Over twenty years, ubiquitous computing research has transformed everyday environments into rich sources of behavioral insight. Today, data-driven models built on mobile, wearable, and physiological sensors can infer activities, routines, affect, and even identity – often beyond what system designers intended. This keynote revisits the field’s evolution through the lens of "the good, the bad, and the unknown": the good –powerful models enabling health, education, and well-being applications; the bad – fragility, bias, and opaque inference pipelines; and the unknown – emergent, unintended capabilities in AI-driven sensing systems. I will outline open research directions for a future where pervasive systems remain trustworthy, meaningful, and aligned with human values.

Concept-based models are an emerging paradigm in deep learning that constrains the inference process to operate through human-interpretable variables, facilitating explainability and human interaction. However, these architectures, on par with popular opaque neural models, fail to account for the true causal mechanisms underlying the target phenomena represented in the data. This hampers their ability to support causal reasoning tasks, limits out-of-distribution generalization, and hinders the implementation of fairness constraints. To overcome these issues, we propose Causally reliable Concept Bottleneck Models (C$^2$BMs), a class of concept-based architectures that enforce reasoning through a bottleneck of concepts structured according to a model of the real-world causal mechanisms. We also introduce a pipeline to automatically learn this structure from observational data and unstructured background knowledge (e.g., scientific literature). Experimental evidence suggests that C$^2$BMs are more interpretable, causally reliable, and improve responsiveness to interventions w.r.t. standard opaque and concept-based models, while maintaining their accuracy.

Human Activity Recognition (HAR) benefits various application domains, including health and elderly care. Traditional HAR involves constructing pipelines reliant on centralized user data, which can pose privacy concerns as they necessitate the uploading of user data to a centralized server. This work proposes multi-frequency Federated Learning (FL) to enable: (1) privacy-aware ML; (2) joint ML model learning across devices with varying sampling frequency. We focus on head-worn devices (e.g., earbuds and smart glasses), a relatively unexplored domain compared to traditional smartwatch- or smartphone-based HAR. Results have shown improvements on two datasets against frequency-specific approaches, indicating a promising future in the multi-frequency FL-HAR task. The proposed network’s implementation is publicly available for further research and development.**

Sensor-based sleep monitoring systems can be used to track sleep behavior on a daily basis and provide feedback to their users to promote health and well-being. Such systems can provide data visualizations to enable self-reflection on sleep habits or a sleep coaching service to improve sleep quality. To provide useful feedback, sleep monitoring systems must be able to recognize whether an individual is sleeping or awake. Existing approaches to infer sleep-wake phases, however, typically assume continuous streams of data to be available at inference time. In real-world settings, though, data streams or data samples may be missing, causing severe performance degradation of models trained on complete data streams. In this paper, we investigate the impact of missing data to recognize sleep and wake, and use regression- and interpolation-based imputation strategies to mitigate the errors that might be caused by incomplete data. To evaluate our approach, we use a data set that includes physiological traces - collected using wristbands -, behavioral data - gathered using smartphones - and self-reports from 16 participants over 30 days. Our results show that the presence of missing sensor data degrades the balanced accuracy of the classifier on average by 10–35 percentage points for detecting sleep and wake depending on the missing data rate. The impu-tation strategies explored in this work increase the performance of the classifier by 4–30 percentage points. These results open up new opportunities to improve the robustness of sleep monitoring systems against missing data.

Internet of Things (IoT) applications build upon resource-constrained, distributed devices that generate data and enable communication. For such applications to be truly trustworthy, it must be ensured that the devices are not compromised by malicious software. Remote attestation (RA), a prominent technique, exploits challenge-response protocols to detect malware on remote devices. Given the increasing scale and number of IoT deployments, recent work on RA has explored collective attestation ofswarmsof devices. However state-of-the-art swarm attestation techniques require trusted hardware which makes them inapplicable to both legacy and next generation IoT deployments without trusted hardware. We present SWARNA, asoftware-basedswarm attestation for IoT devices. After highlighting the challenges in designing such a solution, we present two protocol variants for IEEE 802.15.4 TSCH networks. We assess their performance analytically and empirically through testbed experiments. SWARNA maintains a constant payload size whereas, it increases linearly with the network size for existing solutions requiring trusted hardware. The two protocol variants attest 30 nodes networks, in 6s and 1.5s to 8.2s, respectively, depending on the number of malicious nodes. Further, we demonstrate that attestation traffic has a negligible impact on the packet delivery ratio (0.4 percent drop) of a typical data collection application.

Flow is a positive affective state occurring when individuals are fully immersed into an activity. Being in flow during work activities can lead to higher performance and productivity. Despite the importance of flow at work, few approaches have been proposed for its automatic recognition using sensor data and most existing studies are conducted in laboratory settings with simulated work activities. In this paper, we investigate the use of physiological data, collected using wrist-worn devices, combined with context information, obtained through self-reports, to automatically distinguish between low and high levels of flow. We investigate the role of the context for flow perceptions and in its automatic recognition. Further, we compare the performance of several sensor fusion strategies based on shallow and deep learning. To evaluate our approach we use a data set of 390 activities collected during actual work days. Our results show that using raw blood volume pulse, electrodermal activity and the type of activity as input to a sensor-based late fusion approach, implemented using convolutional neural networks, allows to reach a balanced accuracy of 70.93%.

Head gestures and facial expressions – like, e.g., nodding or smiling – are important indicators of the quality of human interactions in physical meetings as well as in computer-mediated settings. Computer systems able to recognize such behavioral cues can support and improve human interactions. Several researchers have thus tackled the problem of automatically recognizing head gestures and facial expressions, mainly leveraging video data. In this paper, we instead consider inertial signals collected from unobtrusive, ear-mounted devices. We focus on typical activities performed during social interactions – head shaking, nodding, smiling, talking and yawning – and propose a hierarchical classification approach to discriminate them from each other. Further, we investigate whether the transfer of knowledge learned from publicly available datasets leads to further performance improvements. Our results show that the combined use of our hierarchical approach and transfer learning allows the classifier to discriminate head and mouth activities with an F1 score of 84.79, smile, talk and yawn with an F1 score of 45.42, and nodding and head shaking with an F1 score of 88.24, outperforming shallow classifiers by 2-9 percentage points.

This article presents Whisper, a fast and reliable protocol to flood small amounts of data into a multi-hop network. Whisper makes use of synchronous transmissions, a technique first introduced by the Glossy flooding protocol. In contrast to Glossy, Whisper does not let the radio switch from receive to transmit mode between messages. Instead, it makes nodes continuously transmit identical copies of the message and eliminates the gaps between subsequent transmissions. To this end, Whisper embeds the message to be flooded into a signaling packet that is composed of multiple packlets —where a packlet is a portion of the message payload that mimics the structure of an actual packet. A node must intercept only one of the packlets to detect that there is an ongoing transmission and that it should start forwarding the message. This allows Whisper to speed up the propagation of the flood and, thus, to reduce the overall radio-on time of the nodes. Our evaluation on the FlockLab testbed shows that Whisper achieves comparable reliability but 2× lower radio-on time than Glossy. We further show that by embedding Whisper in an existing data collection application, we can more than double the lifetime of the network.

Increasing the traffic visibility, by monitoring network flow packets, provides valuable information for various network management tasks. The mirroring mode of flow packet monitoring requires the switches and routers to duplicate packets of interest, and to send them to flow monitors for in-depth analysis. A common practice to avoid the interference between the original and the mirrored flows is to transmit them separately, in two different planes (data plane and monitoring plane, respectively). In this paper, we aim at reducing the overall cost of transmitting both the original and mirrored flows. Towards that end, we present a generic monitoring framework called REMO. The key idea of REMO is twofold: (i) placing the flow monitors in central locations, and (ii) passing the original flows through the vicinity of the monitors. By doing so, REMO reduces the resources consumed in the monitoring plane, without unworthily increasing the resource consumption in the data plane. The results of extensive numerical simulations show that REMO effectively reduces the overall transmission cost, remarkably outperforming several baseline strategies, particularly when the transmission is more expensive in the monitoring plane.

This demo paper presents EConVi, a framework to support the implementation and visualization of the energy efficient software defined networking (SDN). This framework monitors the network workload, and dynamically changes the operational states of the switches and routing paths of flows, based on the controlling algorithms. The network topology, traffic workloads on each network link and switch, and the operational states of the switches are visualized. EConVi is able to manage the large-scaled networks simulated in the SDN emulators such as Mininet, or interact with DVFS-enabled hardware nodes installed with software switches, as shown in our demo.

A large plethora of models to predict human mobility exists in the literature. The problem of how to select the most appropriate model to solve a specific mobility prediction task has however received only little attention. Yet, a wrong model choice may lead to severe performance losses. In this paper, we address the model selection problem in human mobility prediction and make the following contributions. We present SELECTOR, a generic framework to explore human mobility data and compute both population models and individual models to predict human mobility. The former are models that are adapted to the characteristics of an entire population of users and can be used to overcome the cold-start problem. The latter are prediction models optimized for individual users. We present and analyze the results obtainable using SELECTOR on the Nokia data set, which is one of the largest and richest, publicly available data sets of human mobility data. We show that for many users, generic population models can be used in place of individual models with negligible performance losses. Yet for about 25 percent of the users, individual models perform at least three percentage points better than population models. Thereby, we show that the use of phone context data does not lead to significantly better performance of human mobility predictors with respect to the case in which only temporal and spatial features are used. We further observe that the population models we derive are robust against the demographics of the users and that building different population models for different periods of the day leads to performance improvements. We make SELECTOR publicly available to allow other researchers and practitioners to explore further mobility data sets and to embed the code base of SELECTOR in their applications.

Data center networks (DCNs) continuously evolve to support emerging applications. For instance, server-centric architectures are becoming popular because they improve the modularity and resilience of data centers. In a server-centric DCN, switching functionalities are embedded in each server instead of in dedicated switches. Recently, wireless technologies have been introduced into the design of server-centric DCNs. Wireless interconnects provide great flexibility by creating direct flyways between servers. These wireless links help reducing the length of paths assigned to flows so that fewer links must be powered-on to transfer data. In this paper, we investigate energy-aware scheduling of coflows and antenna directions in a server-centric DCN. Coflows are groups of parallel and high-volume flows generated by big data applications. We assume that the DCN is hybrid, i.e., that it includes both wired and wireless links. We formulate this scheduling problem as an optimization problem and solve it using a mathematical programming solver. We further introduce an heuristic algorithm to reduce computational complexity. Our simulation results show that our optimal solution saves up to 44.9% of energy compared with representative competitors. Our heuristic algorithm achieves lower energy savings but still generates the shortest paths and lowest transferring time of flows.

Emerging protocols for low-power wireless networks increasingly exploit constructive interference and the capture effect. The basic idea is that the synchronous transmission of identical packets by neighboring nodes leads to constructive interference - or at least do not cause destructive interference. This requires that the temporal displacement of packets at receiving nodes is lower than 0.5 μs when employing IEEE 802.15.4 radios. However, commonly used sensor nodes are equipped with cheap and imprecise clocks that show high frequency deviations across nodes, making constructive interference difficult to achieve. Such deviations further increase when individual nodes are exposed to different temperatures. In this paper we introduce Flock, a novel approach to compensate for differences in clock frequency across synchronously transmitting nodes. We implemented Flock in Contiki on the example of Glossy, a flooding protocol based on synchronous transmissions. Our results confirm that Flock can achieve constructive interference on real sensor nodes in over 98% of the cases. Overall, Flock makes protocols that exploit synchronous transmissions more robust to operate even in challenging environments.

Standard Internet communication protocols are key enablers for the Internet of Things (IoT). Recent technological advances have made it possible to run such protocols on resource-constrained devices. Yet these devices often use energy-efficient, low-level communication technologies, like IEEE 802.15.4, which suffer from low-reliability and high latency. These drawbacks can be significantly reduced if communication occurs using concurrent transmissions - a novel communication paradigm for resource-constrained devices. In this paper, we show that Internet protocols like TCP/UDP and CoAP can run efficiently on top of a routing substrate based on concurrent transmissions. We call this substrate LaneFlood and demonstrate its effectiveness through extensive experiments on Flocklab, a publicly available testbed. Our results show that LaneFlood improves upon CXFS - a representative competitor - in terms of both duty cycle and reliability. Furthermore, LaneFlood can transport IoT traffic with an end-to-end latency of less than 300 ms over several hops.

Occupancy monitoring (i.e. sensing whether a building or room is currently occupied) is required by many building automation systems. An automatic heating system may, for example, use occupancy data to regulate the indoor temperature. Occupancy data is often obtained through dedicated hardware such as passive infrared sensors and magnetic reed switches. In this paper, we derive occupancy information from electric load curves measured by off-the-shelf smart electricity meters. Using the publicly available ECO dataset, we show that supervised machine learning algorithms can extract occupancy information with an accuracy between 83% and 94%. To this end we use a comprehensive feature set containing 35 features. Thereby we found that the inclusion of features that capture changes in the activation state of appliances provides the best occupancy detection accuracy.

The use of concurrent transmissions allows protocols to achieve high reliable, ultra-low latency communication in homogeneous wireless sensor networks. However, applications for wireless sensor networks must often operate over heterogeneous nodes. In this work, we provide a first step towards enabling concurrent transmissions in heterogeneous networks. We present a methodology to implement the Glossy communication protocol on a number of different hardware platforms. We further compare the performance of Glossy on two exemplary platforms -- the Tmote Sky and the WiSMote. We show that even small differences in the underlying hardware can influence the performance of Glossy significantly.

Existing measurement frameworks typically assume that the communication protocols and mechanisms running on the devices do not change during network operation. However, recent research efforts show that by enabling devices to switch between protocols and mechanisms at runtime the overall network performance can be improved. In this paper, a novel measurement framework that enables the continuous and consistent measurement of monitoring metrics even across such adaptations in networks is presented. The framework exploits monitoring metrics locally on the devices (i) irrespectively of the used mechanisms or protocols on the devices and (ii) allows other mechanisms and applications in the network to adapt to changes by referring monitoring information from the framework. A proof-of-concept prototype of the measurement framework is used to show that the work represents a promising step towards protocol-independent, adaptive monitoring in dynamic networks.

Several algorithms to predict the next place visited by a user have been proposed in the literature. The accuracy of these algorithms -- measured as the ratio of the number of correct predictions and the number of all computed predictions -- is typically very high. In this paper, we show that this good performance is due to the high predictability intrinsic in human mobility. We also show that most algorithms fail to correctly predict transitions, i.e. situations in which users move between different places. To this end, we analyze the performance of 18 prediction algorithms focusing on their ability to predict transitions. We run our analysis on a data set of mobility traces of 37 users collected over a period of 1.5 years. Our results show that even algorithms achieving an overall high accuracy are unable to reliably predict the next location of the user if this is different from the current one. Building upon our analysis we then present a novel next-place prediction algorithm that can both achieve high overall accuracy and reliably predict transitions. Our approach combines all the 18 algorithms considered in our analysis and achieves its good performance at the cost of a higher computational and memory overhead.

Predicting the arrival and residence time of individuals at their relevant places enables a plethora of novel applications. In this work we first analyze the theoretical predictability of arrival and residence times and then evaluate the performance of eight different residence time predictors. We show that these predictors tend to underestimate the time a user will spend at her relevant places.

Novel protocols that can enable ultra-low latency communication in wireless multihop sensor and actuator networks have recently been presented in the literature. These approaches achieve ultra-low latencies in packet delivery notwithstanding the unreliability of the wireless channel. A systematic methodology to analyze and compare the performance of ultra-low latency communication protocols is however still missing. This work presents our first steps towards the definition of such a methodology. Building upon this work, we aim at designing and implementing a comprehensive benchmarking framework for ultra-low latency communication protocols.

Comparing experimental results obtained on different wireless sensor network deployments is typically very cumbersome and in most cases unfeasible. This is due to the lack of a methodology to describe the properties of network deployments and the experimental conditions under which experiments have been run. Our work focuses on the design and development of a site properties assessment framework, called SiteWork, that aims at providing the means to quickly, automatically and accurately quantify their properties. This poster abstract describes the preliminary design and evaluation of the basic site properties assessment mechanisms provided by SiteWork.

This poster presents the homeset algorithm, a lightweight approach to estimate occupancy schedules of private households. The algorithm relies on the mobile phones of households' occupants to collect Wi-Fi scans. The scans are then used to determine if occupants are at home or not. The algorithm operates in an autonomous fashion using only information available locally on the mobile phones. We validate our approach using a data set from the Nokia Lausanne Data Collection Campaign.

In tree-structured data collection sensor networks, packets are routed towards a sink node by iteratively choosing a node's immediate parent node as the next hop. It is however beyond the scope of these routing protocols to transfer messages along the reverse path, i.e., from the sink to individual nodes in the network. In this paper, we present CBFR, a novel routing scheme that builds upon collection protocols to enable efficient point-to-point communication. We propose the use of space-efficient data structures known as Bloom filters to efficiently store routing tables on the networked devices. In particular, each node in the collection tree stores the addresses of its direct and indirect child nodes in its local Bloom filter. A packet is forwarded down-tree only if the node's local filter indicates the presence of the packet's destination address among the node's descendants. In order to cater for the presence of mobile nodes, we apply the concept of counting Bloom filters to allow for the removal of elements from the filter by means of gradual forgetting. The effectiveness of our approach in achieving both high delivery rates and low overhead is demonstrated by means of simulations and experiments.