On-device machine learning (ML) inference can enable the use of private user data on user devices without revealing them to remote servers. However, a pure on-device solution to private ML inference is impractical for many applications that rely on embedding tables that are too large to be stored on-device. In particular, recommendation models typically use multiple embedding tables each on the order of 1--10 GBs of data, making them impractical to store on-device. To overcome this barrier, we propose the use of private information retrieval (PIR) to efficiently and privately retrieve embeddings from servers without sharing any private information. As off-the-shelf PIR algorithms are usually too computationally intensive to directly use for latency-sensitive inference tasks, we 1) propose novel GPU-based acceleration of PIR, and 2) co-design PIR with the downstream ML application to obtain further speedup. Our GPU acceleration strategy improves system throughput by more than 20× over an optimized CPU PIR implementation, and our PIR-ML co-design provides an over 5× additional throughput improvement at fixed model quality. Together, for various on-device ML applications such as recommendation and language modeling, our system on a single V100 GPU can serve up to 100,000 queries per second---a > 100× throughput improvement over a CPU-based baseline---while maintaining model accuracy.

Recently, there has been an explosive growth of mobile and embedded applications using convolutional neural networks (CNNs). To alleviate their excessive computational demands, developers have traditionally resorted to cloud offloading, inducing high infrastructure costs and a strong dependence on networking conditions. On the other end, the emergence of powerful SoCs is gradually enabling on-device execution. Nonetheless, low- and mid-tier platforms still struggle to run state-of-the-art CNNs sufficiently. In this article, we present DynO, a distributed inference framework that combines the best of both worlds to address several challenges, such as device heterogeneity, varying bandwidth, and multi-objective requirements. Key components that enable this are its novel CNN-specific data packing method, which exploits the variability of precision needs in different parts of the CNN when onloading computation, and its novel scheduler, which jointly tunes the partition point and transferred data precision at runtime to adapt inference to its execution environment. Quantitative evaluation shows that DynO outperforms the current state of the art, improving throughput by over an order of magnitude over device-only execution and up to 7.9× over competing CNN offloading systems, with up to 60× less data transferred.

The unprecedented performance of deep neural networks (DNNs) has led to large strides in various Artificial Intelligence (AI) inference tasks, such as object and speech recognition. Nevertheless, deploying such AI models across commodity devices faces significant challenges: large computational cost, multiple performance objectives, hardware heterogeneity and a common need for high accuracy, together pose critical problems to the deployment of DNNs across the various embedded and mobile devices in the wild. As such, we have yet to witness the mainstream usage of state-of-the-art deep learning algorithms across consumer devices. In this paper, we provide preliminary answers to this potentially game-changing question by presenting an array of design techniques for efficient AI systems. We start by examining the major roadblocks when targeting both programmable processors and custom accelerators. Then, we present diverse methods for achieving real-time performance following a cross-stack approach. These span model-, system- and hardware-level techniques, and their combination. Our findings provide illustrative examples of AI systems that do not overburden mobile hardware, while also indicating how they can improve inference accuracy. Moreover, we showcase how custom ASIC- and FPGA-based accelerators can be an enabling factor for next-generation AI applications, such as multi-DNN systems. Collectively, these results highlight the critical need for further exploration as to how the various cross-stack solutions can be best combined in order to bring the latest advances in deep learning close to users, in a robust and efficient manner.

With smartphones' omnipresence in people's pockets, Machine Learning (ML) on mobile is gaining traction as devices become more powerful. With applications ranging from visual filters to voice assistants, intelligence on mobile comes in many forms and facets. However, Deep Neural Network (DNN) inference remains a compute intensive workload, with devices struggling to support intelligence at the cost of responsiveness. On the one hand, there is significant research on reducing model runtime requirements and supporting deployment on embedded devices. On the other hand, the strive to maximise the accuracy of a task is supported by deeper and wider neural networks, making mobile deployment of state-of-the-art DNNs a moving target.

Federated Learning (FL) has been gaining significant traction across different ML tasks, ranging from vision to keyboard predictions. In large-scale deployments, client heterogeneity is a fact and constitutes a primary problem for fairness, training performance and accuracy. Although significant efforts have been made into tackling statistical data heterogeneity, the diversity in the processing capabilities and network bandwidth of clients, termed system heterogeneity, has remained largely unexplored. Current solutions either disregard a large portion of available devices or set a uniform limit on the model's capacity, restricted by the least capable participants.In this work, we introduce Ordered Dropout, a mechanism that achieves an ordered, nested representation of knowledge in Neural Networks and enables the extraction of lower footprint submodels without the need for retraining. We further show that for linear maps our Ordered Dropout is equivalent to SVD. We employ this technique, along with a self-distillation methodology, in the realm of FL in a framework called FjORD. FjORD alleviates the problem of client system heterogeneity by tailoring the model width to the client's capabilities. Extensive evaluation on both CNNs and RNNs across diverse modalities shows that FjORD consistently leads to significant performance gains over state-of-the-art baselines while maintaining its nested structure.

Attention networks constitute the state-of-the-art paradigm for capturing long temporal dynamics. This paper examines the efficacy of this paradigm in the challenging task of emotion recognition in dyadic conversations. In this work, we introduce a novel attention mechanism capable of inferring the immensity of the effect of each past utterance on the current speaker emotional state. The proposed self-attention network captures the correlation patterns among consecutive encoder network states, thus enabling the robust and effective modeling of temporal dynamics over arbitrary long temporal horizons. We exhibit the effectiveness of our approach considering the challenging IEMOCAP benchmark. We show that, our devised methodology outperforms state-of-the-art alternatives and commonly used approaches, giving rise to promising new research directions in the context of Online Social Network (OSN) analysis tasks.

Despite the soaring use of convolutional neural networks (CNNs) in mobile applications, uniformly sustaining high-performance inference on mobile has been elusive due to the excessive computational demands of modern CNNs and the increasing diversity of deployed devices. A popular alternative comprises offloading CNN processing to powerful cloud-based servers. Nevertheless, by relying on the cloud to produce outputs, emerging mission-critical and high-mobility applications, such as drone obstacle avoidance or interactive applications, can suffer from the dynamic connectivity conditions and the uncertain availability of the cloud. In this paper, we propose SPINN, a distributed inference system that employs synergistic device-cloud computation together with a progressive inference method to deliver fast and robust CNN inference across diverse settings. The proposed system introduces a novel scheduler that co-optimises the early-exit policy and the CNN splitting at run time, in order to adapt to dynamic conditions and meet user-defined service-level requirements. Quantitative evaluation illustrates that SPINN outperforms its state-of-the-art collaborative inference counterparts by up to 2× in achieved throughput under varying network conditions, reduces the server cost by up to 6.8× and improves accuracy by 20.7% under latency constraints, while providing robust operation under uncertain connectivity conditions and significant energy savings compared to cloud-centric execution.

The Internet Para Todos program is working to provide sustainable mobile broadband to 100 M unconnected people in Latin America. In this paper we present our commercial deployment in thousands remote small communities and describe the unique experience of maintaining this infrastructure. We describe the challenges related to managing operations containing the cost in these extreme geographical conditions. We also analyze operational data to understand outage patterns and present typical operational issues in this unique remote community environment. Finally, we present an extension of the operations support system (OSS) leveraging advanced analytics and machine learning with the goal of optimizing network maintenance while reducing costs.

We present DarkneTZ, a framework that uses an edge device's Trusted Execution Environment (TEE) in conjunction with model partitioning to limit the attack surface against Deep Neural Networks (DNNs). Increasingly, edge devices (smartphones and consumer IoT devices) are equipped with pre-trained DNNs for a variety of applications. This trend comes with privacy risks as models can leak information about their training data through effective membership inference attacks (MIAs).

Video sharing platforms like YouTube are increasingly targeted by aggression and hate attacks. Prior work has shown how these attacks often take place as a result of "raids," i.e., organized efforts by ad-hoc mobs coordinating from third-party communities. Despite the increasing relevance of this phenomenon, however, online services often lack effective countermeasures to mitigate it. Unlike well-studied problems like spam and phishing, coordinated aggressive behavior both targets and is perpetrated by humans, making defense mechanisms that look for automated activity unsuitable. Therefore, the de-facto solution is to reactively rely on user reports and human moderation. In this paper, we propose an automated solution to identify YouTube videos that are likely to be targeted by coordinated harassers from fringe communities like 4chan. First, we characterize and model YouTube videos along several axes (metadata, audio transcripts, thumbnails) based on a ground truth dataset of videos that were targeted by raids. Then, we use an ensemble of classifiers to determine the likelihood that a video will be raided with very good results (AUC up to 94%). Overall, our work provides an important first step towards deploying proactive systems to detect and mitigate coordinated hate attacks on platforms like YouTube.

Cyberbullying and cyberaggression are increasingly worrisome phenomena affecting people across all demographics. More than half of young social media users worldwide have been exposed to such prolonged and/or coordinated digital harassment. Victims can experience a wide range of emotions, with negative consequences such as embarrassment, depression, isolation from other community members, which embed the risk to lead to even more critical consequences, such as suicide attempts. In this work, we take the first concrete steps to understand the characteristics of abusive behavior in Twitter, one of today’s largest social media platforms. We analyze 1.2 million users and 2.1 million tweets, comparing users participating in discussions around seemingly normal topics like the NBA, to those more likely to be hate-related, such as the Gamergate controversy, or the gender pay inequality at the BBC station. We also explore specific manifestations of abusive behavior, i.e., cyberbullying and cyberaggression, in one of the hate-related communities (Gamergate). We present a robust methodology to distinguish bullies and aggressors from normal Twitter users by considering text, user, and network-based attributes. Using various state-of-the-art machine-learning algorithms, we classify these accounts with over 90% accuracy and AUC. Finally, we discuss the current status of Twitter user accounts marked as abusive by our methodology and study the performance of potential mechanisms that can be used by Twitter to suspend users in the future.

Botnets in online social networks are increasingly often affecting the regular flow of discussion, attacking regular users and their posts, spamming them with irrelevant or offensive content, and even manipulating the popularity of messages and accounts. Researchers and cybercriminals are involved in an arms race, and new and updated botnets designed to defeat current detection systems are constantly developed, rendering such detection systems obsolete.

With current technology, a number of entities have access to user mobility traces at different levels of spatio-temporal granularity. At the same time, users frequently reveal their location through different means, including geo-tagged social media posts and mobile app usage. Such leaks are often bound to a pseudonym or a fake identity in an attempt to preserve one's privacy. In this work, we investigate how large-scale mobility traces can de-anonymize anonymous location leaks. By mining the country-wide mobility traces of tens of millions of users, we aim to understand how many location leaks are required to uniquely match a trace, how spatio-temporal obfuscation decreases the matching quality, and how the location popularity and time of the leak influence de-anonymization. We also study the mobility characteristics of those individuals whose anonymous leaks are more prone to identification. Finally, by extending our matching methodology to full traces, we show how large-scale human mobility is highly unique. Our quantitative results have implications for the privacy of users' traces, and may serve as a guideline for future policies regarding the management and publication of mobility data.

While developing mobile apps is becoming easier, testing and characterizing their behavior is still hard. On the one hand, the de facto testing tool, called "Monkey," scales well due to being based on random inputs, but fails to gather inputs useful in understanding things like user engagement and attention. On the other hand, gathering inputs and data from real users requires distributing instrumented apps, or even phones with pre-installed apps, an expensive and inherently unscaleable task. To address these limitations we present CHIMP, a system that integrates automated tools and large-scale crowdsourced inputs. CHIMP is different from previous approaches in that it runs apps in a virtualized mobile environment that thousands of users all over the world can access via a standard Web browser. CHIMP is thus able to gather the full range of real-user inputs, detailed run-time traces of apps, and network traffic. We thus describe CHIMP»s design and demonstrate the efficiency of our approach by testing thousands of apps via thousands of crowdsourced users. We calibrate CHIMP with a large-scale campaign to understand how users approach app testing tasks. Finally, we show how CHIMP can be used to improve both traditional app testing tasks, as well as more novel tasks such as building a traffic classifier on encrypted network flows.

Mobile network operators collect a humongous amount of network measurements. Among those, sector Key Performance Indicators (KPIs) are used to monitor the radio access, i.e., the "last mile" of mobile networks. Thresholding mechanisms and synthetic combinations of KPIs are used to assess the network health, and rank sectors to identify the underperforming ones. It follows that the available monitoring methodologies heavily rely on the fine grained tuning of thresholds and weights, currently established through domain knowledge of both vendors and operators. In this paper, we study how to bridge sector KPIs to reflect Quality of Experience (QoE) groundtruth measurements, namely throughput, latency and video streaming stall events. We leverage one month of data collected in the operational network of mobile network operator serving more than 10 million subscribers. We extensively investigate up to which extent adopted methodologies efficiently capture QoE. Moreover, we challenge the current state of the art by presenting data-driven approaches based on Particle Swarm Optimization (PSO) metaheuristics and random forest regression algorithms, to better assess sector performance. Results show that the proposed methodologies outperforms state of the art solution improving the correlation with respect to the baseline by a factor of 3, and improving visibility on underperforming sectors. Our work opens new areas for research in monitoring solutions for enriching the quality and accuracy of the network performance indicators collected at the network edge.

To manage and maintain large-scale cellular networks, operators need to know which sectors underperform at any given time. For this purpose, they use the so-called hot spot score, which is the result of a combination of multiple network measurements and reflects the instantaneous overall performance of individual sectors. While operators have a good understanding of the current performance of a network and its overall trend, forecasting the performance of each sector over time is a challenging task, as it is affected by both regular and non-regular events, triggered by human behavior and hardware failures. In this paper, we study the spatio-temporal patterns of the hot spot score and uncover its regularities. Based on our observations, we then explore the possibility to use recent measurements' history to predict future hot spots. To this end, we consider tree-based machine learning models, and study their performance as a function of time, amount of past data, and prediction horizon. Our results indicate that, compared to the best baseline, tree-based models can deliver up to 14% better forecasts for regular hot spots and 153% better forecasts for non-regular hot spots. The latter brings strong evidence that, for moderate horizons, forecasts can be made even for sectors exhibiting isolated, non-regular behavior. Overall, our work provides insight into the dynamics of cellular sectors and their predictability. It also paves the way for more proactive network operations with greater forecasting horizons.

As the number and the diversity of news outlets on the Web grows, so does the opportunity for "alternative" sources of information to emerge. Using large social networks like Twitter and Facebook, misleading, false, or agenda-driven information can quickly and seamlessly spread online, deceiving people or influencing their opinions. Also, the increased engagement of tightly knit communities, such as Reddit and 4chan, further compounds the problem, as their users initiate and propagate alternative information, not only within their own communities, but also to different ones as well as various social media. In fact, these platforms have become an important piece of the modern information ecosystem, which, thus far, has not been studied as a whole.

Video streaming on mobile devices is prone to a multitude of faults and although well established video Quality of Experience (QoE) metrics such as stall frequency are a good indicator of the problems perceived by the user, they do not provide any insights about the nature of the problem nor where it has occurred. Quantifying the correlation between the aforementioned faults and the users' experience is a challenging task due the large number of variables and the numerous points-of-failure.

A significant fraction of Internet traffic is now encrypted and HTTPS will likely be the default in HTTP/2. However, Transport Layer Security (TLS), the standard protocol for encryption in the Internet, assumes that all functionality resides at the endpoints, making it impossible to use in-network services that optimize network resource usage, improve user experience, and protect clients and servers from security threats. Re-introducing in-network functionality into TLS sessions today is done through hacks, often weakening overall security.

Through their normal operation, cellular networks are a repository of continuous location information from their subscribed devices. Such information, however, comes at a coarse granularity both in terms of space, as well as time. For otherwise inactive devices, location information can be obtained at the granularity of the associated cellular sector, and at infrequent points in time, that are sensitive to the structure of the network itself, and the level of mobility of the device. In this paper, we are asking the question of whether such sparse information can help to identify the paths followed by mobile connected devices throughout the day. If such a task is possible, then we would not only enable continuous mobility path estimation for smartphones, but also for the millions of future connected "things".

Increased user concern over security and privacy on the Internet has led to widespread adoption of HTTPS, the secure version of HTTP. HTTPS authenticates the communicating end points and provides confidentiality for the ensuing communication. However, as with any security solution, it does not come for free. HTTPS may introduce overhead in terms of infrastructure costs, communication latency, data usage, and energy consumption. Moreover, given the opaqueness of the encrypted communication, any in-network value added services requiring visibility into application layer content, such as caches and virus scanners, become ineffective.

In many work environments, serendipitous interactions between members of different groups may lead to enhanced productivity, collaboration and knowledge dissemination. Two factors that may have an influence on such interactions are cultural differences between individuals in highly multicultural workplaces, and the layout and physical spaces of the workplace itself. In this work, we investigate how these two factors may facilitate or hinder inter-group interactions in the workplace. We analyze traces collected using wearable electronic badges to capture face-to-face interactions and mobility patterns of employees in a research laboratory in the UK. We observe that those who interact with people of different roles tend to come from collectivist cultures that value relationships and where people tend to be comfortable with social hierarchies, and that some locations in particular are more likely to host serendipitous interactions, knowledge that could be used by organizations to enhance communication and productivity.

The layouts of the buildings we live in shape our everyday lives. In office environments, building spaces affect employees' communication, which is crucial for productivity and innovation. However, accurate measurement of how spatial layouts affect interactions is a major challenge and traditional techniques may not give an objective view.

Mobile phones play a pivotal role in supporting ubiquitous and unobtrusive sensing of human activities. However, maintaining a highly accurate record of a user's behavior throughout the day imposes significant energy demands on the phone's battery. In this paper, we present the design, implementation, and evaluation of METIS: an adaptive mobile sensing platform that efficiently supports social sensing applications. The platform implements a novel sensor task distribution scheme that dynamically decides whether to perform sensing on the phone or in the infrastructure, considering the energy consumption, accuracy, and mobility patterns of the user. By comparing the sensing distribution scheme with sensing performed solely on the phone or exclusively on the fixed remote sensors, we show, through benchmarks using real traces, that the opportunistic sensing distribution achieves over 60% and 40% energy savings, respectively. This is confirmed through a real world deployment in an office environment for over a month: we developed a social application over our frameworks, that is able to infer the collaborations and meetings of the users. In this setting the system preserves over 35% more battery life over pure phone sensing.

The application of complex network models to communication systems has led to several important results: nonetheless, previous research has often neglected to take into account their temporal properties, which in many real scenarios play a pivotal role. At the same time, network robustness has come extensively under scrutiny. Understanding whether networked systems can undergo structural damage and yet perform efficiently is crucial to both their protection against failures and to the design of new applications. In spite of this, it is still unclear what type of resilience we may expect in a network which continuously changes over time. In this work, we present the first attempt to define the concept of temporal network robustness: we describe a measure of network robustness for time-varying networks and we show how it performs on different classes of random models by means of analytical and numerical evaluation. Finally, we report a case study on a real-world scenario, an opportunistic vehicular system of about 500 taxicabs, highlighting the importance of time in the evaluation of robustness. Particularly, we show how static approximation can wrongly indicate high robustness of fragile networks when adopted in mobile time-varying networks, while a temporal approach captures more accurately the system performance.

Nowadays companies increasingly aggregate location data from different sources on the Internet to offer location-based services such as estimating current road traffic conditions, and finding the best nightlife locations in a city. However, these services have also caused outcries over privacy issues. As the volume of location data being aggregated expands, the comfort of sharing one's whereabouts with the public at large will unavoidably decrease. Existing ways of aggregating location data in the privacy literature are largely centralized in that they rely on a trusted location-based service. Instead, we propose a piece of software (SpotMe) that can run on a mobile phone and is able to estimate the number of people in geographic locations in a privacy-preserving way: accurate estimations are made possible in the presence of privacy-conscious users who report, in addition to their actual locations, a very large number of erroneous locations. The erroneous locations are selected by a randomized response algorithm. We evaluate the accuracy of SpotMe in estimating the number of people upon two very different realistic mobility traces: the mobility of vehicles in urban, suburban and rural areas, and the mobility of subway train passengers in Greater London. We find that erroneous locations have little effect on the estimations (in both traces, the error is below 18% for a situation in which more than 99% of the locations are erroneous), yet they guarantee that users cannot be localized with high probability. Also, the computational and storage overheads for a mobile phone running Spot Me are negligible, and the communication overhead is limited.

The application of complex network theory to communication systems has led to several important results. Nonetheless, previous research has often neglected to take into account their temporal properties, which in many real scenarios play a pivotal role. Mainly because of mobility, transmission delays or protocol design, a communication network should not be considered only as a static entity. At the same time, network robustness has come extensively under scrutiny. Understanding whether networked systems can undergo structural damage and yet perform efficiently is crucial to both their protection against failures and to the design of new applications. In spite of this, it is still unclear what type of resilience we may expect in a network that continuously changes over time. In this work we present the first attempt to define the concept of temporal network robustness: we describe a measure of network robustness for time-varying networks and we show how it performs on different classes of random models by means of analytical and numerical evaluation. Particularly, we show how static approximation can wrongly indicate high robustness of fragile networks when adopted in mobile time-varying networks, while a temporal approach captures more accurately the system performance.

Road congestion results in a huge waste of time and productivity for millions of people. A possible way to deal with this problem is to have transportation authorities distribute traffic information to drivers, which, in turn, can decide (or be aided by a navigator) to route around congested areas. Such traffic information can be gathered by relying on static sensors placed at specific road locations (e.g., induction loops and video cameras) or by having single vehicles report their location, speed, and travel time. While the former approach has been widely exploited, the latter has come about only more recently; consequently, its potential is less understood. For this reason, in this paper, we study a realistic test case that allows the evaluation of the effectiveness of such a solution. As part of this process, (a) we designed a system that allows vehicles to crowd-source traffic information in an ad hoc manner, allowing them to dynamically reroute based on individually collected traffic information; (b) we implemented a realistic network-mobility simulator that allowed us to evaluate such a model; and (c) we performed a case study that evaluates whether such a decentralized system can help drivers to minimize trip times, which is the main focus of this paper. This study is based on traffic survey data from Portland, OR, and our results indicate that such navigation systems can indeed greatly improve traffic flow. Finally, to test the feasibility of our approach, we implemented our system and ran some real experiments at UCLA's C-Vet test bed.

Nowadays, the navigation systems available on cars are becoming more and more sophisticated. They greatly improve the experience of drivers and passengers by enabling them to receive map and traffic updates, news feeds, advertisements, media files, etc. Unfortunately, the bandwidth available to each vehicle with the current technology is severely limited. There have been many reports on the inability of 3G networks to cope with large size file downloads, especially in dense and mobile settings. A possible alternative is provided by WiFi access points (APs) that are being installed in several countries along the main routes and in popular areas. Although this approach significantly increases the available bandwidth, it still does not provide a fully satisfactory solution due to the limited transmission range (usually a few hundred meters). In this paper we present a novel routing protocol, based on opportunistic vehicle to vehicle communication, to enable efficient multi-hop routing capabilities between mobile vehicles and APs. Unlike prior work, this protocol fully supports two- way communication, i.e., the traditional vehicle-to-AP as well as the more challenging AP-to-vehicle. We leverage the information offered by the navigation system in terms of final destination and path, to i) route packets to the closest AP and ii) to route replies back to the moving vehicle efficiently.

An increasing number of sensor networks have been deployed to monitor a variety of conditions and situations. At the same time, more and more applications are starting to rely on the data from sensor networks to provide users with (near) real-time information and conditions. This increasing demand of users for accurate information about natural and surrounding phoenomena is creating a business case for application providers.

Content-based information dissemination has a potential number of applications in vehicular networking, including advertising, traffic and parking notifications and emergency announcements. In this paper we describe a protocol for content based information dissemination in hybrid (i.e., partially structureless) vehicular networks. The protocol allows content to ldquostickrdquo to areas where vehicles need to receive it. The vehicle's subscriptions indicate the driver's interests about types of content and are used to filter and route information to affected vehicles. The publications, generated by other vehicles or by central servers, are first routed into the area, then continuously propagated for a specified time interval. The protocol takes advantage of both the infrastructure (i.e., wireless base stations), if this exists, and the decentralized vehicle-to-vehicle communication technologies. We evaluate our approach by simulation over a number of realistic vehicular traces based scenarios. Results show that our protocol achieves high message delivery while introducing low overhead, even in scenarios where no infrastructure is available.

Vehicular networks can be seen as an example of hybrid delay tolerant network where a mixture of infostations and vehicles can be used to geographically route the information messages to the right location. In this paper we present a forwarding protocol which exploits both the opportunistic nature and the inherent characteristics of the vehicular network in terms of mobility patterns and encounters, and the geographical information present in navigator systems of vehicles. We also report about our evaluation of the protocol over a simulator using realistic vehicular traces and in comparison with other geographical routing protocols.