimal runtime configurations along with associated tradeoffs. A goal programming approach is also proposed to choose Pareto optimal solutions with respect to diverse QoS domains.

Due to the large success of wireless networks and handheld devices, the ambient intelligence (AmI) paradigm is becoming a reality. One of the most challenging objectives to achieve in AmI environments is to enable a user to perform a task by composing on the fly networked services available at a specific time and place. Towards this goal, we propose a solution based on semantic Web services, and we show how service capabilities described as conversations can be integrated to perform a user task that is also described as a conversation, further meeting the QoS requirements of the user task. Experimental results show that the runtime overhead of our algorithm is reasonable, and further, that QoS-awareness improves its performance.

The Amigo project aims to develop a networked home system enabling the ambient intelligence / pervasive computing vision by effectively integrating devices and their hosted services in today’s home. The Amigo system architecture poses limited technology-specific restrictions, supporting interoperability among heterogeneous services.