Human Activity Recognition (HAR) is vital for proactive health monitoring, but current systems are hindered by fixed-length time windows and the labor-intensive process of collecting annotated data which complicates effective HAR classifier development. To address these challenges, we contribute LLMSEG, a novel two-stage framework that integrates dynamic segmentation and HAR classification. Utilizing an LLM-based segmentation approach with signature activities, LLMSEG dynamically adjusts window lengths based on contextual activity, significantly enhancing accuracy over traditional methods. Additionally, it generates sensor data descriptions enriched with contextual cues, enabling effective few-shot classification without extensive labeled datasets. Systematic evaluations across various HAR datasets demonstrate that LLMSEG outperforms fixed-window methods, balances expressivity and efficiency, and generalizes well across activities. Performance can be further improved by carefully designed contextual prompts. Furthermore, LLMSEG supports robust deployment on both on-device (Raspberry Pi) and edge (laptop) devices. Overall, LLMSEG enhances the generality and reliability of HAR solutions while accommodating diverse deployment scenarios.

Large language models (LLMs) are reshaping the educational landscape, particularly in online learning environments where student supervision is often limited. Early evidence and anecdotal reports suggest that the use of AI-generated content is highly prevalent among students. However, definitive statistics remain elusive, primarily due to the challenges associated with distinguishing between AI-generated and human-generated responses. Establishing clear evidence and effective mechanisms for identifying AI-generated responses is crucial for understanding the significance of this challenge and for developing policies to address it. To tackle these issues, we present a large-scale empirical study on the prevalence of AI-generated content in online education. Our study analyzes over 4045 student responses from an introductory MOOC on the Internet of Things, employing textual analysis techniques to evaluate various metrics for identifying AI-generated responses and understanding their characteristics. Our findings reveal that a significant majority of student responses (up to 90.1%) exhibit strong similarities to AI-generated content in both wording and contextual meaning, regardless of the specific LLM or similarity metric employed. In terms LLMs usage, DeepSeek, Gemini, and Grok are the three most popular LLMs used to generate responses.

Large-scale datasets, particularly those collected from smart devices and Internet of Things sensors, usually exhibit significant temporal and spatial sparsity, resulting in high amounts of missing data. Unless addressed in the analysis, this sparsity can result in substantial gaps and biases as well as limit the generalizability of conclusions drawn from such data. To address this challenge in data quality, we contribute the Sparsity-Aware Reconstruction Framework (SARF) as a novel and unified data fusion and reconstruction framework that enhances data quality and addresses sparsity. SARF analyzes datasets, partitioning the data into segments with similar characteristics, and reconstructs the data in each segment individually by selecting a reconstruction technique that is tailored to the internal temporal-spatial characteristics of the dataset. Through extensive experiments on two representative datasets - mobile application measurements and IoT sensor data from low-cost air quality sensors - we demonstrate that the targeted adaptation of reconstruction strategies employed by SARF significantly enhances the quality of reconstructed data. Our results show the robustness of SARF's performance across spatiotemporal variations, outperforming current state-of-the-art methods by margins up to 68% on average (74% for compressive sensing, 53% for convolutional sparse coding, 78% for deep learning). These findings underscore SARF's potential to enhance datadriven insights across multiple domains, paving the way for more robust analyses of sparsity-affected datasets.

We present SNAKE, an innovative method that harnesses heat transferred from human touch interactions to estimate product quality. SNAKE offers an accessible and cost-effective solution that seamlessly integrates with existing retail practices; for example, it can be integrated with scales and cashiers already present in shops. Rigorous and systematic experiments demonstrate that SNAKE achieves a high level of accuracy (83%) and outperforms optical sensing and WiFi sensing baselines. We also provide evidence that SNAKE can capture touch interactions of different durations and maintain consistency across diverse user profiles and operating environments. To assess the potential for practical impact, we also carry out an additional user study (N = 100) which suggests that SNAKE has potential to improve consumer purchasing decisions by at least 25% and reduce food waste (or increase promotional opportunities) by 10%–15%. In summary, our contribution offers a novel solution for leveraging smart IoT solutions to support retailing and foster sustainable retail practices.

We study user's valuations of smartphone sensing resources and the factors mediating them through a systematic auction study with 108 bids from$N=18$participants, two resource use conditions [fixed battery (FB) and variable battery (VB)] and three sensors (camera, microphone, and GPS) with differing energy and privacy costs. We use a second-price sealed-bid reverse auction as this allows us to elicit the participants’ truthful perceived value for sharing resources. We show that most users would be willing to share even highly-privacy intrusive sensors if they are sufficiently compensated. At the FB level, participants placed much lower value for sharing GPS (€13) than camera (€30) or microphone (€32.5). The values people place on sharing access to resources generally reflect four considerations: 1) the perceived value of the sensor type; 2) the value of the data captured by the sensor; 3) the impact of sharing on the device; and 4) personal variations related to sharing motives, personal tendencies, and the broader sharing context. We address the practical impact of our results by presenting two case studies (collaborative sensing and collaborative AI). Finally, we derive design implications for sharing sensing resources on personal devices.

AI-assisted interactive annotation is a powerful way to facilitate data annotation—a prerequisite for constructing robust AI models. While AI-assisted interactive annotation has been extensively studied in static settings, less is known about its usage in dynamic scenarios where the annotators operate under time and cognitive constraints, e.g., while detecting suspicious or dangerous activities from real-time surveillance feeds. Understanding how AI can assist annotators in these tasks and facilitate consistent annotation is paramount to ensure high performance for AI models trained on these data. We address this gap in interactive machine learning (IML) research, contributing an extensive investigation of the benefits, limitations, and challenges of AI-assisted annotation in dynamic application use cases. We address both the effects of AI on annotators and the effects of (AI) annotations on the performance of AI models trained on annotated data in real-time video annotations. We conduct extensive experiments that compare annotation performance at two annotator levels (expert and non-expert) and two interactive labeling techniques (with and without AI assistance). In a controlled study with \(N=34\) annotators and a follow-up study with 51,963 images and their annotation labels being input to the AI model, we demonstrate that the benefits of AI-assisted models are greatest for non-expert users and for cases where targets are only partially or briefly visible. The expert users tend to outperform or achieve similar performance as the AI model. Labels combining AI and expert annotations result in the best overall performance as the AI reduces overflow and latency in the expert annotations. We derive guidelines for the use of AI-assisted human annotation in real-time dynamic use cases.

Underwater environments are emerging as a new frontier for data science thanks to an increase in deployments of underwater sensor technology. Challenges in operating computing underwater combined with a lack of high-speed communication technology covering most aquatic areas means that there is a significant delay between the collection and analysis of data. This in turn limits the scale and complexity of the applications that can operate based on these data. In this article, we develop underwater fog computing support using low-cost micro-clouds and demonstrate how they can be used to deliver cost-effective support for data-heavy underwater applications. We develop a proof-of-concept micro-cloud prototype and use it to perform extensive benchmarks that evaluate the suitability of underwater micro-clouds for diverse underwater data science scenarios. We conduct rigorous tests in both controlled and field deployments, using river and sea waters. We also address technical challenges in enabling underwater fogs, evaluating the performance of different communication interfaces and demonstrating how accelerometers can be used to detect the likelihood of communication failures and determine which communication interface to use. Our work offers a cost-effective way to increase the scale and complexity of underwater data science applications, and demonstrates how off-the-shelf devices can be adopted for this purpose.

Humans tend to spend most of their life indoors, making the quality of indoor environments essential for human health and wellbeing. While several solutions for monitoring the indoor environment have been proposed, ranging from infrastructure-based monitoring solutions to cameras, these tend to require separate installation, making the sensors difficult to maintain and upgrade. In this article, we introduce the idea of using smart plants as an easy-to-deploy and affordable solution for monitoring the indoor environment. Plants are typically deployed close to humans and they increasingly are placed in containers that integrate sensors, such as soil moisture, temperature, humidity, and CO2 sensors. We demonstrate how these sensors can be used as an alternative technology for monitoring—and enriching—indoor spaces without needing to install proprietary sensors or other technology. Specifically, we show how smart plants can be used to estimate overall CO2 accumulation, occupancy information, and whether people use protective face masks or not. We also establish a research roadmap for the use of smart plants to monitor indoor environments.

We contribute MIDAS as a novel sensing solution for characterizing everyday objects using thermal dissipation. MIDAS takes advantage of the fact that anytime a person touches an object, it results in heat transfer. By capturing and modeling the dissipation of the transferred heat, e.g., through the decrease in the captured thermal radiation, MIDAS can characterize the object and determine its material. We validate MIDAS through extensive empirical benchmarks and demonstrate that MIDAS offers an innovative sensing modality that can recognize a wide range of materials – with up to 83% accuracy – and generalize to variations in the people interacting with objects.

The ever-increasing ubiquity of smart devices is creating new opportunities for people to interact and engage with digital information using multiple devices. In the simplest case this can refer to choosing which device to use for a particular task (e.g., phone, laptop or smartwatch), whereas a more complex example is simultaneously taking advantage of the capabilities of different devices (e.g., laptop and smart TV). Despite these types of opportunities becoming increasing available, currently the full potential of multi-device interactions is not being realized as people struggle to take advantage of them. As our first contribution, we study people’s willingness to engage with multi-device interactions and rank the factors that mediate this response through an online survey (N = 60). Our results show that users are strongly in favour of using multiple devices, but lack the awareness or information to engage with them, or feel that establishing the interactions is too laborious and would disrupt the fluidity of the interactions. Motivated by this result, as our second contribution we design and evaluate intelligent shifting cues, visualizations that offer information about available interaction opportunities and how to establish them, and study how they influence users willingness to engage in multi-device interactions. Results of our study show that the cues can be effective in helping people to engage with multiple devices, but that the suitability of the proposed device and fit with task are important mediating factors. We end the paper by deriving design implications for intelligent systems that can support people in engaging with multi-device interactions.

Pervasive availability of programmable smart de-vices is giving rise to sensing and computing scenarios that involve collaboration between multiple devices. Maximizing the benefits of collaboration requires careful selection of devices with whom to collaborate as otherwise collaboration may be interrupted prematurely or be sub-optimal for the characteristics of the task at hand. Existing research on collaborative scenarios has mostly focused on providing mechanisms that can establish and harness collaboration, without considering how to maximally benefit from it. In this paper, we contribute by developing COSINE as a novel approach for selecting collaborators in multi-device computing scenarios. COSINE identifies and recommends collaborators based on a novel information theoretic measure based on Markov trajectory entropy. Rigorous experimental benchmarks carried out using a large-scale dataset of device-to-device encounters demonstrate that COSINE can significantly improve collaboration benefits compared to current state-of-the-art solutions, increasing expected duration of collaboration and reducing variability of collaborations.

We contribute by quantifying the effect of network latency and battery consumption on mobile app performance and retention, i.e., user's decisions to continue or stop using apps. We perform our analysis by fusing two large-scale crowdsensed datasets collected by piggybacking on information captured by mobile apps. We find that app performance has an impact in its retention rate. Our results demonstrate that high energy consumption and high latency decrease the likelihood of retaining an app. Conversely, we show that reducing latency or energy consumption does not guarantee higher likelihood of retention as long as they are within reasonable standards of performance. However, we also demonstrate that what is considered reasonable depends on what users have been accustomed to, with device and network characteristics, and app category playing a role. As our second contribution, we develop a model for predicting retention based on performance metrics. We demonstrate the benefits of our model through empirical benchmarks which show that our model not only predicts retention accurately, but generalizes well across application categories, locations and other factors moderating the effect of performance.