In this paper, we present a new method for learned data cleaning. In contrast to existing methods, our method learns to clean data in the latent space. The main idea is that we (1) shape the latent space such that we know the area where clean data resides and (2) learn latent operators trained on error repair (Lopster) which shift erroneous data (e.g., table rows with noise, outliers, or missing values) in their latent representation back to a "clean" region, thus abstracting the complexities of the input domain. When formulating data cleaning as a simple shift operation in latent space, we can repair all types of errors using the same method which makes it more robust than other methods. Importantly, with our method, we can handle errors that are unseen during the training of our error repair model. We do not rely on an external error detection method as seen in the state-of-the-art, instead, we handle both detection and repair within the Lopster framework. In our evaluation, we show that our approach outperforms existing cleaning methods even when trained on only a subset of the errors that occur in the dirty data.

Recently, named data networking (NDN) has been introduced to connect the world of computing devices via naming data instead of their containers. Through this strategic change, NDN brings several new features to network communication, including in-network caching, multipath forwarding, built-in multicast, and data security. Despite these unique features of NDN networking, there exist plenty of opportunities for continuing developments, especially with packet forwarding and caching. In this context, we introduce liteNDN, a novel forwarding and caching strategy for NDN networks. liteNDN comprises a cooperative forwarding strategy through which NDN routers share their knowledge, i.e. data names and interfaces, to optimize their packet forwarding decisions. Subsequently, liteNDN leverages that knowledge to estimate the probability of each downstream path to swiftly retrieve the requested data. Additionally, liteNDN exploits heuristics, such as routing costs and data significance, to make proper decisions about caching normal as well as segmented packets. The proposed approach has been extensively evaluated in terms of the data retrieval latency, network utilization, and the cache hit rate. The results showed that liteNDN, compared to conventional NDN forwarding and caching strategies, achieves much less latency while reducing the unnecessary traffic and caching activities.

Nowadays, autonomous driving and driver assistance applications are being developed at an accelerated pace. This rapid growth is primarily driven by the potential of such smart applications to significantly improve safety on public roads and offer new possibilities for modern transportation concepts. Such indispensable applications typically require wireless connectivity between the vehicles and their surroundings, i.e. roadside infrastructure and cloud services. Nevertheless, such connectivity to external networks exposes the internal systems of individual vehicles to threats from remotely-launched attacks. In this realm, it is highly crucial to identify any misbehavior of the software components which might occur owing to either these threats or even software/hardware malfunctioning. In this paper, we introduce AutoSec, a host-based anomaly detection algorithm which relies on observing four timing parameters of the executed software components to accurately detect malicious behavior on the operating system level. To this end, AutoSec formulates the task of detecting anomalistic executions as a clustering problem. Specifically, AutoSec devises a hybrid clustering algorithm for grouping a set of collected timing traces resulted from executing the legitimate code. During the runtime, AutoSec simply classifies a certain execution as an anomaly, if its timing parameters are distant enough from the boundaries of the predefined clusters. To show the effectiveness of AutoSec, we collected timing traces from a testbed composed of a set of real and virtual control units communicating over a CAN bus. We show that using our proposed AutoSec, compared to baseline methods, we can identify up to 21% less false positives and 18% less false negatives.

Recently, indoor modeling has gained increased attention, thanks to the immense need for realizing efficient indoor location-based services. Indoor environments differ from outdoor spaces in two aspects: spaces are smaller and there are many structural objects such as walls, doors, and furniture. To model the indoor environments in a proper manner, novel data acquisition concepts and data modeling algorithms have been devised to meet the requirements of indoor spatial applications. In this realm, several research efforts have been exerted. Nevertheless, these efforts mostly suffer either from adopting impractical data acquisition methods or from being limited to 2D modeling. To overcome these limitations, we introduce the MapSense approach, which automatically derives indoor models from 3D point clouds collected by individuals using mobile devices, such as Google Tango, Apple ARKit, and Microsoft HoloLens. To this end, MapSense leverages several computer vision and machine learning algorithms for precisely inferring the structural objects. In MapSense, we mainly focus on improving the modeling accuracy through adopting formal grammars that encode design-time knowledge, i.e., structural information about the building. In addition to modeling accuracy, MapSense considers the energy overhead on the mobile devices via developing a probabilistic quality model through which the mobile devices solely upload high-quality point clouds to the crowd-sensing servers. To demonstrate the performance of MapSense, we implemented a crowd-sensing Android App to collect 3D point clouds from two different buildings by six volunteers. The results showed that MapSense can accurately infer the various structural objects while drastically reducing the energy overhead on the mobile devices.

Internet of Things (IoT) is expanding at a rapid rate where it allows for virtually endless opportunities and connections to take place. In general, IoT opens the door to a myriad of applications but also to many challenges. One of the major challenges is how to efficiently retrieve the sensory data from "resources-limited" IoT devices. Such devices typically have a restricted energy budget, which broadly hinders their direct connection to the Internet. In this realm, modern mobile devices, e.g. smartphones, tablets, smartwatches, have been harnessed to bridge between the low-power IoT devices and the Internet. However, the current vision which mainly relies on designing siloed gateways, i.e. a separate gateway/App for each IoT device, is certainly impractical, especially with the rapid growth in the number of IoT devices. Furthermore, the energy efficiency of the smart mobile devices hosting the IoT gateways has to be thoroughly considered. To tackle these challenges, we introduce GaaS (Gateway as a Service), a cross-platform gateway architecture for opportunistically retrieving sensory data from the low-power IoT sensors. Through Bluetooth low energy radios, GaaS is capable of simultaneously connecting to several nearby IoT sensors. To this end, we devise two distinct priority-based scheduling algorithms, namely the EP-WSM and FEP-AHP schedulers, which rank the detected IoT sensors, before estimating the connection time for each IoT sensor. The intuition behind ranking the IoT sensors is to improve the data retrieval rate from these sensors together with reducing the energy overhead on the mobile devices. Additionally, GaaS encompasses a self-adaptive engine to automatically balance the trade-off between energy efficiency and data retrieval rate through switching between schedulers according to the runtime dynamics. To demonstrate the effectiveness of GaaS, we implemented an IoT testbed to evaluate the energy consumption, the latency, and the data retrieval rate. The results show that using GaaS, compared to siloed gateways, we can identify up to 18% savings in the consumed energy while requiring much less data retrieval time.

Recently, mobile crowdsensing has become an appealing paradigm thanks to the ubiquitous presence of powerful mobile devices. Indoor mapping, as an example of crowdsensingdriven applications, is essential to provide many indoor locationbased services, such as emergency response, security, and tracking/navigation in large buildings. In this realm, 3D point clouds stand as an optimal data type which can be crowdsensed-using currently-available mobile devices, e.g. Google Tango, Microsoft Hololens and Apple ARKit-to generate floor plans with different levels of detail, i.e. 2D and 3D mapping. However, collecting such bulky data from "resources-limited" mobile devices can significantly harm their energy efficiency. To overcome this challenge, we introduce GreenMap, an energy-aware architectural framework for automatically mapping the interior spaces using crowdsensed point clouds with the support of structural information encoded in formal grammars. GreenMap reduces the energy overhead through projecting the point clouds to several filtration steps on the mobile devices. In this context, GreenMap leverages the potential of approximate computing to reduce the computational cost of data filtering while maintaining a satisfactory level of modeding accuracy. To this end, we propose two approximation strategies, namely DyPR and SuFFUSION. To demonstrate the effectiveness of GreenMap, we implemented a crowdsensing Android App to collect 3D point clouds from two different buildings. We show that GreenMap achieves significant energy savings of up to 67.8%, compared to the baseline methods, while generating comparable floor plans.

Recently, crowd-sensing has rapidly been evolved thanks to the technological advancement in personal mobile devices. This emerging technology opens the door for numerous applications to collect sensory data from the crowd. To provide people with a motive for participating in data acquisition, the crowd-sensing systems have to sidestep burdening the resources allocated to the mobile devices, i.e. computing power and energy budget. In this paper, we propose iSense, a novel framework for reducing the energy costs of participating in crowd-sensing. We mainly target the superfluous energy overhead on the mobile devices to sense and report their position information to the back-end servers. To relieve such an overhead, iSense entirely offloads the localization burden to the crowd-sensing servers. In this manner, iSense enables the utilization of advanced localization approaches thanks to the high resources of the crowd-sensing servers. To this end, iSense opportunistically exploits the “already-existent” network signaling exchanged frequently between the mobile devices and the WiFi networks or the cellular networks. To collect the localization data, we implement a lightweight data collection algorithm on a set of off-the-shelves access points. As a case study, we implement a two-step localization method, including a coarse- and a fine-grained localization. In this regard, compressed sensing is employed to estimate the fine-grained solution. To assess the effectiveness of iSense, we implemented a testbed to evaluate the energy consumption and the localization accuracy with different mobility and usage patterns. The results show that using iSense, compared to some baseline methods, we can identify up to 95% savings in the consumed energy.

Recently, several approaches have been proposed to automatically model indoor environments. Most of such efforts principally rely on the crowd to sense data such as motion traces, images, and WiFi footprints. However, large datasets are usually required to derive precise indoor models which can negatively affect the energy efficiency of the mobile devices participating in the crowd-sensing system. Furthermore, the aforementioned data types are hardly suitable for deriving 3D indoor models. To overcome these challenges, we propose GraMap, a QoS-aware automatic indoor modeling approach through crowd-sensing 3D point clouds. GraMap exploits a recently-developed sensors fusion mechanism, namely Tango technology, to cooperatively collect point clouds from the crowd. Afterward, a set of backend servers extracts the required geometrical information to derive indoor models.

This paper aims at developing a novel compression technique which breaks the "downward spiral" between compression ratio and recovery precision. Based on the previously developed Fuzzy Transform Compression (FTC), we design and implement a modified version of the algorithm, referred to as Fuzzy Compression Adaptive Transform (FuzzyCAT). The crux of FuzzyCAT is to adapt the transform parameters to the signal's curvature inferred from the time derivativesA comparative study with the Lightweight Temporal Compression (LTC) technique revealed that transmission costs of the FuzzyCAT nodes are much less than that of the LTC, which makes it an outstanding candidate for data compression in Wireless Sensor Networks.

Energy efficiency is an important goal for Wireless Sensor Network (WSN) designers. However, successful implementations of such networks are highly dependent on the enabling technologies, as well as on the provisioning of Quality of Service (QoS) in the network. In this paper, we propose a novel strategy, referred to as the lifetime planning for achieving best-effort QoS. Simultaneously, an adequate lifetime required to complete the assigned task is reached. The core idea is to sidestep lifetime maximization strategies in which sensor nodes continue functioning even after the fulfillment of the required task. In these cases, we could deliberately bound the operational lifetime to the expected task lifetime. As a result, more energy can be spent throughout the entire task lifetime for enhancing the provided service qualities. An analytical QoS model is engineered to validate the QoS's "conflicts-free" nature of lifetime planning. The proposed strategy is feasible via the design of QoS boundaries at design-time. During run-time, the controllable parameters are modulated by a proactive adaptation mechanism. To demonstrate the effectiveness of our design, we conduct an intensive performance evaluation using an office monitoring scenario in a cluster-tree WSN topology. The scenario has been designed in the Contiki network simulator Cooja using Tmote sky motes. Furthermore, we examine the profit of adopting our strategy relative to fixed heuristics and blind adaptation.