Based on the collective input of Dagstuhl Seminar (21342), this paper presents a comprehensive discussion on AI methods and capabilities in the context of edge computing, referred as Edge AI. In a nutshell, we envision Edge AI to provide adaptation for data-driven applications, enhance network and radio access, and allow the creation, optimisation, and deployment of distributed AI/ML pipelines with given quality of experience, trust, security and privacy targets. The Edge AI community investigates novel ML methods for the edge computing environment, spanning multiple sub-fields of computer science, engineering and ICT. The goal is to share an envisioned roadmap that can bring together key actors and enablers to further advance the domain of Edge AI. @en

Widespread adoption of mobile augmented reality (AR) and virtual reality (VR) applications depends on their smoothness and immersiveness. Modern AR applications applying computationally intensive computer vision algorithms can burden today's mobile devices, and cause high energy consumption and/or poor performance. To tackle this challenge, it is possible to offload part of the computation to nearby devices at the edge. However, this calls for smart task placement strategies in order to efficiently use the resources of the edge infrastructure. In this paper, we introduce Nimbus --- a task placement and offloading solution for a multi-tier, edge-cloud infrastructure where deep learning tasks are extracted from the AR application pipeline and offloaded to nearby GPU-powered edge devices. Our aim is to minimize the latency experienced by end-users and the energy costs on mobile devices. Our multifaceted evaluation, based on benchmarked performance of AR tasks, shows the efficacy of our solution. Overall, Nimbus reduces the task latency by ~4x and the energy consumption by ~77% for real-time object detection in AR applications. We also benchmark three variants of our offloading algorithm, disclosing the trade-off of centralized versus distributed execution.@en

Human mobility shapes our daily lives, our urban environment and even the trajectory of a global pandemic. While various aspects of human mobility and inter-personal contact duration have already been studied separately, little is known about how these two key aspects of our daily lives are fundamentally connected. Better understanding of such interconnected human behaviors is crucial for studying infectious diseases, as well as opportunistic content forwarding. To address these deficiencies, we conducted a study on a mobile social network of human mobility and contact duration, using data from 71 persons based on GPS and Bluetooth logs for 2 months in 2018. We augment these data with location APIs, enabling a finer granular characterization of the users’ mobility in addition to contact patterns. We model stops durations to reveal how time-unbounded-stops (e.g., bars or restaurants) follow a log-normal distribution while time-bounded-stops (e.g., offices, hotels) follow a power-law distribution. Furthermore, our analysis reveals contact duration adheres to a log-normal distribution, which we use to model the duration of contacts as a function of the duration of stays. We further extend our understanding of contact duration during trips by modeling these times as a Weibull distribution whose parameters are a function of trip length. These results could better inform models for information or epidemic spreading, helping guide the future design of network protocols as well as policy decisions.@en

For road safety, detecting and reacting efficiently to road hazards is crucial and yet challenging due to practical restrictions such as limited data availability, which relies on network support. Moreover, from a system perspective we lack a computational model capable of providing to vehicles reliable and real-time assessment of the road context. As autonomous vehicles become widespread, the safety issues are further aggravated by the gap between cloud, roadside infrastructure and road users in terms of communication latency, software-hardware compatibility and data interoperability. To tackle this, we present ECCO: an orchestration framework that enables edge-cloud collaborative computing for road context assessment. ECCO can create on-demand task execution pipelines spanning multiple, potentially resource-constrained edge-nodes with the smart IoT infrastructure support. Our prototype lays the groundwork to support new services, which can use more efficiently the road infrastructure and deliver safety-critical applications for road users.@en

For indoor IoT environments, spontaneous device associations are of particular interest where users establish a connection in an ad-hoc manner to enable serendipitous interaction. For instance, between a user's personal device and devices the user encounters in the surrounding environment. Our system for device grouping named DevLoc takes advantage of ubiquitous light sources around us to perform continuous device grouping based on the similarity of light signals. To control the spatial granularity of user's proximity, we provide a configuration framework to manage the lighting infrastructure through customized visible light communication. We support two modes of device associations to achieve a binding between different entities: device-to-device and device-to-area allowing either proximity-based or location-based services. Our device grouping includes several methods where in general the machine learning based signal similarity performs best compared to distance and correlation metrics.@en

Unikernels are a new lightweight virtualization technology born as an alternative to virtual machines and containers. Geared towards service provisioning for the Internet of Things (IoT) and edge computing, they offer extremely small memory footprint and strong isolation properties. However, the unikernels ecosystem is still in its infancy and lacks quintessential functionalities found in more well-established virtualization technologies. For example, stateful migration is a highly desired feature for mobile edge services in distributed environments which is not yet supported by unikernels. This is one of the shortcomings preventing us from reaping the full benefits of unikernels outside of stateless applications. In this work, we aim bridging this gap with MirageManager: A ready-to-deploy unikernel migration system enabling lossless migration supported by a function-level, application logic check-pointing library of our design. Our evaluation results show that MirageManager is able to lower the service downtime by ∼80%, and drastically reduce the state transfer data by almost ∼100% when comparing against Podman. Additionally, MirageManager also beats Podman, a container-based engine, in parallel service migration across constrained edge networks reducing the overall migration time by up to ∼6x. @en

Bluetooth trackers, or tags, have quickly become ubiquitous and widely supported by multiple vendors. Beyond their original design of finding lost objects, these devices have the ability to extend the capabilities of current wireless smart devices. Since its launch in 2019, Apple’s FindMy enables any devices from their brand to be easily tracked by more than 1 billion active iPhones and iPads on the market. While convenient, these systems may even serve further uses, including as a result of this work, crowd sensing and a side channel for mobile communication. But they also raise privacy concerns for their users. In this paper, we demonstrate how Apple FindMy can be used as a privacy-friendly tool for crowd monitoring, and how it may inadvertently leak information on a person’s location in case of deliberate tracking. Additionally, we design and evaluate a proof of concept protocol, using the Apple FindMy and a crafted tag using a simple microcontroller. We show how such system could be used to transmit information at very low bit rates, while the devices transporting the information remain unaware of this covert channel, yielding an out of band communication channel.@en

Visible Light Communication (VLC)emerges as a communication technology for Internet of Things (IoT)services with appealing benefits not present in existing radio-based communication. However, current VLC designs commonly require dedicated LED lights to emit modulated light beams which entail high energy overhead and unpleasant visual experiences due to the perceptible light blinking effects for end users. This greatly limits the deployment and applicable scenarios of VLC. In this paper, we design and develop LocalVLC, a practical and low-cost VLC system that can be used as a standard light source to augment smart IoT services. LocalVLC introduces a novel Morse-code inspired modulation scheme that can operate on off-the-shelf LEDs with low energy overhead. It can effectively overcome the light flickering by encoding data into high frequency light pulses without requiring extra processing hardware such as FPGA or micro-controller. We have implemented and evaluated a full-fledged system prototype based on LocalVLC design. Under practical settings, our LocalVLC prototype can support up to 10 meters of range, and attain reasonable throughput (up to 1.4 Kbps)with low error rate and energy consumption. Comparing with the widely adopted Manchester encoding, LocalVLC yields 8x improvement on both throughput and energy consumption. In addition, we demonstrate the practicality of LocalVLC through two IoT use cases where we developed two lightweight LocalVLC-based solutions using low-cost off-the-shelf hardware to exemplify the usage of LocalVLC for indoor service discovery and smart home key management.@en

Detecting and reacting efficiently to road condition hazards are challenging given practical restrictions such as limited data availability and lack of infrastructure support. In this paper, we present an edge-cloud chaining solution that bridges the cloud and road infrastructures to enhance road safety. We exploit the roadside infrastructure (e.g., smart lampposts) to form a processing chain at the edge nodes and transmit the essential context to approaching vehicles providing what we refer as road fingerprinting. We approach the problem from two angles: first we focus on semantically defining how an execution pipeline spanning edge and cloud is composed, then we design, implement and evaluate a working prototype based on our assumptions. In addition, we present experimental insights and outline open challenges for next steps.@en

Pervasive computing environments are characterized by a plethora of sensing and communication-enabled devices that diffuse themselves among different users. Built-in sensors and telecommunication infrastructure allow co-presence detection. In turn, co-presence detection enables context-aware applications, like those for social networking among close-by users, and for modeling human behavior. We aim to support developers building better context-aware applications by a deepened understanding of which set of context information is appropriate for co-presence detection. We have gathered a multimodal dataset for proximity sensing, including several proximity verification sets, like Bluetooth, Wi-Fi, and GSM encounters, to be able to associate sensor's data with a spatial granularity. We show that sensor modalities are suitable to recognize the spatial adjacency of users with different spatio-temporal granularity. We find that individual user mobility has only a minor, negligible effect on co-presence detection. In contrast, the heterogeneity of device's sensor hardware has a major negative impact on co-presence detection. To reveal energy pitfalls with respect to usability, we perform an energy analysis pertaining to the usage stemming from different sensors for co-presence detection. @en

Understanding and predicting mobility are essential for the design and evaluation of future mobile edge caching and networking. Consequently, research on human mobility prediction has drawn significant attention in the last decade. Employing information-theoretic concepts and machine learning methods, earlier research has shown evidence that human behavior can be highly predictable. Whether high predictability manifests itself for different modes of device usage, across spatial and temporal dimensions is still debatable. Despite existing studies, more investigations are needed to capture intrinsic mobility characteristics constraining predictability, to explore more dimensions (e.g. device types) and spatiotemporal granularities, especially with the change in human behavior and technology. We investigate practical predictability of next location visitation across three different dimensions: device type, spatial granularity and temporal spans using an extensive longitudinal dataset, with fine spatial granularity (AP level) covering 16 months. The study reveals device type as an important factor affecting predictability. Ultra-portable devices such as smartphones have”on-the-go” mode of usage (and hence dubbed”Flutes”), whereas laptops are”sit-to-use” (dubbed”Cellos”). The goal of this study is to investigate practical prediction mechanisms to quantify predictability as an aspect of human mobility modeling, across time, space and device types. We apply our systematic analysis to wireless traces from a large university campus. We compare several algorithms using varying degrees of temporal and spatial granularity for the two modes of devices; Flutes vs. Cellos. Through our analysis, we quantify how the mobility of Flutes is less predictable than the mobility of Cellos. In addition, this pattern is consistent across various spatio-temporal granularities, and for different methods (Markov chains, neural networks/deep learning, entropy-based estimators). This work substantiates the importance of predictability as an essential aspect of human mobility, with direct application in predictive caching, user behavior modeling and mobility simulations.@en

Mobility is a fundamental characteristic of human society that shapes various aspects of our everyday interactions. This pervasiveness of mobility makes it paramount to understand factors that govern human movement and how it varies across individuals. Currently, factors governing variations in personal mobility are understudied with existing research focusing on explaining the aggregate behaviour of individuals. Indeed, empirical studies have shown that the aggregate behaviour of individuals follows a truncated Lévy-flight model, but little understanding exists of the laws that govern intra-individual variations in mobility resulting from transportation choices, social interactions, and exogenous factors such as location-based mobile applications. Understanding these variations is essential for improving our collective understanding of human mobility, and the factors governing it. In this article, we study the mobility laws of location-based gaming—an emerging and increasingly popular exogenous factor influencing personal mobility. We analyse the mobility changes considering the popular PokémonGO application as a representative example of location-based games and study two datasets with different reporting granularity, one captured through location-based social media, and the other through smartphone application logging. Our analysis shows that location-based games, such as PokémonGO, increase mobility—in line with previous findings—but the characteristics governing mobility remain consistent with a truncated Lévy-flight model and that the increase can be explained by a larger number of short-hops, i.e., individuals explore their local neighborhoods more thoroughly instead of actively visiting new areas. Our results thus suggest that intra-individual variations resulting from location-based gaming can be captured by re-parameterization of existing mobility models.@en

The number of deployed Internet of Things (IoT) devices is steadily increasing to manage and interact with community assets of smart cities, such as transportation systems and power plants. This may lead to degraded network performance due to the growing amount of network traffic and connections generated by various IoT devices. To tackle these issues, one promising direction is to leverage the physical proximity of communicating devices and inter-device communication to achieve low latency, bandwidth efficiency, and resilient services. In this work, we aim at enhancing the performance of indoor IoT communication (e.g., smart homes, SOHO) by taking advantage of emerging technologies such as visible light and ultrasound. This approach increases the network capacity, robustness of network connections across IoT devices, and provides efficient means to enable distance-bounding services. We have developed communication modules using off-the-shelf components for visible light and ultrasound and evaluate their network performance and energy consumption. In addition, we show the efficacy of our communication modules by applying them in a practical indoor IoT scenario to realize secure IoT group communication.@en

Telecommunication (Telco) outdoor position recovery aims to localize outdoor mobile devices by leveraging measurement report (MR) data. Unfortunately, Telco position recovery requires sufficient amount of MR samples across different areas and suffers from high data collection cost. For an area with scarce MR samples, it is hard to achieve good accuracy. In this paper, by leveraging the recently developed transfer learning techniques, we design a novel Telco position recovery framework, called sf TLoc, to transfer good models in the carefully selected source domains (those fine-grained small subareas) to a target one which originally suffers from poor localization accuracy. Specifically, sf TLoc introduces three dedicated components: 1) a new coordinate space to divide an area of interest into smaller domains, 2) a similarity measurement to select best source domains, and 3) an adaptation of an existing transfer learning approach. To the best of our knowledge, sf TLoc is the first framework that demonstrates the efficacy of applying transfer learning in the Telco outdoor position recovery. To exemplify, on the 2G GSM and 4G LTE MR datasets in Shanghai, sf TLoc outperforms a non-transfer approach by 27.58 and 26.12 percent less median errors, and further leads to 47.77 and 49.22 percent less median errors than a recent fingerprinting approach NBL. @en

Telecommunication (Telco) outdoor position recovery aims to localize outdoor mobile devices by leveraging measurement report (MR) data. Unfortunately, Telco position recovery requires sufficient amount of MR samples across different areas and suffers from high data collection cost. For an area with scarce MR samples, it is hard to achieve good accuracy. In this paper, by leveraging the recently developed transfer learning techniques, we design a novel Telco position recovery framework, called sf TLoc, to transfer good models in the carefully selected source domains (those fine-grained small subareas) to a target one which originally suffers from poor localization accuracy. Specifically, sf TLoc introduces three dedicated components: 1) a new coordinate space to divide an area of interest into smaller domains, 2) a similarity measurement to select best source domains, and 3) an adaptation of an existing transfer learning approach. To the best of our knowledge, sf TLoc is the first framework that demonstrates the efficacy of applying transfer learning in the Telco outdoor position recovery. To exemplify, on the 2G GSM and 4G LTE MR datasets in Shanghai, sf TLoc outperforms a non-transfer approach by 27.58 and 26.12 percent less median errors, and further leads to 47.77 and 49.22 percent less median errors than a recent fingerprinting approach NBL. @en

Telecommunication (Telco) outdoor position recovery aims to localize outdoor mobile devices by leveraging measurement report (MR) data. Unfortunately, Telco position recovery requires sufficient amount of MR samples across different areas and suffers from high data collection cost. For an area with scarce MR samples, it is hard to achieve good accuracy. In this paper, by leveraging the recently developed transfer learning techniques, we design a novel Telco position recovery framework, called sf TLoc, to transfer good models in the carefully selected source domains (those fine-grained small subareas) to a target one which originally suffers from poor localization accuracy. Specifically, sf TLoc introduces three dedicated components: 1) a new coordinate space to divide an area of interest into smaller domains, 2) a similarity measurement to select best source domains, and 3) an adaptation of an existing transfer learning approach. To the best of our knowledge, sf TLoc is the first framework that demonstrates the efficacy of applying transfer learning in the Telco outdoor position recovery. To exemplify, on the 2G GSM and 4G LTE MR datasets in Shanghai, sf TLoc outperforms a non-transfer approach by 27.58 and 26.12 percent less median errors, and further leads to 47.77 and 49.22 percent less median errors than a recent fingerprinting approach NBL. @en

Future computing environments are embedded with many sensors for applications like augmented reality. Much of the deployed Internet of Things (IoT) technology is designed to be invisible. To support user's privacy awareness, a map of surrounding sensing devices is beneficial to determine the nature of data collection taking place in any given area. Moreover, security and governance issues are among the challenges IoT poses to organizations which might not know exactly which IoT devices are connected to their network. For instance, many employees bringing their own devices to the workplace. We explore the feasibility to use small COTS drones to create indoor maps of wireless devices. These comprehensive device maps serve as basis for device localization and monitoring to enhance user privacy and network security. We analyze the impact of our device management platform at the drone's energy consumption and evaluate the device detection rate, explored area, and localization error. Due to the restricted battery capacity of the drone, we simulate larger areas with a varying number of IoT devices to highlight the limits of our drone-based device management platform regarding area exploration and reachable IoT devices.@en

Towards private proximity services we realized a set of proximity services at different spatial resolutions. For small-scale (0.5 m) securing remote access to smart homes and for mid-scale (10-20 m) to manage nearby Internet of Things (IoT) devices and offer fine-grained service discovery in indoor environments. Regarding large-scale services (100 m) we implemented a device grouping via similarity of light patterns ambient sound Wi-Fi signals and ultrasound communication which is naturally restricted by spatial barriers. To improve user's privacy from a system point of view we analyzed different security mechanisms in the domain of device-to-device (D2D) communication such as access control location privacy. Based on visible light communication (VLC) we are implemented and tested a system for private indoor service discovery and distance-bounding authorization. Furthermore we examined the feasibility of homomorphic encryption for time-series data like visible light patterns.@en