We contribute a novel model evaluation technique that divides available measurements into training and testing sets in a way that adheres to the requirements imposed on professional monitoring stations. We perform extensive and systematic experiments with a wide range of state-of-the-art calibration models to demonstrate that our approach provides accurate insights about the performance of calibration models in real-world deployments, while at the same time highlighting issues with evaluation techniques used in previous works. Among others, our results show that although trained and tested in the same location, calibration errors can exhibit deviation up to 116% depending on the evaluation protocol that is being adopted. We also demonstrate that models trained with continuous data can suffer up to 76% greater error when tested with data coming from diverse environmental conditions. In contrast, when models are trained and tested with our method, the variability of errors is significantly reduced and the robustness of calibration models is significantly improved. The overall performance improvements depend on pollutant concentration, ranging from 10% for low concentrations to 90% for high concentrations that represent conditions that are most dangerous for human health.

The outbreak of Covid-19 changed the world as well as human behavior. In this article, we study the impact of Covid-19 on smartphone usage. We gather smartphone usage records from a global data collection platform called Carat, including the usage of mobile users in North America from November 2019 to April 2020. We then conduct the first study on the differences in smartphone usage across the outbreak of Covid-19. We discover that Covid-19 leads to a decrease in users' smartphone engagement and network switches, but an increase in WiFi usage. Also, its outbreak causes new typical diurnal patterns of both memory usage and WiFi usage. Additionally, we investigate the correlations between smartphone usage and daily confirmed cases of Covid-19. The results reveal that memory usage, WiFi usage, and network switches of smartphones have significant correlations, whose absolute values of Pearson coefficients are greater than 0.8. Moreover, smartphone usage behavior has the strongest correlation with the Covid-19 cases occurring after it, which exhibits the potential of inferring outbreak status. By conducting extensive experiments, we demonstrate that for the inference of outbreak stages, both Macro-F1 and Micro-F1 can achieve over 0.8. Our findings explore the values of smartphone usage data for fighting against the epidemic.

The significance of air pollution and the problems associated with it are fueling deployments of air quality monitoring stations worldwide. The most common approach for air quality monitoring is to rely on environmental monitoring stations, which unfortunately are very expensive both to acquire and to maintain. Hence, environmental monitoring stations are typically sparsely deployed, resulting in limited spatial resolution for measurements. Recently, low-cost air quality sensors have emerged as an alternative that can improve the granularity of monitoring. The use of low-cost air quality sensors, however, presents several challenges: They suffer from cross-sensitivities between different ambient pollutants; they can be affected by external factors, such as traffic, weather changes, and human behavior; and their accuracy degrades over time. Periodic re-calibration can improve the accuracy of low-cost sensors, particularly with machine-learning-based calibration, which has shown great promise due to its capability to calibrate sensors in-field. In this article, we survey the rapidly growing research landscape of low-cost sensor technologies for air quality monitoring and their calibration using machine learning techniques. We also identify open research challenges and present directions for future research.

Today, an increasing number of systems produce, process, and store personal and intimate data. Such data has plenty of potential for entirely new types of software applications, as well as for improving old applications, particularly in the domain of smart healthcare. However, utilizing this data, especially when it is continuously generated by sensors and other devices, with the current approaches is complex—data is often using proprietary formats and storage, and mixing and matching data of different origin is not easy. Furthermore, many of the systems are such that they should stimulate interactions with humans, which further complicates the systems. In this article, we introduce the Human Data Model—a new tool and a programming model for programmers and end users with scripting skills that help combine data from various sources, perform computations, and develop and schedule computer-human interactions. Written in JavaScript, the software implementing the model can be run on almost any computer either inside the browser or using Node.js. Its source code can be freely downloaded from GitHub, and the implementation can be used with the existing IoT platforms. As a whole, the work is inspired by several interviews with professionals, and an online survey among healthcare and education professionals, where the results show that the interviewed subjects almost entirely lack ideas on how to benefit the ever-increasing amount of data measured of the humans. We believe that this is because of the missing support for programming models for accessing and handling the data, which can be satisfied with the Human Data Model.

As smartphones are increasingly an integral part of daily life, recent literature suggests a deeper relationship between personality traits and smartphone usage. However, this relationship depends on many complex factors such as geographic location, demographics, or cultural influence, just to name a few. These factors provide crucial knowledge for e.g. usage support, recommendations, marketing, general usage improvements. We use six months of application usage data from 739 Android smartphone user together with the IPIP 50-item Big Five personality traits questionnaire. As our main contribution, we show that even category-level aggregated application usage can predict Big Five traits at up to 86%–96% prediction fit in our sample. Our results show the effect of personality traits on application usage (mean error improvement on random guess 17.0%). We also identify which application usage data best describe the Big Five personality traits. Our work enables future personality-driven research, and shows that when studying personality, application categories can provide sufficient predictions in general traits.

Air pollution is a major problem in urban areas, where high population density is accompanied with excess anthropomorphic emissions impacting the environment and increasing health effects. Highly accurate air quality monitoring stations have been used to monitor the severity of the problem and warn citizens. However, air quality can vary sharply even within the same city block, and pollution exposure can vary even 30% between individuals living in the same residence. Therefore, a dense deployment of air quality sensors is needed to detect these variations, and protect citizens from overexposure. Low-cost air quality sensors make it possible to densely instrument a city and detect hot spots as they happen. However, thus far limited information exists on their accuracy and practicability. In this paper, we conduct a 44-day measurement campaign to assess performance of low-cost air quality monitors under different environmental conditions. As practical use case, we consider pollution hot spot detection. Our results show that the mean error of low-cost sensors is small, but the variation in error is significantly larger than with reference sensors. We also show that the accuracy is sufficient for applications relying on variations in air quality index values, such as hot spot detection.

Poor indoor air quality is a significant burden to society that can cause health issues and decrease productivity. According to research, indoor air quality is intrinsically linked with human activity and mobility. Indeed, mobility is directly linked with transfer of small particles (e.g. PM2.5) and extent of activity affects production of CO2. Currently, however, estimation of indoor quality is difficult, requiring deployment of highly specialized sensing devices which need to be carefully placed and maintained. In this paper, we contribute by examining the suitability of infrastructure-based motion detectors for indoor air quality estimation. Such sensors are increasingly being deployed into smart environments, e.g., to control lighting and ventilation for energy management purposes. Being able to take advantage of these sensors would thus provide a cost-effective solution for indoor quality monitoring without need for deploying additional sensors. We perform a feasibility study considering measurements collected from a smart office environment having a dense deployment of motion detectors and correlating measurements obtained from motion detectors against air quality values. We consider two main pollutants, PM2.5and CO2, and demonstrate that there indeed is a connection between extent of movement and PM2.5concentration. However, for CO2, no relationship can be established, mostly due to difficulties in separating between people passing by and those residing long-term in the environment.

Wearable thermal imaging is emerging as a powerful and increasingly affordable sensing technology. Current thermal imaging solutions are mostly based on uncooled forward looking infrared (FLIR), which is susceptible to errors resulting from warming of the camera and the device casing it. To mitigate these errors, a blackbody calibration technique where a shutter whose thermal parameters are known is periodically used to calibrate the measurements. This technique, however, is only accurate when the shutter's temperature remains constant over time, which rarely is the case. In this paper, we contribute by developing a novel deep learning based calibration technique that uses battery temperature measurements to learn a model that allows adapting to changes in the internal thermal calibration parameters. Our method is particularly effective in continuous sensing where the device casing the camera is prone to heating. We demonstrate the effectiveness of our technique through controlled benchmark experiments which show significant improvements in thermal monitoring accuracy and robustness.

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.

Popularity of mobile apps is traditionally measured by metrics such as the number of downloads, installations, or user ratings. A problem with these measures is that they reflect usage only indirectly. Indeed, retention rates, i.e., the number of days users continue to interact with an installed app, have been suggested to predict successful app lifecycles. We conduct the first independent and large-scale study of retention rates and usage trends on a dataset of app-usage data from a community of 339,842 users and more than 213,667 apps. Our analysis shows that, on average, applications lose 65% of their users in the first week, while very popular applications (top 100) lose only 35%. It also reveals, however, that many applications have more complex usage behaviour patterns due to seasonality, marketing, or other factors. To capture such effects, we develop a novel app-usage trend measure which provides instantaneous information about the popularity of an application. Analysis of our data using this trend filter shows that roughly 40% of all apps never gain more than a handful of users ( Marginal apps). Less than 0.1% of the remaining 60% are constantly popular ( Dominant apps), 1% have a quick drain of usage after an initial steep rise ( Expired apps), and 6% continuously rise in popularity ( Hot apps). From these, we can distinguish, for instance, trendsetters from copycat apps. We conclude by demonstrating that usage behaviour trend information can be used to develop better mobile app recommendations.

We contribute by developing GeoMatch as a novel, scalable, and efficient big-data pipeline for large-scale map matching on Apache Spark. GeoMatch improves existing spatial big data solutions by utilizing a novel spatial partitioning scheme inspired by Hilbert space-filling curves. Thanks to the partitioning scheme, GeoMatch can effectively balance operations across different processing units and achieve significant performance gains. We demonstrate the effectiveness of GeoMatch through rigorous and extensive benchmarks that consider data sets containing large-scale urban spatial data sets ranging from 166, 253 to 3.78 billion location measurements. Our results show over 17-fold performance improvements compared to previous works while achieving better processing accuracy than current solutions (97.48%).

While mobile apps have become an integral part of everyday life, little is known about the factors that govern their usage. Particularly the role of geographic and cultural factors has been understudied. This article contributes by carrying out a large-scale analysis of geographic, cultural, and demographic factors in mobile usage. We consider app usage gathered from 25,323 Android users from 44 countries and 54,776 apps in 55 categories, and demographics information collected through a user survey. Our analysis reveals significant differences in app category usage across countries and we show that these differences, to large degree, reflect geographic boundaries. We also demonstrate that country gives more information about application usage than any demographic, but that there also are geographic and socio-economic subgroups in the data. Finally, we demonstrate that app usage correlates with cultural values using the Value Survey Model of Hofstede as a reference of cross-cultural differences.

This demo presents MegaSense, an air pollution monitoring system for realizing low-cost, near real-time and high resolution spatio-temporal air pollution maps of urban areas. MegaSense involves a novel hierarchy of multi-vendor distributed air quality sensors, in which accurate sensors calibrate lower cost sensors. Current low-cost air quality sensors suffer from measurement drift and they have low accuracy. We address this significant open problem for dense urban areas by developing a calibration scheme that detects and automatically corrects drift. MegaSense integrates with the 5G cellular network and leverages mobile edge computing for sensor management and distributed pollution map creation. We demonstrate MegaSense with two sensor types, a state of the art air quality monitoring station and a low-cost sensor array, with calibration between the two to improve the accuracy of the low-cost device. Participants can interact with the sensors and see air quality changes in real-time, and observe the mechanism to mitigate sensor drift. Our re-calibration method minimizes the error for NO2 and O3 81% of the time (vs single calibration) and reduces the mean relative error by 25%-45%.

Computational offloading can improve user experience of mobile apps through improved responsiveness and reduced energy footprint. A fundamental challenge in offloading is to distinguish situations where offloading is beneficial from those where it is counterproductive. Currently, offloading decisions are predominantly based on profiling performed on individual devices. While significant gains have been shown in benchmarks, these gains rarely translate to real-world use due to the complexity of contexts and parameters that affect offloading. We contribute by proposing crowdsensed evidence traces as a novel mechanism for improving the performance of offloading systems. Instead of limiting to profiling individual devices, crowdsensing enables characterizing execution contexts across a community of users, providing better generalisation and coverage of contexts. We demonstrate the feasibility of using crowdsensing to characterize offloading contexts through an analysis of two crowdsensing datasets. Motivated by our results, we present the design and development of the EMCO toolkit and platform as a novel solution for computational offloading. Experiments carried out on a testbed deployment in Amazon EC2 Ireland demonstrate that EMCO can consistently accelerate app execution while at the same time reduce energy footprint. We also demonstrate that EMCO provides better scalability than current cloud platforms, being able to serve a larger number of clients without variations in performance. Ourframework, use cases, and tools are available as open source from GitHub.

The quality of service of a mobile application is critical to ensure user satisfaction. Techniques have been proposed to accomplish adaptation of quality of service dynamically. However, there is still a limited understanding about how to provide a utility model for code execution. One key challenge is modeling the level of quality in the code execution that can be provisioned by the cloud. Since the allocation of cloud resources has a cost, it is important to optimize cloud usage. We propose a software-defined networking approach that allows modeling and controlling code acceleration of a mobile application deployed across multiple type of devices. By segregating the computational requirements of the mobile application into groups, we were able to define the acceleration needed by each group of devices. As the computational requirements of a device can change across time, a mobile device can be re-assigned to another group based on demand. Our SDN approach implements a model that allows the system to predict workload based on acceleration groups. Evaluating our system in a real testbed showed that it is possible to predict workload and allocate optimal resources to handle that workload with 87.5% accuracy.

Many applications of machine learning, for example in health care, would benefit from methods that can guarantee privacy of data subjects. Differential privacy (DP) has become established as a standard for protecting learning results. The standard DP algorithms require a single trusted party to have access to the entire data, which is a clear weakness, or add prohibitive amounts of noise. We consider DP Bayesian learning in a distributed setting, where each party only holds a single sample or a few samples of the data. We propose a learning strategy based on a secure multi-party sum function for aggregating summaries from data holders and the Gaussian mechanism for DP. Our method builds on an asymptotically optimal and practically efficient DP Bayesian inference with rapidly diminishing extra cost.

The Carat project started in 2012 has collected over 1.5 TB of data from over 850,000 mobile users all over the world. The project uses Apache Thrift to transmit data, and Apache Spark to run data analysis tasks, and the gist of the Carat analysis method has been published. While the Carat application code is open source, the data is much harder to share because of its size and privacy concerns. This paper outlines the challenges in sharing such a large-scale dataset with detailed information about smart devices, applications, and their users, and presents some solutions to these challenges.

Minimizing the energy consumption of mobile devices for wireless network access is important. In this article, we analyze the energy efficiency of a new set of applications which use Virtual Private Network (VPN) tunnels for secure communication. First, we discuss the energy efficiency of a number of VPN applications from a large scale deployment of 500 K devices. We next measure the energy consumption of some of these applications with different use cases. Finally, we demonstrate that a VPN tunnel can be instrumented for enhanced energy efficiency with multimedia streaming applications. Our results indicate energy savings of 40% for this class of applications.

The question “Where has my battery life gone?” remains a common source of frustration for many smartphone users. With the increased complexity of smartphone applications, and the increasing number of system settings affecting them, understanding and optimizing battery use has become a difficult chore. The present paper develops a novel approach for constructing energy models from crowdsourced measurements. In contrast to previous approaches, which have focused on the effect of a specific sensor, system setting or application, our approach can simultaneously capture relationships between multiple factors, and provide a unified view of the energy state of the mobile device. We demonstrate the validity of using crowdsourced measurements for constructing battery models through a combination of large-scale analysis of a dataset containing battery discharge and system state measurements and hardware power measurements. The results indicate that the models captured by our approach are both in line with previous studies on battery consumption and empirical measurements, providing a cost-effective way to construct energy models during normal operations of the device. The analysis also provides several new insights about battery consumption. For example, our analysis shows the energy use of high CPU activity with automatic screen brightness is actually higher (resulting in around 9 minutes shorter battery lifetime on average) than with a medium CPU load and manual screen brightness; a Wi-Fi signal strength drop of one bar can result in a battery life loss of over 13%; and a smartphone sitting in the sun can experience over 50% worse battery life than one indoors in cool conditions.

Mobile devices have limited battery life, and numerous battery management applications are available that aim to improve it. This paper examines a large-scale mobile battery awareness application, called Carat, to see how it changes user behavior with long-term use. We conducted a survey of current Carat Android users and analyzed their interaction logs. The results show that long-term Carat users save more battery, charge their devices less often, learn to manage their battery with less help from Carat, have a better understanding of how Carat works, and may enjoy competing against other users. Based on these findings, we propose a set of guidelines for mobile battery awareness applications: battery awareness applications should make the reasoning behind their recommendations understandable to the user, be tailored to retain long-term users, take the audience into account when formulating feedback, and distinguish third-party and system applications.

There is little information from independent sources in the public domain about mobile malware infection rates. The only previous independent estimate (0.0009%) [11], was based on indirect measurements obtained from domain-name resolution traces. In this paper, we present the first independent study of malware infection rates and associated risk factors using data collected directly from over 55,000 Android devices. We find that the malware infection rates in Android devices estimated using two malware datasets (0.28% and 0.26%), though small, are significantly higher than the previous independent estimate. Based on the hypothesis that some application stores have a greater density of malicious applications and that advertising within applications and cross-promotional deals may act as infection vectors, we investigate whether the set of applications used on a device can serve as an indicator for infection of that device. Our analysis indicates that, while not an accurate indicator of infection by itself, the application set does serve as an inexpensive method for identifying the pool of devices on which more expensive monitoring and analysis mechanisms should be deployed. Using our two malware datasets we show that this indicator performs up to about five times better at identifying infected devices than the baseline of random checks. Such indicators can be used, for example, in the search for new or previously undetected malware. It is therefore a technique that can complement standard malware scanning. Our analysis also demonstrates a marginally significant difference in battery use between infected and clean devices.

We aim to detect and diagnose energy anomalies, abnormally heavy battery use. This paper describes a collaborative black-box method, and an implementation called Carat, for diagnosing anomalies on mobile devices. A client app sends intermittent, coarse-grained measurements to a server, which correlates higher expected energy use with client properties like the running apps, device model, and operating system. The analysis quantifies the error and confidence associated with a diagnosis, suggests actions the user could take to improve battery life, and projects the amount of improvement. During a deployment to a community of more than 500,000 devices, Carat diagnosed thousands of energy anomalies in the wild. Carat detected all synthetically injected anomalies, produced no known instances of false positives, projected the battery impact of anomalies with 95% accuracy, and, on average, increased a user's battery life by 11% after 10 days (compared with 1.9% for the control group).

Current smartphones have a storage capacity of several gigabytes. More and more information is stored on mobile devices. To meet the challenge of information organization, we turn to desktop search. Users often possess multiple devices, and synchronize (subsets of) information between them. This makes file synchronization more important. This paper presents Dessy, a desktop search and synchronization framework for mobile devices. Dessy supports synchronization of search results, individual files, and directory trees. It allows finding and synchronizing files that reside on remote computers, or the Internet. The contributions of this paper include an energy usage evaluation of the system. Dessy is closely integrated with the Syxaw file synchronizer, which provides efficient file and metadata synchronization, optimizing network usage.

The storage capacity of smartphones has reached tens of gigabytes, while the search functionality remains simple. We have designed a search and synchronization framework for mobile devices, called Dessy. Dessy has been designed with mobility and device constraints in mind. It requires only MIDP 2.0 Mobile Java with File Connection support, and Java 1.5 on desktop machines. This paper demonstrates the application in practice, using multiple devices and synchronizing files between a desktop computer, a laptop, smartphones, and the Internet. Smartphones and laptops are able to search for files hosted on other devices as well as on the Internet. Unnecessary search operations are avoided using Bloom filters.

In this paper we describe modifications to a natural language grocery retrieval system, introduced in our earlier work. We also compare our system against an off-the-shelf retrieval tool, and show that our system is significantly better for top-ranked retrieval results.

We introduce a grocery retrieval system that maps shopping lists written in natural language into actual products in a grocery store. We have developed the system using nine months of shopping basket data from a large Finnish supermarket. To evaluate the system, we used 70 real shopping lists gathered from customers of the supermarket. Our system achieves over 80% precision for products at rank one, and the precision is around 70% for products at rank 5.