Generative AI is being increasingly integrated into web search for the convenience it provides users. In this work, we aim to understand how generative AI disrupts web search by retrieving and presenting the information and sources differently from traditional search engines. We introduce a public benchmark dataset of 11,500 user queries to support our study and future research of generative search. We compare the search results returned by Google's search engine, the accompanying AI Overview (AIO), and Gemini Flash 2.5 for each query. We have made several key findings. First, we find that for 51.5% of representative, real-user queries, AIOs are generated, and are displayed above the organic search results. Controversial questions frequently result in an AIO. Second, we show that the retrieved sources are substantially different for each search engine (<0.2 average Jaccard similarity). Traditional Google search is significantly more likely to retrieve information from popular or institutional websites in government or education, while generative search engines are significantly more likely to retrieve Google-owned content. Third, we observe that websites that block Google's AI crawler are significantly less likely to be retrieved by AIOs, despite having access to the content. Finally, AIOs are less consistent when processing two runs of the same query, and are less robust to minor query edits. Our findings have important implications for understanding how generative search impacts website visibility, the effectiveness of generative engine optimization techniques, and the information users receive. We call for revenue frameworks to foster a sustainable and mutually beneficial ecosystem for publishers and generative search providers.

We present CryptGNN, a secure and effective inference solution for third-party graph neural network (GNN) models in the cloud, which are accessed by clients as ML as a service (MLaaS). The main novelty of CryptGNN is its secure message passing and feature transformation layers using distributed secure multi-party computation (SMPC) techniques. CryptGNN protects the client's input data and graph structure from the cloud provider and the third-party model owner, and it protects the model parameters from the cloud provider and the clients. CryptGNN works with any number of SMPC parties, does not require a trusted server, and is provably secure even if P -1 out of P parties in the cloud collude. Theoretical analysis and empirical experiments demonstrate the security and efficiency of CryptGNN.

High-accuracy and low-latency indoor place prediction for mobile users can enable a wide range of applications for domains such as assisted living and smart homes. In this paper, we propose GoPlaces, a practical indoor place prediction system that works on mobile devices without requiring any new infrastructure. GoPlaces does not rely on servers or specialized localization infrastructure, except for a single cheap off-the-shelf WiFi access point that supports ranging with Round Trip Time (RTT) protocol. GoPlaces enables personalized place naming and prediction, and it protects users’ location privacy. It fuses inertial sensor data with distances estimated using the WiFi-RTT protocol to predict the indoor places a user will visit. GoPlaces employs an attention-based BiLSTM model to detect user’s current trajectory, which is then used together with historical information stored in a prediction tree to infer user’s future places. We implemented GoPlaces in Android and evaluated it in several indoor spaces. The experimental results demonstrate prediction accuracy as high as 86%. Furthermore, they show GoPlaces is feasible in real life because it has low latency and low resource consumption on the phones.

This paper introduces SGCode, a flexible prompt-optimizing system to generate secure code with large language models (LLMs). SGCode integrates recent prompt-optimization approaches with LLMs in a unified system accessible through front-end and back-end APIs, enabling users to 1) generate secure code, which is free of vulnerabilities, 2) review and share security analysis, and 3) easily switch from one prompt optimization approach to another, while providing insights on model and system performance. We populated SGCode on an AWS server with PromSec, an approach that optimizes prompts by combining an LLM and security tools with a lightweight generative adversarial graph neural network to detect and fix security vulnerabilities in the generated code. Extensive experiments show that SGCode is practical as a public tool to gain insights into the trade-offs between model utility, secure code generation, and system cost. SGCode has only a marginal cost compared with prompting LLMs. SGCode is available at: http://3.131.141.63:8501/.

Multi-task learning (MTL) enables simultaneous learning of related tasks, enhancing the generalization performance of each task and facilitating faster training and inference on resource-constrained devices. Federated Learning (FL) can further enhance performance by enabling collaboration among devices, effectively leveraging distributed data to improve model performance, while ensuring that the raw data remains on the respective devices. However, conventional FL is inadequate for handling MTL models trained on different sets of tasks. This paper proposes FedMTL, a new FL aggregation technique that handles task heterogeneity across users. FedMTL generates personalized MTL models based on task similarities, which are determined by analyzing the parameters for the task-specific layers of the trained models. To prevent privacy leakage through these model parameters and to protect the privacy of the task types, FedMTL employs low-overhead algorithms that are adaptable to existing techniques for secure aggregation. Extensive experiments on three datasets demonstrate that FedMTL performs better than state-of-the-art approaches. Additionally, we implement the FedMTL aggregation algorithm using secure multi-party computation, showing that it can achieve the same accuracy with the plain-text version while preserving privacy.

This paper presents a study of GDPR compliance under the Interactive Advertising Bureau Europe's Transparency and Consent Framework (TCF). This framework provides digital advertising market participants a standard for sharing users' privacy consent choices. TCF is widely used across the Internet, and this paper presents a thorough experimental evaluation of both the compliance of websites with TCF and its impact on user privacy. We reviewed 2,230 websites that use TCF and accepted the automatic decline of user consent by our data collection system. Unlike previous work on GDPR compliance, we found that most websites using TCF properly record the user's consent choice. However, we found that 72.8% of the websites that were TCF compliant claimed legitimate interest as a rationale for overriding the consent choice. While legitimate interest is legal under GDPR, previous studies have shown that most users disagreed with how it is being used to collect data. Additionally, analysis of cookies set to the browsers indicates that TCF may not fully protect user privacy even when websites are compliant. Our research provides regulators and publishers with a data collection and analysis system to monitor compliance, detect non-compliance, and examine questionable practices of circumventing user consent choices using legitimate interest.

This article presents the design, implementation, and evaluation ofFLSys, a mobile-cloud federated learning (FL) system, which can be a key component for an open ecosystem of FL models and apps. FLSys is designed to work on smart phones with mobile sensing data. It balances model performance with resource consumption, tolerates communication failures, and achieves scalability. In FLSys, different DL models with different FL aggregation methods can be trained and accessed concurrently by different apps. Furthermore, FLSys provides advanced privacy preserving mechanisms and a common API for third-party app developers to access FL models. FLSys adopts a modular design and is implemented in Android and AWS cloud. We co-designed FLSys with a human activity recognition (HAR) model. HAR sensing data was collected in the wild from 100+ college students during a 4-month period. We implemented HAR-Wild, a CNN model tailored to mobile devices, with a data augmentation mechanism to mitigate the problem of non-Independent and Identically Distributed data. A sentiment analysis model is also used to demonstrate that FLSys effectively supports concurrent models. This article reports our experience and lessons learned from conducting extensive experiments using simulations, Android/Linux emulations, and Android phones that demonstrate FLSys achieves good model utility and practical system performance.

Federated Learning (FL) is a privacy-preserving distributed deep learning paradigm that involves substantial communication and computation effort, which is a problem for resource-constrained mobile and IoT devices. Model pruning/sparsification develops sparse models that could solve this problem, but existing sparsification solutions cannot satisfy at the same time the requirements for low bidirectional communication overhead between the server and the clients, low computation overhead at the clients, and good model accuracy, under the FL assumption that the server does not have access to raw data to fine-tune the pruned models. We propose Complement Sparsification (CS), a pruning mechanism that satisfies all these requirements through a complementary and collaborative pruning done at the server and the clients. At each round, CS creates a global sparse model that contains the weights that capture the general data distribution of all clients, while the clients create local sparse models with the weights pruned from the global model to capture the local trends. For improved model performance, these two types of complementary sparse models are aggregated into a dense model in each round, which is subsequently pruned in an iterative process. CS requires little computation overhead on the top of vanilla FL for both the server and the clients. We demonstrate that CS is an approximation of vanilla FL and, thus, its models perform well. We evaluate CS experimentally with two popular FL benchmark datasets. CS achieves substantial reduction in bidirectional communication, while achieving performance comparable with vanilla FL. In addition, CS outperforms baseline pruning mechanisms for FL.

Online publishers typically sell ad impressions through auctions held in ad exchanges in real-time, i.e., real-time bidding (RTB). A publisher will accept the winning bid if it is higher than a given reserve price for an ad impression. Setting an appropriate reserve price for an ad impression is critical for publishers’ revenue generation, but also challenging. While this problem has been studied for second-price auctions, it lacks studies for first-price auctions, the de facto industry standard since 2019. This paper proposes a machine learning model that determines the optimal reserve prices for individual ad impressions in real-time. It uses a multi-task learning framework to predict the lower bounds of the highest bids with a coverage probability, using only the data available to publishers. The experiments using data from a large international publisher show that the proposed model outperforms the comparison systems on generating revenue.

This paper proposes Zone-based Federated Learning (ZoneFL) to simultaneously achieve good model accuracy while adapting to user mobility behavior, scaling well as the number of users increases, and protecting user data privacy. ZoneFL divides the physical space into geographical zones mapped to a mobile-edge-cloud system architecture for good model accuracy and scalability. Each zone has a federated training model, called a zone model, which adapts well to data and behaviors of users in that zone. Benefiting from the FL design, the user data privacy is protected during the ZoneFL training. We propose two novel zone-based federated training algorithms to optimize zone models to user mobility behavior: Zone Merge and Split (ZMS) and Zone Gradient Diffusion (ZGD). ZMS optimizes zone models by adapting the zone geographical partitions through merging of neighboring zones or splitting of large zones into smaller ones. Different from ZMS, ZGD maintains fixed zones and optimizes a zone model by incorporating the gradients derived from neighboring zones' data. ZGD uses a self-attention mechanism to dynamically control the impact of one zone on its neighbors. Extensive analysis and experimental results demonstrate that ZoneFL significantly outperforms traditional FL in two models for heart rate prediction and human activity recognition. In addition, we developed a ZoneFL system using Android phones and AWS cloud. The system was used in a heart rate prediction field study with 63 users for 4 months, which demonstrated the feasibility of ZoneFL in real-life.

Page placement mechanisms have long been used to reduce cache conflict misses. They become more important in clouds where the emerging way-based cache partitioning is used for better workload isolation but at a cost of increased cache conflicts. However, page placement mechanisms become ineffective in virtualized environments, such as clouds, because the real locations of memory pages (i.e., their host physical addresses) are hidden from guest OSs. The paper proposesxPlaceas a solution to reestablish page placement mechanisms under the nested virtualization configuration. To keep high portability and low overhead,xPlacefollows an approach that creates a synergy between the host and guest VMs, such that the page placement mechanism inside each guest VM becomes effective even if its page placement decisions are made based on the guest physical addresses of memory pages. The paper addresses the technical issues for implementing this approach in the nested virtualization setting, particularly how to create the synergy with the obstacle created by guest hypervisors sitting between the host and guest VMs. Evaluation based on the prototype implementation and diverse real world applications shows thatxPlacecan greatly reduce cache conflicts and improve application performance in the nested environment.

The fast increase in ad-blocker usage has resulted in significant revenue loss for online publishers. To mitigate this, many publishers implement the Wall strategy, where an adblock user is asked to whitelist the intended webpage. If the user refuses, the result is a loss-loss situation: the user is denied access to content, and the publisher cannot receive revenue. An alternative strategy, called AAX, is to show only acceptable ads to users. However, acceptable ads generate less revenue than regular ads. This article proposes personalized counter ad-blocking that dynamically chooses a counter ad-blocking strategy for individual users. To implement it, we propose a novel deep learning-based whitelist prediction model. Adblock users predicted to whitelist a page receive the Wall strategy; the others receive the AAX strategy. The proposed Deep Ad-Block Whitelist Network (DAWN) for whitelist prediction captures page characteristics, user interests in pages and their sensitivity to ads, reflected in historic behavior, using a deep learning mechanism. Furthermore, DAWN leverages multi-task learning on whitelist prediction and dwell-time prediction to boost performance. DAWN's effectiveness is validated on a real-world dataset provided by Forbes Media. The experimental results demonstrate the advantages of the proposed counter ad-blocking policy over existing policies on revenue generation and user engagement.

High-accuracy and low-latency indoor place prediction for mobile users is crucial to enable applications for assisted living, emergency services, smart homes, and augmented reality. Previous studies on indoor place prediction use complex infrastructure with multiple visual/wireless anchors or multiple wireless access points. These localization techniques are difficult to deploy, may negatively impact user privacy through location tracking, and their data collection is not suitable for personalized place prediction. To solve these challenges, this paper proposes GoPlaces, a novel app that fuses inertial sensor data with WiFi-RTT estimated distances to predict the future indoor places visited by a user. GoPlaces does not require any infrastructure, except for one cheap off-the-shelf WiFi access point that supports ranging with RTT. In addition, it enables personalized place naming and prediction through its on-the-phone data collection and protects users' location privacy because user's data never leaves the phone. GoPlaces uses an attention-based bidirectional long short-term memory model to detect user's current trajectory, which is then used together with historical information stored in a prediction tree to infer user's future places. We implemented GoPlaces in Android and evaluated it in several indoor spaces. The experimental results demonstrate prediction accuracy as high as 92%, low latency, and low resource consumption on the phones.

Cruising for vacant curbside parking spaces causes waste of time, frustration, waste of fuel, and pollution. This problem has been addressed by centralized solutions that perform parking assignments and communicate them to drivers’ smart phones. These solutions suffer, however, from two intrinsic problems: scalability, as the server has to perform intensive computation and communication with the drivers; and privacy, as the drivers have to disclose their destinations to the server. This article proposes DFPS, a distributed mobile system for free parking assignment. DFPS solves the scalability problem by using the drivers’ smart phones to cooperatively compute the parking assignments, and a centralized dispatcher to receive and distribute parking requests to the network of smart phones. The phones of the parked drivers in DFPS are structured in a K-D tree to serve parking requests in a distributed fashion. DFPS removes the computation from the dispatcher and substantially reduces its communication load. DFPS solves the privacy problem through an entropy-based cloaking technique that runs on drivers’ smart phones and conceals drivers’ destinations from the dispatcher. The evaluation demonstrates that DFPS is scalable and obtains better travel time than a centralized system, while protecting the privacy of drivers’ destinations.

Fine-grained location prediction on smart phones can be used to improve app/system performance. Application scenarios include video quality adaptation as a function of the 5G network quality at predicted user locations, and augmented reality apps that speed up content rendering based on predicted user locations. Such use cases require prediction error in the same range as the GPS error, and no existing works on location prediction can achieve this level of accuracy. We propose Federated Meta-Location Learning (FMLL) on smart phones for fine-grained location prediction, based on GPSt races collected on the phones. FMLL has three components: a meta-location generation module, a prediction model, and a federated learning framework. The meta-location generation module represents the user location data as relative points in an abstract 2D space, which enables learning across different physical spaces. The model fuses Bidirectional Long Short-Term Memory (BiLSTM) and Convolutional Neural Networks (CNN), where BiLSTM learns the speed and direction of the mobile users, and CNN learns information such as user movement preferences. The framework runs on the phones of the users and also on a server that coordinates learning from all users in the system. FMLL uses federated learning to protect user privacy and reduce bandwidth consumption. Our experimental results, using a dataset with over 600,000 users, demonstrate that FMLL outperforms baseline models in terms of prediction accuracy. We also demonstrate that FMLL works well in conjunction with transfer learning, which enables model reusability. Finally, benchmark results on Android phones demonstrate FMLL’s feasibility in real life.

Traditional page recommendation models are endangered by stricter privacy regulations, such as the General Data Protection Regulation (GDPR). The performance of these models suffer when only a part of the users share their personal data, such as cookies, with web servers, while the rest of the users choose to opt-out from sharing these data. Furthermore, these models are not designed to provide recommendations for users who do not share their data. This paper addresses the question of how to provide good page recommendations to all users, independent of their privacy attitudes. We propose Fed4Rec, a privacy-preserving framework for page recommendation based on federated learning (FL) and model-agnostic meta-learning (MAML), which allows machine learning models to train on data collected from both public users, who share data with the server, and private users, who do not share data with the server. Fed4Rec enables recommendations for both public users, computed at the server, and private users, computed at their local devices. Private users' data are stored only on user devices and never shared with the server. FL is used to train on local data, and Fed4Rec shares with the server only partial model parameters, computed on local devices. MAML is used to jointly train on the public data and the model parameters from the private users. We compare Fed4Rec against several baseline frameworks, using a publicly available dataset from a large news portal. The results show that Fed4Rec outperforms the baselines in terms of recommendation accuracy. We also conduct one ablation study to examine the impact of varying the ratio between the number of public and private users. Fed4Rec performs better than the baselines for all ratios, but it is especially beneficial w hen t he p ercentage of public users is low.

Cellular traffic prediction enables operators to adapt to traffic demand in real-time for improving network resource utilization and user experience. To predict cellular traffic, previous studies either applied Recurrent Neural Networks (RNN) at individual base stations or adapted Convolutional Neural Networks (CNN) to work at grid-cells in a geographically defined grid. These solutions do not consider explicitly the effect of handover on the spatial characteristics of the traffic, which may lead to lower prediction accuracy. Furthermore, RNN solutions are slow to train, and CNN-grid solutions do not work for cells and are difficult to apply to base stations. This paper proposes a new prediction model, STGCN-HO, that uses the transition probability matrix of the handover graph to improve traffic prediction. STGCN-HO builds a stacked residual neural network structure incorporating graph convolutions and gated linear units to capture both spatial and temporal aspects of the traffic. Unlike RNN, STGCN-HO is fast to train and simultaneously predicts traffic demand for all base stations based on the information gathered from the whole graph. Unlike CNN-grid, STGCN-HO can make predictions not only for base stations, but also for cells within base stations. Experiments using data from a large cellular network operator demonstrate that our model outperforms existing solutions in terms of prediction accuracy.

The fast growing ad-blocker usage results in large revenue decrease for ad-supported online websites. Facing this problem, many online publishers choose either to cooperate with ad-blocker software companies to show acceptable ads or to build a wall that requires users to whitelist the site for content access. However, there is lack of studies on the impact of these two counter-ad-blocking strategies on user behaviors. To address this issue, we conduct a randomized field experiment on the website of Forbes Media, a major US media publisher. The ad-blocker users are divided into a treatment group, which receives the wall strategy, and a control group, which receives the acceptable ads strategy. We utilize the difference-in-differences method to estimate the causal effects. Our study shows that the wall strategy has an overall negative impact on user engagements. However, it has no statistically significant effect on high-engaged users as they would view the pages no matter what strategy is used. It has a big impact on low-engaged users, who have no loyalty to the site. Our study also shows that revisiting behavior decreases over time, but the ratio of session whitelisting increases over time as the remaining users have relatively high loyalty and high engagement. The paper concludes with discussions of managerial insights for publishers when determining counter-ad-blocking strategies.

With cloud assistance, mobile apps can offload their resource-demanding computation tasks to the cloud. This leads to a scenario where computation tasks in the same program run concurrently on both the mobile device and the cloud. An important challenge is to ensure that the tasks are able to access and share the files on both the mobile and the cloud in a manner that is efficient, consistent, and transparent to locations. Existing distributed file systems and network file systems do not satisfy these requirements. Current systems for offloading tasks either do not support file access for offloaded tasks or do not offload tasks with file access. The paper addresses this issue by designing and implementing an application-level file system called Overlay File System (OFS). To improve efficiency, OFS maintains and buffers local copies of data sets on both the cloud and the mobile device. OFS ensures consistency and guarantees that all the reads get the latest data. It combines write-invalidate and write-update policies to effectively reduce the network traffic incurred by invalidating/updating stale data copies and to reduce the execution delay when the latest data cannot be accessed locally. To guarantee location transparency, OFS creates a unified view of the data that is location independent and is accessible as local storage. We overcome the challenges caused by the special features of mobile systems on an application-level file system, like the lack of root privilege and state loss when application is killed due to the shortage of resource and implement an easy to deploy prototype of OFS. The paper tests the OFS prototype on Android OS with a real mobile app and real mobile user traces. Extensive experiments show that OFS can effectively support consistent file accesses from computation tasks, no matter whether they are on a mobile device or offloaded to the cloud. In addition, OFS reduce both file access latency and network traffic incurred by file accesses.

The fast increase in ad blocker usage results in large revenue loss for online publishers and advertisers. Many publishers initialize counter-ad-blocking strategies, where a user has to choose either whitelisting the publisher's web site in their ad blocker or leaving the site without accessing the content. This paper aims to predict the user whitelisting behavior, which can help online publishers to better assess users' interests and design corresponding strategies. We present several techniques for personalized whitelist prediction for a target user and a target web page. Our prediction models are evaluated on real-world data provided by a large online publisher, Forbes Media. The best prediction performance was achieved using the gradient boosting regression tree model, which also demonstrated robustness and efficiency.

The ever-evolving mobile applications need more and more computing resources to smooth user experience and sometimes meet delay requirements. Therefore, mobile devices (MDs) are gradually having difficulties to complete all tasks in time due to the limitations of computing power and battery life. To cope with this problem, mobile edge computing (MEC) systems were created to help with task processing for MDs at nearby edge servers. Existing works have been devoted to solving MEC task offloading problems, including those with simple delay constraints, but most of them neglect the coexistence of deadline-constrained and delay- sensitive tasks (i.e., the diverse delay sensitivities of tasks). In this paper, we propose an actor-critic based deep reinforcement learning (ADRL) model that takes the diverse delay sensitivities into account and offloads tasks adaptively to minimize the total penalty caused by deadline misses of deadline-constrained tasks and the lateness of delay-sensitive tasks. We train the ADRL model using a real data set that consists of the diverse delay sensitivities of tasks. Our simulation results show that the proposed solution outperforms several heuristic algorithms in terms of total penalty, and it also retains its performance gains under different system settings.

The revenue of online display advertising in the U.S. is projected to be 7.9 billion U.S. dollars by 2022. One main way of display advertising is through real-time bidding (RTB). In RTB, an ad exchange runs a second price auction among multiple advertisers to sell each ad impression. Publishers usually set up a reserve price, the lowest price acceptable for an ad impression. If there are bids higher than the reserve price, then the revenue is the higher price between the reserve price and the second highest bid; otherwise, the revenue is zero. Thus, a higher reserve price can potentially increase the revenue, but with higher risks associated. In this paper, we study the problem of estimating the failure rate of a reserve price, i.e., the probability that a reserve price fails to be outbid. The solution to this problem have managerial implications to publishers to set appropriate reserve prices in order to minimizes the risks and optimize the expected revenue. This problem is highly challenging since most publishers do not know the historical highest bidding prices offered by RTB advertisers. To address this problem, we develop a parametric survival model for reserve price failure rate prediction. The model is further improved by considering user and page interactions, and header bidding information. The experimental results demonstrate the effectiveness of the proposed approach.

This paper aims to provide an efficient solution for people in a city who drive their cars to visit several destinations, where they need to park for a while, but do not care about the visiting order. This instance of the multi-destination route planning problem is novel in terms of its constraints: the real-time traffic conditions and the real-time free parking conditions in the city. The paper proposes a novel Multi-Destination Vehicle Route Planning (MDVRP) system to optimize the travel time for all drivers. MDVRP's design has two components: a mobile app running on the drivers' smart phones that submits real-time route requests and guides the drivers toward destinations, and a server in the cloud that optimizes the routes by finding the most efficient order to visit the destinations. MDVRP uses TDTSP-FPA, an algorithm that finds the fastest route to the next destination and also assigns free curbside parking spaces that minimize the total travel time for drivers. We evaluate MDVRP using a driver trip dataset that contains real vehicular mobility traces of over two million drivers from the city of Cologne, Germany. By learning the spatio-temporal distribution of real driver destinations from this dataset, we build a novel experimental platform that simulates real, multi-destination driver trips. Extensive simulations executed over this platform demonstrate that TDTSP-FPA delivers the best performance when compared to three baseline algorithms.

Computation offloading is employed by mobile apps running over resource-constrained devices to leverage the cloud in overcoming their resource limits. The advent of the Multi-access Edge Computing (MEC) paradigm further extends the potential opportunities of mobile-cloud offloading, allowing new service provisioning scenarios, such as mobile gaming and multimedia, where responsiveness of mobile devices at the network edge significantly benefits from low latency interactions. However, state-of-the-art offloading platforms for MEC architectures have not addressed the technical challenge of supporting specific file systems for this MEC-enabled class of applications, with components running at three hosting environments, i.e., mobile, edge, and cloud. This paper proposes the Mobile Edge File System (MEFS), an application-level distributed file system designed to be highly resilient and able to efficiently maintain consistency among the mobile, edge, and cloud entities. MEFS supports application handoff through live migration as end devices move between edges. The cloud transparently helps with recovery from faulty edge nodes or in the case of unavailability of edges in the user's proximity. We implemented a MEFS prototype in Android along with MEFS-based MEC-enabled mobile apps. The experimental results show how MEFS can achieve low latency and low overhead.

Display advertising is the most important revenue source for publishers in the online publishing industry. The ad pricing standards are shifting to a new model in which ads are paid only if they are viewed. Consequently, an important problem for publishers is to predict the probability that an ad at a given page depth will be shown on a user's screen for a certain dwell time. This paper proposes deep learning models based on Long Short-Term Memory (LSTM) to predict the viewability of any page depth for any given dwell time. The main novelty of our best model consists in the combination of bi-directional LSTM networks, encoder-decoder structure, and residual connections. The experimental results over a dataset collected from a large online publisher demonstrate that the proposed LSTM-based sequential neural networks outperform the comparison methods in terms of prediction performance.

Recent research has proposed middleware to enable efficient distributed apps over mobile-cloud platforms. This paper presents a Context-Aware File Discovery Service (CAFDS) that allows distributed mobile-cloud applications to find and access files of interest shared by collaborating users. CAFDS enables programmers to search for files defined by context and content features, such as location, creation time, or the presence of certain object types within an image file. CAFDS provides low-latency through a cloud-based metadata server, which uses a decision tree to locate the nearest files that satisfy the context and content features requested by applications. We implemented CAFDS in Android and Linux. Experimental results show CAFDS achieves substantially lower latency than peer-to-peer solutions that cannot leverage context information.

Applying Machine Learning (ML) to business applications for automation usually faces difficulties when integrating diverse ML dependencies and services, mainly because of the lack of a common ML framework. In most cases, the ML models are developed for applications which are targeted for specific business domain use cases, leading to duplicated effort, and making reuse impossible. This paper presents Acumos, an open platform capable of packaging ML models into portable containerized microservices which can be easily shared via the platform's catalog, and can be integrated into various business applications. We present a case study of packaging sentiment analysis and classification ML models via the Acumos platform, permitting easy sharing with others. We demonstrate that the Acumos platform reduces the technical burden on application developers when applying machine learning models to their business applications. Furthermore, the platform allows the reuse of readily available ML microservices in various business domains.

This paper presents Sentio, a distributed middle-ware designed to provide mobile apps with seamless connectivity to remote sensors when the sensing code and the sensors are not physically on the same device, e.g., when the sensing code is offloaded to the cloud. Sentio presents the apps with virtual sensors that are mapped to remote physical sensors. Virtual sensors can be composed into higher-level sensors, which fuse sensing data from multiple physical sensors. Furthermore, they are mapped to the best available physical sensors when the app starts and re-mapped transparently to other physical sensors at runtime in response to context changes. Sentio was designed to work without modifications to the operating system and to provide low-latency access to remote sensors, which is beneficial to apps with real time-requirements such as mobile games. We have built a prototype of Sentio on Android. We have also developed four apps based on Sentio to understand the programming effort and evaluate the performance. The development of the apps shows that complex sensing tasks can be implemented quickly, benefiting from Sentio's high-level API. The experimental results show that Sentio achieves good real-time performance.

A half of online display ads are not rendered viewable because the users do not scroll deep enough or spend sufficient time at the page depth where the ads are placed. In order to increase the marketing efficiency and ad effectiveness, there is a strong demand for viewability prediction from both advertisers and publishers. This paper aims to predict the dwell time for a given triplet based on historic data collected by publishers. This problem is difficult because of user behavior variability and data sparsity. To solve it, we propose predictive models based on Factorization Machines and Field‐aware Factorization Machines in order to overcome the data sparsity issue and provide flexibility to add auxiliary information such as the visible area of a user's browser. In addition, we leverage the prior dwell time behavior of the user within the current page view, that is, time series information, to further improve the proposed models. Experimental results using data from a large web publisher demonstrate that the proposed models outperform comparison models. Also, the results show that adding time series information further improves the performance.

Online display advertising has becomes a billion-dollar industry, and it keeps growing. Advertisers attempt to send marketing messages to attract potential customers via graphic banner ads on publishers' webpages. Advertisers are charged for each view of a page that delivers their display ads. However, recent studies have discovered that more than half of the ads are never shown on users' screens due to insufficient scrolling. Thus, advertisers waste a great amount of money on these ads that do not bring any return on investment. Given this situation, the Interactive Advertising Bureau calls for a shift toward charging by viewable impression, i.e., charge for ads that are viewed by users. With this new pricing model, it is helpful to predict the viewability of an ad. This paper proposes two probabilistic latent class models (PLC) that predict the viewability of any given scroll depth for a user-page pair. Using a real-life dataset from a large publisher, the experiments demonstrate that our models outperform comparison systems.

Centralized solutions for vehicular traffic re-routing to alleviate congestion suffer from two intrinsic problems: scalability, as the central server has to perform intensive computation and communication with the vehicles in real-time; and privacy, as the drivers have to share their location as well as the origins and destinations of their trips with the server. This article proposes DIVERT, a distributed vehicular re-routing system for congestion avoidance. DIVERT offloads a large part of the re-routing computation at the vehicles, and thus, the re-routing process becomes practical in real-time. To take collaborative re-routing decisions, the vehicles exchange messages over vehicular ad hoc networks. DIVERT is a hybrid system because it still uses a server and Internet communication to determine an accurate global view of the traffic. In addition, DIVERT balances the user privacy with the re-routing effectiveness. The simulation results demonstrate that, compared with a centralized system, the proposed hybrid system increases the user privacy by 92 percent on average. In terms of average travel time, DIVERT's performance is slightly less than that of the centralized system, but it still achieves substantial gains compared to the no re-routing case. In addition, DIVERT reduces the CPU and network load on the server by 99.99 and 95 percent, respectively.

The amount of time spent by users at specific page depths within webpages, called dwell time, can be used by web publishers to decide where to place online ads and what type of ads to place at different depths within a webpage. This paper presents a model to predict the dwell time for a given "user, webpage, depth" triplet based on historic data collected by publishers. Dwell time prediction is difficult due to user behavior variability and data sparsity. We adopt the Factorization Machines model because it is able to capture the interaction between users and webpages, overcome the data sparsity issue, and provide flexibility to add auxiliary information such as the visible area of a user's browser. Experimental results using data from a large web publisher demonstrate that our model outperforms deterministic and regression-based comparison models.

With cloud assistance, a mobile application can offload its resource-demanding computation tasks to the cloud (public cloud, cloudlet, or personal cloud, etc). This leads to a scenario where computation tasks in the same application run concurrently on both the mobile device and the cloud. These tasks need to save, read, and write files on both the mobile device and the cloud. An important challenge is to ensure that the tasks are able to access and share the files in a manner that is efficient, consistent, and transparent to locations. The paper addresses this issue by designing an application-level file system called Overlay File System (OFS). To improve efficiency, OFS maintains and buffers local copies of data sets on both the cloud and the mobile device. OFS ensures consistency and guarantees that all the reads get the latest data. It combines write-invalidate and write-update policies to effectively reduce the network traffic incurred by invalidating/updating stale data copies and to reduce the application delay when the latest data cannot be accessed locally. To guarantee location transparency, OFS creates an unified view of the data that is location independent and is accessible as local storage. Our experiments show that OFS can effectively support task offloading and efficient execution of offloaded tasks by significantly decreasing both file access latency and network traffic incurred by file accesses.

Mobile participatory sensing could be used in many applications such as vehicular traffic monitoring, pollution tracking, or even health surveying. However, its success depends on finding a solution for querying large numbers of users which protects user location privacy and works in real-time. This paper presents PAMPAS, a privacy-aware mobile distributed system for efficient data aggregation in mobile participatory sensing. In PAMPAS, mobile devices enhanced with secure hardware, called secure probes (SPs), perform distributed query processing, while preventing users from accessing other users' data. A supporting server infrastructure (SSI) coordinates the inter-SP communication and the computation tasks executed on SPs. PAMPAS ensures that SSI cannot link the location reported by SPs to the user identities even if SSI has additional background information. In addition to its novel system architecture, PAMPAS also proposes two new protocols for privacy-aware location-based aggregation and adaptive spatial partitioning of SPs that work efficiently on resource-constrained SPs. Our experimental results and security analysis demonstrate that these protocols are able to collect the data, aggregate them, and share statistics or derived models in real-time, without any location privacy leakage.

Catastrophic disasters can destroy large regions and, in the process, leave many victims isolated from the rest of the world. Recovering the communication infrastructure is typically slow and expensive, which is not suitable for emergency response. Multihop wireless access networks have the potential to quickly provide Internet connectivity to victims, but so far no simple and practical solution has been proposed to help people configure these networks easily. We are pursuing the approach of utilizing wireless virtualization techniques to establish wireless access networks on-the-fly using on-site mobile devices. While our previous work has demonstrated proof-of-concept solutions, it lacked fundamental communication abstractions, a rigorous design, and a thorough analysis on the effectiveness of these solutions. The main new contributions of this article are: (1) the wireless multihop communication abstraction (WMCA) as a fundamental communication concept for a practical tree-based disaster recovery access network (TDRAN), (2) the complete design and implementation details of TDRAN, and (3) a comprehensive analysis of the effectiveness of the proposed approach based on field experiments, both in indoor and outdoor settings, at different sites in Japan. The results demonstrate the effectiveness of the proposed solution for on-site configuration of wireless access networks, as it can easily extend to 20 hops by 15 m-distance and 16 hops by 30 m-distance networks, which result in 300 m and 480 m (respectively) in radius or about 1 km in diameter. This work also confirms that our approach is ready for realization as a real disaster recovery solution.

As a massive industry, display advertising delivers advertisers' marketing messages to attract customers through graphic banners on webpages. Advertisers are charged by ad serving, where their ads are shown in web pages. However, recent studies show that about half of the ads were actually never seen by users because they do not scroll deep enough to bring the ads in-view. Thus, the ad pricing standards are shifting to a new model: ads are paid if they are in view, not just being served. To the best of our knowledge, this paper is the first to address the important problem of ad viewability prediction which can improve the performance of guaranteed ad delivery, real-time bidding, as well as recommender systems. We analyze a real-life dataset from a large publisher, identify a number of features that impact the scroll depth for a given user and a page, and propose a probabilistic latent class model that predicts the viewability of any given scroll depth for a user-page pair. The experiments demonstrate that our model outperforms comparison systems based on singular value decomposition and logistic regression, in terms of prediction quality and training time.

An unprecedented information wealth produced by online social networks, further augmented by location/collocation data, is currently fragmented across different proprietary services. Combined, it can accurately represent the social world and enable novel socially aware applications. We present Prometheus, a socially aware peer-to-peer service that collects social information from multiple sources into a multigraph managed in a decentralized fashion on user-contributed nodes, and exposes it through an interface implementing nontrivial social inferences while complying with user-defined access policies. Simulations and experiments on PlanetLab with emulated application workloads show the system exhibits good end-to-end response time, low communication overhead, and resilience to malicious attacks.

Traffic congestion causes driver frustration and costs billions of dollars annually in lost time and fuel consumption. This paper presents three traffic re-routing strategies designed to be incorporated in a cost-effective and easily deployable vehicular traffic guidance system that reduces the effect of traffic congestions. This system collects real-time traffic data from vehicles and road-side sensors and computes proactive, individually-tailored re-routing guidance which is pushed to vehicles when signs of congestion are observed on their route. Extensive simulation results over two urban road networks show that all three strategies, namely multipath load balancing considering future vehicle positions (EBkSP), random multipath load balancing (RkSP), and dynamic shortest path (DSP), significantly decrease the average travel time. EBkSP is the best, with as much as 104% improvement compared to the "no re-routing" baseline. Additionally, it lowers with 34% the re-routing frequency compared to the other strategies. Finally, all strategies offer good improvements even when many drivers ignore the guidance or when the system adoption rate is relatively low.

Accurate lane recognition for moving vehicles is important for lane keeping and lane changing assistance systems. Additionally, this information could be leveraged by Intelligent Transportation Systems to suggest lane changes for improved traffic load balancing across lanes. This paper presents a position estimation algorithm for moving vehicles based on RFID (Radio Frequency Identification) active sensors placed on roadsides and lane boundaries, and multiple on-car RFID receiver antennas. To improve localization accuracy, the algorithm proposes two novel ideas: (1) compute pair-wise position estimates using the RSSI (Received Signal Strength Indication) of all pairs of signals received from RFIDs, and (2) compute the final position as a weighted average of these pair-wise estimates using a dynamic weighting function that assigns higher weights to positions estimated based on closer RFIDs. The results from our field experiments indicate that the proposed method achieves 0.7-meter localization accuracy when RFIDs are placed at 0.5-meter intervals and a vehicle has 8 antennas. This accuracy allows a moving vehicle to recognize which lane it is in. The localization accuracy of the proposed method was found to be mostly stable for any type of road shape and any number of lanes. A further 14% accuracy improvement is achieved when RFIDs are placed at 0.25-meter intervals and the RFIDs located farther than 30-meter are excluded from computation.

Message ferries (MF) have been proposed as special nodes to improve communication in sparse, intermittent networks, such as delay-and disruption-tolerant networks (DTNs) by providing non-random message transfer opportunities to participating nodes. However, considering the mobility of nodes and the speed of the ferry, merely ensuring non-random encounters with hosts and the ferry do not guarantee optimal performance as the contact duration may not be enough to ensure adequate message transfers. The end-to-end performance could suffer significantly if message sizes are large. To address this issue, we present two methods that improve the performance of ferry-assisted DTNs: ferry access points (FAPs) and sticky transfers. FAPs supplement a ferry-assisted DTN with rendezvous points that increase the contact opportunities between mobile nodes and the ferry. Sticky transfers, upon mutual agreement, allow two encountered nodes to remain within transmission range of each other until they are able to complete necessary message transfers. This improves the number of successfully forwarded messages by allowing more messages to be transferred during the contact duration. Sticky transfers inherently reduce message transfer aborts. Our simulation results demonstrate that particular FAP placements improve the delivery ratio by as much as 25% for sparse networks and reduce the delay by as much as 40% for relatively dense networks. Sticky transfers on top of FAP-based DTNs achieved 100% delivery ratio for most situations, except in the case of highly-loaded, dense networks. Additionally, using sticky transfers resulted in lower delays by as much as 65% compared to non-sticky transfer cases.

The widespread adoption of smart phones allows for the seamless capture of social interactions on a scale that was once impossible. Co-presence, collected using Bluetooth on the phones, faithfully represents such real-world social interactions. This social information can be transformed into communities, which can be leveraged into applications such as recommender systems and collaborative tools. However, correctly identifying communities is difficult. This paper presents TIE, a visualization tool that enables effective review of detected communities. With TIE, we can visualize the social interaction of a set of people over time. Also, TIE can overlay detected community events in a usable way over the underlying social interactions. Further, it allows us to investigate specific social interaction events and see how well detected communities match those events. Lastly, it enables the comparison of different sets of detected communities by interactively switching between overlays. TIE has proven useful in evaluating our community detection algorithms and has been invaluable in identifying strengths and weaknesses of these algorithms. Beyond our needs, TIE is usable for other data sets that can be reduced to temporal interaction events such as multiplayer game communities, SMS interactions, and paper co-authorship.

To ensure fair and secure communication in Mobile Ad hoc Networks (MANETs), the applications running in these networks must be regulated by proper communication policies. However, enforcing policies in MANETs is challenging because they lack the infrastructure and trusted entities encountered in traditional distributed systems. This paper presents the design and implementation of a policy enforcing mechanism based on Satem, a kernel-level trusted execution monitor built on top of the Trusted Platform Module. Under this mechanism, each application or protocol has an associated policy. Two instances of an application running on different nodes may engage in communication only if these nodes enforce the same set of policies for both the application and the underlying protocols used by the application. In this way, nodes can form trusted application-centric networks. Before allowing a node to join such a network, Satem verifies its trustworthiness of enforcing the required set of policies. Furthermore, Satem protects the policies and the software enforcing these policies from being tampered with. If any of them is compromised, Satem disconnects the node from the network. We demonstrate the correctness of our solution through security analysis, and its low overhead through performance evaluation of two MANET applications.

Mobile ad hoc networks can be leveraged to provide ubiquitous services capable of acquiring, processing, and sharing real-time information from the physical world. Unlike Internet services, these services have to survive frequent and unpredictable faults such as disconnections, crashes, or users tur

Nodes in MANETs lack the protection offered by firewalls in infrastructure-based networks because malicious nodes can roam into the vicinity of another node and start launching attacks. This paper presents a distributed mechanism that allows trusted nodes to create protected networks in MANETs. A protected network is created to run a specific application and enforce a common network access control policy associated with that application. To become a member in the protected network, a node has to demonstrate its trustworthiness by proving its ability to enforce policies. Attacks from untrusted nodes are impossible because these nodes are not allowed to establish wireless links with member nodes. Attacks from member nodes are stopped at the originators by the network policy. The trusted execution of all programs involved in policy enforcement is guaranteed by a kernel agent. We demonstrate the correctness of our solution through security analysis and its feasibility through a prototype implementation tested over an IEEE 802.11 ad hoc network.

Ad hoc networks can be used not only as data carriers for mobile devices but also as providers of a new class of services specific to ubiquitous computing environments. Building services in ad hoc networks, however, is challenging due to the rapidly changing operating contexts, which often lead to situations where a node hosting a certain service becomes unsuitable for hosting the service execution any longer. We propose a novel model of service provisioning in ad hoc networks based on the concept of context- aware migratory services. Unlike a regular service that executes always on the same node, a migratory service can migrate to different nodes in the network in order to accomplish its task. The migration is triggered by changes of the operating context, and it occurs transparently to the client application. We designed and implemented a framework for developing migratory services. We built TJam, a proof-of-concept migratory service that predicts traffic jams in a given region of a highway by using only car-to-car short-range wireless communication. The experimental results obtained over an ad hoc network of personal digital assistants (PDAs) show the effectiveness of our approach in the presence of frequent disconnections. We also present simulation results that demonstrate the benefits of migratory services in large-scale networks compared to a statically centralized approach.

Web services and service oriented architectures are becoming the de facto standard for Internet computing. A main problem faced by users of such services is how to ensure that the service code is trusted. While methods that guarantee trusted service code execution before starting a client-service transaction exist, there is no solution for extending this assurance to the entire lifetime of the transaction. This paper presents Satem, a Service-aware trusted execution monitor that guarantees the trustworthiness of the service code across a whole transaction. The Satem architecture consists of an execution monitor residing in the operating system kernel on the service provider platform, a trust evaluator on the client platform, and a service commitment protocol. During this protocol, executed before every transaction, the client requests and verifies against its local policy a commitment from the service platform that promises trusted code execution. Subsequently, the monitor enforces this commitment for the duration of the transaction. To initialize the trust on the monitor, we use the Trusted Platform Module specified by the Trusted Computing Group. We implemented Satem under the Linux 2.6.12 kernel and tested it for a Web service and DNS. The experimental results demonstrate that Satem does not incur significant overhead to the protected services and does not impact the unprotected services

EZCab is a proof-of-concept ubiquitous computing application that allows people to book nearby cabs using their cell phones or PDAs equipped with short-range wireless network interfaces. EZCab discovers and books free cabs using mobile ad hoc networks of vehicles. We have implemented an EZCab prototype on top of smart messages, a middleware architecture based on execution migration, which we had developed to provide a common execution environment for outdoor ubiquitous computing applications. The experimental and simulation results have demonstrated the feasibility of EZCab

Spatial programming (SP) is a space-aware programming model for outdoor distributed embedded systems. Central to SP are the concepts of space and spatial reference, which provide applications with a virtual resource naming in networks of embedded systems. A network resource is referenced using its expected physical location and properties. Together with other SP features, such as reference consistency and access timeout, they help programmers cope with highly dynamic network configurations in a network-transparent fashion. We present the SP design and its implementation using smart messages, a lightweight software architecture similar to mobile agents, that we developed for networks of embedded systems. We also describe the implementation and evaluation of a simple SP application over a testbed consisting of HP iPAQs running Linux and equipped with 802.11 cards for wireless communication. The experimental results indicate that SP is a viable programming model for outdoor distributed computing.

Recent advances in wireless technology allow Java-enabled devices, such as Smart Phones and PDAs, to create mobile ad hoc networks, over which distributed applications can be executed. Although Java shields the programmers from the heterogeneity of the hardware platforms, a common middleware architecture is needed to support a cooperative execution environment in these networks. In this paper, we present a portable runtime system for smart messages (SMs), a middleware architecture based on execution migration, that we designed and implemented on top of an unmodified Java virtual machine. To facilitate portability, we have designed a lightweight migration mechanism based on Java bytecode instrumentation. This mechanism is suitable for mobile ad hoc networks where limited bandwidth and mobility impose constraints on the amount of data transferred. The experimental results for applications executed over wireless networks of HP iPAQs demonstrate the feasibility of our portable runtime system.

In this paper, we present the design and implementation of Smart Messages, a distributed computing platform for networks of embedded systems based on execution migration. A Smart Message (SM) is a user-defined distributed program which executes on nodes of interest, named by their properties, and uses an explicit lightweight migration to reach these nodes. During migrations, an SM carries its code and execution state, and it self-routes at each intermediate node between two nodes of interest. The nodes in the network cooperate to support the SM execution by providing a virtual machine and a shared memory region addressable by names (tag space). To illustrate the flexibility of SMs to program real world applications, we describe EZCab, an application for booking cabs in densely populated urban areas. We also present experimental results to quantify the performance achieved by the SM prototype.

Smart messages (SMs) are dynamic collections of code and data that migrate to nodes of interest in the network and execute on these nodes. A key challenge in programming pervasive computing environments using SMs is the ability to route SMs to nodes named by content. This paper describes the SM self-routing mechanism, which allows SMs to route themselves at each hop in the path toward a node of interest. SM applications can choose among multiple content based routing algorithms, switch dynamically the routing algorithm in use, or even implement the best suited routing algorithm for their needs. The main benefits provided by self-routing are high flexibility and resilience to adverse network conditions. To demonstrate these benefits, we present proof-of-concept implementation, simulation results, and analysis for the SM self-routing mechanism using several content-based routing algorithms. We also show preliminary results for SM routing algorithms executed over our SM prototype implementation.

The next generation of computing systems will be embedded, in a virtually unbounded number, and dynamically connected. The current software, network architectures, and their associated programming models are not suitable for this scenario. This paper presents a distributed computing model, Cooperative Computing, and the Smart Messages architecture for programming large networks of embedded systems. In Cooperative Computing, distributed applications are dynamic collections of migratory execution units, called Smart Messages, working to achieve a common goal. Virtually any user-defined distributed application can be implemented using our model. We present preliminary results for our prototype implementation as well as simulation results for two previously proposed applications for sensor networks, Directed Diffusion and SPIN, implemented using Smart Messages.

We propose a system architecture and a computing model, based on Smart Messages (SMs), for computation and communication in large networks of embedded systems. In this model, communication is realized by sending SMs in the network. These messages are comprised of code, which is executed at each hop in the path of the message, and data which the message carries in the network. The execution at each hop determines the next hop in the message's path - SMs are responsible for their own routing.