Abstract Clustering clients into groups with relatively homogeneous data distributions is a key strategy for improving federated learning under non-independent and identically distributed data. However, most state-of-the-art clustering approaches require clients to possess labeled datasets and perform substantial local computation, limiting their applicability in real-world settings. To address these limitations, we introduce CoLEDS , a method for profiling unlabeled client datasets with minimal computational overhead. CoLEDS trains a model using a contrastive learning objective defined across multiple clients and optimized in a distributed fashion through joint client–server coordination. The resulting model embeds key properties of client datasets into low-dimensional vectors that are shared with the server for clustering. Extensive empirical evaluation shows that these profiles accurately capture latent dataset characteristics. By clustering clients based on these representations, CoLEDS yields federatively trained models that are better aligned with individual data distributions and enables appropriate model assignment even for clients that do not participate in federated training.

Abstract Deep learning (DL) powers numerous applications on ubiquitous edge devices, but its high resource demands pose a challenge. Approximate computing is often proposed to alleviate this, yet such calculation usually suffers from a fixed level of accuracy loss. We propose a novel control-theoretic approach for predictive DL inference on resource constrained devices. Our system dynamically adjusts approximation levels based on a trade-off between resource utilisation and accuracy, considering future demands. Extensive experiments across diverse domains - human activity recognition, acoustic scene profiling, and computer vision - with various neural network architectures and approximation techniques, demonstrate that our approach achieves up to 50% energy savings while maintaining the desired inference accuracy and incurring minimal runtime overhead. Furthermore, we showcase our method in a real-world deployment on low-power edge devices and confirm its superiority over current state-of-the-art solutions.

Progressing beyond centralized AI is of paramount importance, yet, distributed AI solutions, in particular various federated learning (FL) algorithms, are often not comprehensively assessed, which prevents the research community from identifying the most promising approaches and practitioners from being convinced that a certain solution is deployment-ready. The largest hurdle towards FL algorithm evaluation is the difficulty of conducting real-world experiments over a variety of FL client devices and different platforms, with different datasets and data distribution, all while assessing various dimensions of algorithm performance, such as inference accuracy, energy consumption, and time to convergence, to name a few. In this paper, we present CoLExT, a real-world testbed for FL research. CoLExT is designed to streamline experimentation with custom FL algorithms in a rich testbed configuration space, with a large number of heterogeneous edge devices, ranging from single-board computers to smartphones, and provides real-time collection and visualization of a variety of metrics through automatic instrumentation. According to our evaluation, porting FL algorithms to CoLExT requires minimal involvement from the developer, and the instrumentation introduces minimal resource usage overhead. Furthermore, through an initial investigation involving popular FL algorithms running on CoLExT, we reveal previously unknown trade-offs, inefficiencies, and programming bugs.

Runtime-tunable context-dependent network compression would make mobile deep learning (DL) adaptable to often varying resource availability, input “difficulty”, or user needs. The existing compression techniques significantly reduce the memory, processing, and energy tax of DL, yet, the resulting models tend to be permanently impaired, sacrificing the inference power for reduced resource usage. The existing tunable compression approaches, on the other hand, require expensive re-training, do not support arbitrary strategies for adapting the compression and do not provide mobile-ready implementations. In this paper we present Mobiprox, a framework enabling mobile DL with flexible precision. Mobiprox implements tunable approximations of tensor operations and enables runtime-adaptable approximation of the individual network layers. A profiler and a tuner included with Mobiprox identify the most promising neural network approximation configurations leading to the desired inference quality with the minimal use of resources. Furthermore, we develop control strategies that depending on contextual factors, such as the input data difficulty, dynamically adjust the approximation levels across a mobile DL model’s layers. We implement Mobiprox in Android OS and through experiments in diverse mobile domains, including human activity recognition and spoken keyword detection, demonstrate that it can save up to 15% system-wide energy with a minimal impact on the inference accuracy.

While the evolution of mobile computing is experiencing considerable growth, it is at the same time seriously threatened by the limitations of battery technology, which does not keep pace with the evergrowing increase in energy requirements of mobile applications. Yet, with the limits of human perception and the diversity of requirements that individuals may have, a question arises of whether the effort should be made to always deliver the highest quality result to a mobile user? In this work we investigate how a user's physical activity, the spatial/temporal properties of the video, and the user's personality traits interact and jointly influence the minimal acceptable playback resolution. We conduct two studies with 45 participants in total and find out that the minimal acceptable resolution indeed varies across different contextual factors. Our predictive models inferring the lowest acceptable playback resolution, together with the reduced power consumption we measure at lower resolutions, open an opportunity for saving a mobile's energy through context-adaptable approximate computing.

Billions of sensor-enabled computing devices open tremendous opportunities for AI-powered context-aware services. Yet, democratizing AI so that heterogeneous devices can enjoy deep learning (DL) revolution requires significant reduction in computing and energy burden posed by DL models. Inspired by the concept of compressive sensing (CS), and guided by the observation that reduced sampling rates often suffice for successful classification, we devise an adaptive CS-DL pipeline. Our approach dynamically adjusts the sensing rate according to input properties and performs classification through an input-flexible DL model, demonstrating classification accuracy rates on par with uncompressed models while using up to 46% less battery energy.

Multimodal data harvested by the Internet of Things sensors has recently been utilised for behavioural biometrics and consequently user authentication. While strengthening the security, these data nevertheless present a privacy threat to users whose behaviour can now be modelled in detail, thus allowing the authenticating authority to not only know who is present, but also what the present person is doing. In this work we reconsider IoT sensor-based authentication and provide a solution mitigating the privacy risk associated with unnecessary information leaks. Our approach harnesses adversarial learning and identifies such a projection of the data that maintains identity separability, yet obfuscates activity separability, thus ensuring that the authenticating authority can successfully identify the user, but not her actions within the sensed environment. We evaluate our approach on a real-world dataset of three activities performed by fifteen users and show that the activity obfuscation is achieved without compromising identification capabilities.

The smartphone has revolutionised the way we receive news, enabling on-demand, personalised content to be viewed in a range of different situations. Yet, while the content of the news is often adapted to the user’s preferences and the current environment (e.g. location), the actual interface of a mobile newsreader app often remains the same across users and contexts of use. In this work we first collect and examine real-world mobile news reading data to uncover the way contextual factors affect the perception of different aspects of the newsreader app interface, and then develop a method for modelling personalised context-dependent viewing preferences. Through a four-week long user study we demonstrate that our reinforcement and active learning-based personalisation approach leads to 26% higher user acceptance as compared to a generic context-aware mobile newsreader interface adaptation model.

Untimely interaction in critical moments, content that does not capture a user’s attention, and missed opportunities for delivering relevant information are just some of the issues caused by ubiquitous computing devices’ inability to understand their users cognitive engagement. However, current means of inferring cognitive load are, with a few exceptions, based on intrusive methods requiring physical contact of the measurement equipment and the user, thus of limited applicability in ubiquitous computing. In this paper we propose Wi-Mind, a system for remote cognitive load assessment through wireless sensing. Wi-Mind is based on a software-defined radio radar that measures sub-millimeter movements related to a person’s breathing and heart beats, which, in turn allows us to infer the person’s cognitive load. We built and tested the system with 23 volunteers engaged in different tasks. Initial results show that while Wi-Mind manages to detect whether one is engaged in a cognitively demanding task, the inference of the exact cognitive load level remains challenging.

Ubiquitous computing is moving from context-awareness to context-prediction. In order to build truly anticipatory systems developers have to deal with many challenges, from multimodal sensing to modeling context from sensed data, and, when necessary, coordinating multiple predictive models across devices. Novel expressive programming interfaces and paradigms are needed for this new class of mobile and ubiquitous applications. In this paper we present FutureWare, a middleware for seamless development of mobile applications that rely on context prediction. FutureWare exposes an expressive API to lift the burden of mobile sensing, individual and group behavior modeling, and future context querying, from an application developer. We implement FutureWare as an Android library, and through a scenario-based testing and a demo app we show that it represents an efficient way of supporting anticipatory applications, reducing the necessary coding effort by two orders of magnitude.

Mobile computing evolution is critically threatened by the limitations of the battery technology, which does not keep pace with the increase in energy requirements of mobile applications. A novel approach for reducing the energy appetite of mobile apps comes from the approximate Computing field, which proposes techniques that in a controlled manner sacrifice computation accuracy for higher energy savings. Building on this philosophy we propose a context-aware mobile video quality adaptation that reduces the energy needed for video playback, while ensuring that a user’s quality expectations with respect to the mobile video are met. We confirm that the decoding resolution can play a significant role in reducing the overall power consumption of a mobile device and conduct a user study with 22 participants to investigate how the context in which a video is played modulates a user’s quality expectations. We discover that a user’s physical activity and the spatial/temporal properties of the video interact and jointly influence the minimal acceptable playback resolution, paving the way for context-adaptable approximate mobile computing.

Extreme citizen science is a bottom up practice used to empower people by supporting them, via processes and technological tools, to find solutions for local problems, but also to tackle major sustainability challenges of the 21st century. Methods and tools based on mobile computing have been utilized by communities in various parts of the world, from the Congo Basin through the Amazonian rain forest. However, extreme citizen science initiatives often face severe challenges as pre-designed technological solutions prove to be non-transferable to peculiar environments of rural developing regions. In this paper we collect and investigate evidence from the implementation of various extreme citizen science initiatives in the developing world. Our aim is to identify key obstacles towards their successful realization, mainly focusing on the problem of user interaction with mobile computing solutions. We conduct interviews with nine experienced researchers who all performed extensive fieldwork within these initiatives, and who reflect on the technology interaction, knowledge organization, inter-cultural, social and usability issues. Based on our analysis we report among others, symptomatic difficulties with abstractions, representational hierarchies, and navigation commands, as well as potential improvements that mobile technology developers can implement in order to create a more inclusive environment for extreme citizen science.

An unexpected increase in natural disasters has prompted a large interest in governments and organisations to utilise ICT for many different purposes such as preparation, impact mitigation, loss reduction and relief efforts. This paper presents initial work on studying disaster scenarios from device level perspective to characterise network infrastructural behaviour during extraordinary situations. We find connectivity challenges during disasters and observe sharp decline of quality metrics and loss of station quantity between ordinary and extraordinary time periods. We also make distinctions between usual and unusual behaviour seen during ordinary and extraordinary situations.

Notifications are extremely beneficial to users, but they often demand their attention at inappropriate moments. In this paper we present an in-situ study of mobile interruptibility focusing on the effect of cognitive and physical factors on the response time and the disruption perceived from a notification. Through a mixed method of automated smartphone logging and experience sampling we collected 10372 in-the-wild notifications and 474 questionnaire responses on notification perception from 20 users. We found that the response time and the perceived disruption from a notification can be influenced by its presentation, alert type, sender-recipient relationship as well as the type, completion level and complexity of the task in which the user is engaged. We found that even a notification that contains important or useful content can cause disruption. Finally, we observe the substantial role of the psychological traits of the individuals on the response time and the disruption perceived from a notification.

An increasing number of notifications demanding the smartphone user's attention, often arrive at an inappropriate moment, or carry irrelevant content. In this paper we present a study of mobile user interruptibility with respect to notification content, its sender, and the context in which a notification is received. In a real-world study we collect around 70,000 instances of notifications from 35 users. We group notifications according to the applications that initiated them, and the social relationship between the sender and the receiver. Then, by considering both content and context information, such as the current activity of a user, we discuss the design of classifiers for learning the most opportune moment for the delivery of a notification carrying a specific type of information. Our results show that such classifiers lead to a more accurate prediction of users' interruptibility than an alternative approach based on user-defined rules of their own interruptibility.

The mobile phone represents a unique platform for interactive applications that can harness the opportunity of an immediate contact with a user in order to increase the impact of the delivered information. However, this accessibility does not necessarily translate to reachability, as recipients might refuse an initiated contact or disfavor a message that comes in an inappropriate moment.

Smartphone sensing enables inference of physical context, while online social networks (OSNs) allow mobile applications to harness users' interpersonal relationships. However, OSNs and smartphone sensing remain disconnected, since obstacles, including the synchronization of mobile sensing and OSN monitoring, inefficiency of smartphone sensors, and privacy concerns, stand in the way of merging the information from these two sources.

The increased demand for wireless connectivity emphasizes the necessity of efficient wireless communication as resources such as the available spectrum and energy reserves become limiting factors for network proliferation. Recent advancements in software-defined radio enable high flexibility of the physical layer allowing fine grained transmission adjustments. Although communication efficiency can greatly benefit from physical layer flexibility, modern wireless protocols can neither handle these new opportunities nor allocate resources according to the overlying application needs. In this work we develop WhiteRate, a method for physical layer parameter adaptation that efficiently utilizes available energy and spectrum resources, while maintaining the desired quality of communication. Our solution adjusts the modulation and coding scheme, and channel width to achieve a communication profile that matches application requirements. We implement WhiteRate in GNUradio and evaluate it in both indoor and outdoor environments. We demonstrate improvements on two important fronts: spectrum utilization and energy efficiency. Moreover, we show that by using WhiteRate, both benefits can be achieved simultaneously.

Mobile telephony brings clear economic and social benefits to its users. As handsets have become more affordable, ownership has reached staggering numbers, even in the most remote areas of the world. However, network coverage is often lacking in low population densities and low income rural areas of the developing world, where big telecoms often defer from deploying expensive infrastructure. To solve this coverage gap, we propose VillageCell, a low-cost alternative to high-end cell phone networks. VillageCell relies on software defined radios and open-source solutions to provide free local and cheap long-distance communication for remote regions. Our architecture is simple and easy to deploy, yet robust and requires no modification to GSM handsets. Through measuring the call quality metrics and the system capacity under a realistic rural-area network load, we show that VillageCell is indeed an attractive solution for rural area voice connectivity.

Recent advancements in wireless transmission have enabled networks with a high level of physical layer flexibility. Unfortunately, these new opportunities are not harnessed by modern wireless systems. Due to inefficient resource allocation, systems typically encounter problems such as spectrum scarcity, energy depletion or low quality of service. In this paper we consider the problem of physical layer parameter adaptation in a flexible wireless system. We observe that for many practical purposes the acceptable quality of communication depends on the interplay among the packet loss ratio, energy savings and spectrum utilization. We harness this fact and propose a context-aware physical layer parameter adaptation solution, WhiteRate. Our solution adjusts the modulation level, coding scheme and channel width to achieve the communication profile that matches application requirements. We implement WhiteRate in GNUradio and evaluate it in both indoor and outdoor environments. We demonstrate improvements on two important fronts: spectrum utilization and energy efficiency. Moreover, we show that by using WhiteRate, both benefits can be achieved simultaneously.

Online Social Networks (OSN) are fun, popular, and socially significant. An integral part of their success is the immense size of their global user base. To provide a consistent service to all users, Facebook, the world’s largest OSN, is heavily dependent on centralized U.S. data centers, which renders service outside of the U.S. sluggish and wasteful of Internet bandwidth. In this paper, we investigate the detailed causes of these two problems and identify mitigation opportunities. Because details of Facebook’s service remain proprietary, we treat the OSN as a black box and reverse engineer its operation from publicly available traces. We find that contrary to current wisdom, OSN state is amenable to partitioning and that its fine grained distribution and processing can significantly improve performance without loss in service consistency. Through simulations of reconstructed Facebook traffic over measured Internet paths, we show that user requests can be processed 79% faster and use 91% less bandwidth. We conclude that the partitioning of OSN state is an attractive scaling strategy for Facebook and other OSN services.