Client-side Solution Research Articles

FedViT: Federated continual learning of vision transformer at edge

Deep Neural Networks (DNNs) have been ubiquitously adopted in internet of things and are becoming an integral part of our daily life. When tackling the evolving learning tasks in real world, such as classifying different types of objects, DNNs face the challenge to continually retrain themselves according to the tasks on different edge devices. Federated continual learning (FCL) is a promising technique that offers partial solutions but yet to overcome the following difficulties: the significant accuracy loss due to the limited on-device processing, the negative knowledge transfer caused by the limited communication of non-IID (non-Independent and Identically Distributed) data, and the limited scalability on the tasks and edge devices. Moreover, existing FCL techniques are designed for convolutional neural networks (CNNs), which have not utilized the full potential of newly emerged powerful vision transformers (ViTs). Considering ViTs depend heavily on training data diversity and volume, we hypothesize ViTs are well-suited for FCL where data arrives continually. In this paper, we propose FedViT, an accurate and scalable federated continual learning framework for ViT models, via a novel concept of signature task knowledge. FedViT is a client-side solution that continuously extracts and integrates the knowledge of signature tasks which are highly influenced by the current task. Each client of FedViT is composed of a knowledge extractor, a gradient restorer and, most importantly, a gradient integrator. Upon training for a new task, the gradient integrator ensures the prevention of catastrophic forgetting and mitigation of negative knowledge transfer by effectively combining signature tasks identified from the past local tasks and other clients’ current tasks through the global model. We implement FedViT in PyTorch and extensively evaluate it against state-of-the-art techniques using popular federated continual learning benchmarks. Extensive evaluation results on heterogeneous edge devices show that FedViT improves model accuracy by 88.61% without increasing model training time, reduces communication cost by 61.55%, and achieves more improvements under difficult scenarios such as large numbers of tasks or clients, and training different complex ViT models.

Cross-User Similarities in Viewing Behavior for 360° Video and Caching Implications

The demand and usage of 360° video services are expected to increase. However, despite these services being highly bandwidth intensive, not much is known about the potential value that basic bandwidth saving techniques such as server or edge-network on-demand caching (e.g., in a CDN) could have when used for delivery of such services. This problem is both important and complicated as client-side solutions have been developed that split the full 360° view into multiple tiles, and adapt the quality of the downloaded tiles based on the user’s expected viewing direction and bandwidth conditions. This article presents new trace-based analysis methods that incorporate users’ viewports (the area of the full 360° view the user actually sees), a first characterization of the cross-user similarities of the users’ viewports, and a trace-based analysis of the potential bandwidth savings that caching-based techniques may offer under different conditions. Our analysis takes into account differences in the time granularity over which viewport overlaps can be beneficial for resource saving techniques, compares and contrasts differences between video categories, and accounts for uncertainties in the network conditions and the prediction of the future viewing direction when prefetching. The results provide substantial insight into the conditions under which overlap can be considerable and caching effective, and inform the design of new caching system policies tailored for 360° video.

Learning-aided client association control for high-density WLANs

As wireless local area network (WLAN) continues to become popular, there is an increasing number of clients with huge data traffic demands. Especially, some high client-density environments are emerging, such as industrial plants, stadiums, and event centers, which poses significant challenges in terms of client association control. Under such environments, conventional client-side solutions that select access points (APs) according to simple indicators such as signal strength may result in poor network performance, and although some centralized association control mechanisms are proposed, it is still difficult that a large amount of complex global network status information needs to be effectively and efficiently utilized. To meet these challenges, we investigate the online centralized association control problem that aims to improve user quality of experience (QoE), and propose a deep reinforcement learning (DRL) aided solution, called Wi-OAC, where an image-like state pattern is designed to achieve state reformulation for deep Q-network (DQN), and the double DQN and dueling DQN strategies are combined to improve convergence speed. On the basis of offline training, Wi-OAC can determine the proper AP-client associations for the arriving clients. Both simulation experiments and real-world experiments have been conducted to validate the effectiveness of Wi-OAC. In real-world experiments, we build a Wi-OAC testbed including 3 APs and 54 clients in less than 10.5 m2 area, and the results show that Wi-OAC can significantly improve the performance on the client throughput, AP load balancing and user QoE.

FakeAP Detector: An Android-Based Client-Side Application for Detecting Wi-Fi Hotspot Spoofing

Network spoofing is becoming a common attack in wireless networks. The trend is going high due to an increase in Internet users. Similarly, there is a rapid growth of numbers in mobile devices in the working environments and on most official occasions. The trends pose a huge threat to users since they become the prime target of attackers. More unfortunately, mobile devices have weak security measures due to their limited computational powers. Current approaches to detect spoofing attacks focus on personal computers and rely on the network hosts’ capacity, leaving guest users with mobile devices at risk. Some approaches on Android-based devices demand root privilege, which is highly discouraged. This paper presents an Android-based client-side solution to detect the presence of fake access points in a perimeter using details collected from probe responses. Our approach considers the difference in security information and signal level of an access point (AP). We present the detection in three networks, (i) open networks, (ii) closed networks and (iii) networks with captive portals. As a departure from existing works, our solution does not require root access for detection, and it is developed for portability and better performance. Experimental results show that our approach can detect fake access points with an accuracy of 99% and 99.7% at an average of 24.64 and 7.78 milliseconds in open and closed networks, respectively.

A solution to detect the existence of a malicious rogue AP

A malicious rogue AP works like an evil twin; however, instead of using a good twin to connect to the Internet, a malicious rogue AP uses a 3G/4G mobile network to connect to the Internet. While administrators have sufficient information to distinguish rogue APs, it is difficult for client users to know whether they are using a wireless network with malicious an AP. To solve evil twin problems at client-side, many solutions make their detection based on some time metrics or evil twin features. However, time metrics may be influenced by pre-fetching, network topology, traffic volume, or network types. And the evil twin features such as packet forwarding cannot distinguish malicious rogue APs because they behave just like a legitimate AP. To solve above problem, this paper proposes an active user-side solution, called Wi-Fi Malicious Rogue AP Finder (RAF). RAF can be installed in any computer or laptop without any special requirement. RAF detect the existence of a malicious rogue AP based on different reverse traceroute information collected by a remote server. To the best of our knowledge, RAF is the first one client-side solution which could detect malicious rogue APs based on path information but not time metrics.

A machine learning based approach for phishing detection using hyperlinks information

This paper presents a novel approach that can detect phishing attack by analysing the hyperlinks found in the HTML source code of the website. The proposed approach incorporates various new outstanding hyperlink specific features to detect phishing attack. The proposed approach has divided the hyperlink specific features into 12 different categories and used these features to train the machine learning algorithms. We have evaluated the performance of our proposed phishing detection approach on various classification algorithms using the phishing and non-phishing websites dataset. The proposed approach is an entirely client-side solution, and does not require any services from the third party. Moreover, the proposed approach is language independent and it can detect the website written in any textual language. Compared to other methods, the proposed approach has relatively high accuracy in detection of phishing websites as it achieved more than 98.4% accuracy on logistic regression classifier.

Supporting Mobile VR in LTE Networks

Mobile virtual reality (VR) headsets (e.g., Google Cardboard and Samsung Gear VR) seek to offer "anytime, anywhere" panorama, immerse 3D experiences for users. In this work, we study the viability of supporting mobile VR over operational 4G LTE networks, where the device provides pose information to the edge servers to offload graphical processing. We find that, contrary to common perceptions, wireless bandwidth is not the latency bottleneck for medium-quality VR. Instead, the signaling operations, which facilitate wireless data delivery, constitute a bulk portion of the latency. We report findings that challenge five common beliefs on VR network latency in LTE under both static and mobile scenarios, and quantify their impact. We design LTE-VR, a client-side solution to medium-quality VR over LTE. LTE-VR leverages cross-layer design and rich side channel information to reduce various latency sources in the signaling operations. Our prototype evaluation has confirmed its viability in 4G LTE. We discuss its applicability to the upcoming 5G.

Q-FDBA: improving QoE fairness for video streaming

Multiplayer video streaming scenario can be seen everywhere today as the video traffic is becoming the “killer” traffic over the Internet. The Quality of Experience fairness is critical for not only the users but also the content providers and ISP. Consequently, a QoE fairness adaptive method of multiplayer video streaming is of great importance. Previous studies focus on client-side solutions without network global view or network-assisted solution with extra reaction to client. In this paper, a pure network-based architecture using SDN is designed for monitoring network global performance information. With the flexible programming and network mastery capacity of SDN, we propose an online Q-learning-based dynamic bandwidth allocation algorithm Q-FDBA with the goal of QoE fairness. The results show the Q-FDBA could adaptively react to high frequency of bottleneck bandwidth switches and achieve better QoE fairness within a certain time dimension.

On Choosing Server- or Client-Side Solutions for BFT

Byzantine Fault Tolerant (BFT) protocols have the ability to work correctly even when up to a threshold f of system servers are compromised. This makes them appealing for the construction of critical systems connected to the Internet, which are constantly a target for cyber attacks. BFT protocols differ based on the kind of application, deployment settings, performance, access control mechanisms, number of servers in the system, and protocol implementation. The large number of protocols present in the literature and their differences make it difficult for a system builder to choose the solution that best satisfies the requirements of the system that he wants to build. In particular, the main difference among BFT protocols lies in their system models: server-side versus client-side. In the server-side model each client relies on the system to consistently order and replicate updates, while in the client-side model each client actively participates in the protocol. In this article, we classify BFT protocols as server-side or client-side. We analyze the trade-offs between the two models, describe systems that use these models and the trade-offs they choose, highlight the research gaps, and provide guidelines to system builders in order to choose the solution that best satisfies their needs.

Minimizing Communication Interruptions Using Smart Proactive Channel Scanning Over IEEE 802.11 WLANs

Wireless connectivity is becoming pervasive in our lives as wireless networks and mobile devices have been undergoing a magnificent evolution. In this context, consumers demand to access to the wireless communication services anytime and anywhere with high performance. IEEE 802.11 wireless local area networks (WLANs) have a direct impact on this widespread use. Nevertheless, as any standard procedure, WLAN also has its own drawbacks. In IEEE 802.11 wireless networks, periodic channel scanning is an essential procedure to discover available access points in the vicinity and to achieve fast handover. However, this procedure leads to unnecessary overhead in wireless networks and also interrupts ongoing communications of stations. In this paper, we propose a smart proactive channel scanning scheme that allows stations to achieve fast handover with minimized communication interruptions. In the proposed scheme, periodic channel scanning is scheduled based on adaptive channel scanning intervals and activated after a frame transmission only if the mobile station is on move or the received signal strength indicator value drops below a pre-specified threshold. The proposed scheme is a client-side software solution that any station can benefit from, without changing the standard in any existing IEEE 802.11 environment. Analytical and simulation results show that the proposed channel scanning scheme reduces the channel scanning overhead dramatically, minimizes communication interruptions and improves the overall throughput of stations.

A Practical PSM Scheme for Varying Server Delay

Today, many individuals use smartphones or other battery-powered mobile devices equipped with Wi-Fi to access the Internet. Since the network interface cards (NICs) often use a substantial portion of available energy, schemes such as the 802.11 power saving mode (PSM) have been used to limit the amount of time a NIC is awake in an operable state so as to extend battery life. However, since the NIC cannot retrieve packets while sleeping, PSM may negatively impact packet delay. These delays are also unbounded and can significantly increase communication time, particularly in a mobile environment where server delays may change drastically. To mitigate this, we present PSM-AW: a PSM with adaptive wake-up. PSM-AW is a client-side solution that allows the client device to sleep for a maximum time interval while still keeping a tight bound on performance. We present and prove a bound on performance using PSM-AW and demonstrate its effectiveness through extensive simulations and experiments.

InCHlib - interactive cluster heatmap for web applications.

BackgroundHierarchical clustering is an exploratory data analysis method that reveals the groups (clusters) of similar objects. The result of the hierarchical clustering is a tree structure called dendrogram that shows the arrangement of individual clusters. To investigate the row/column hierarchical cluster structure of a data matrix, a visualization tool called ‘cluster heatmap’ is commonly employed. In the cluster heatmap, the data matrix is displayed as a heatmap, a 2-dimensional array in which the colour of each element corresponds to its value. The rows/columns of the matrix are ordered such that similar rows/columns are near each other. The ordering is given by the dendrogram which is displayed on the side of the heatmap.ResultsWe developed InCHlib (Interactive Cluster Heatmap Library), a highly interactive and lightweight JavaScript library for cluster heatmap visualization and exploration. InCHlib enables the user to select individual or clustered heatmap rows, to zoom in and out of clusters or to flexibly modify heatmap appearance. The cluster heatmap can be augmented with additional metadata displayed in a different colour scale. In addition, to further enhance the visualization, the cluster heatmap can be interconnected with external data sources or analysis tools. Data clustering and the preparation of the input file for InCHlib is facilitated by the Python utility script inchlib_clust.ConclusionsThe cluster heatmap is one of the most popular visualizations of large chemical and biomedical data sets originating, e.g., in high-throughput screening, genomics or transcriptomics experiments. The presented JavaScript library InCHlib is a client-side solution for cluster heatmap exploration. InCHlib can be easily deployed into any modern web application and configured to cooperate with external tools and data sources. Though InCHlib is primarily intended for the analysis of chemical or biological data, it is a versatile tool which application domain is not limited to the life sciences only.Electronic supplementary materialThe online version of this article (doi:10.1186/s13321-014-0044-4) contains supplementary material, which is available to authorized users.

Hybrid Client Side Phishing Websites Detection Approach

Phishing tricks to steal personal or credential information by entering victims into a forged website similar to the original site, and urging them to enter their information believing that this site is legitimate. The number of internet users who are becoming victims of phishing attacks is increasing beside that phishing attacks have become more sophisticated. In this paper we propose a client-side solution to protect against phishing attacks which is a Firefox extension integrated as a toolbar that is responsible for checking whether recipient website is trusted or not by inspecting URLs of each requested webpage. If the site is suspicious the toolbar is going to block it. Every URL is evaluated corresponding to features extracted from it. Three heuristics (primary domain, sub domain, and path) and Naïve Bayes classification using four lexical features combined with page ranking received from two different services (Alexa, and Google page rank) used to classify URL. The proposed method requires no server changes and will prevent internet users from fraudulent sites especially from phishing attacks based on deceptive URLs. Experimental results show that our approach can achieve 48% accuracy ratio using a test set of 246 URL, and 87.5% accuracy ratio by excluding NB addition tested over 162 URL.

Cross Site Scripting– Client Side Solution

Cross site scripting (XSS) is a common security problem of web applications where an attacker can inject scripting code into the output of the application that is then sent to a user's web browser. In the web browser, this scripting code is executed and used to transfer sensitive data to a third-party. Today's solutions attempt to prevent XSS on the server side, for example, by inspecting and modifying the data sent to and from the web application. The presented solution, on the other hand, stops XSS attacks on the browser/ client side by tracking the use of sensitive information in the JavaScript engine of the web browser. If sensitive information is about to be transferred to a third party, the user can decide if this should be allowed or not. The existing client- side solutions degrade the performance of client's system resulting in a poor web surfing experience. In this project provides a client side solution that uses a step by step approach to protect cross site scripting, without degrading much the user's web browsing experience.

Taking the PubChem web sketcher to the next level

The input of chemical structures in the context of Web applications is a well-known interface problem in chemical information processing. Currently, this is typically achieved by using either Java applets, or platform- and browser-specific plug-ins - both approaches require installation of custom software components which may not be possible in many use cases. Pure client-side JavaScript solutions have recently been presented, but these require the latest generation of browsers, and are significantly limited in the functionality they can provide. The first popular Web-bases structure sketcher to break out of these moulds was the PubChem structure input tool. It is based on a server-based image streaming model, backward-compatible down to IE6 in a fully browser- and platform-agnostic fashion, and usable without any installation of client software. This software has been in continuous development since it was first deployed about five years ago. The latest version, which still maintains exceptional legacy browser compatibility, has evolved into an information hub which can directly query dozens of Internet chemistry databases. It now supports advanced HTML5 features, such as drag&drop of structure data files and pasting of ChemDraw or ISIS/Draw data on the clipboard. An innovative operation mode for use on touchscreens with limited resolution and no mouse, and voice control for structure input on small devices or for handicapped users are other recently added features.

Efficient Querying of Distributed RDF Sources in Mobile Settings based on a Source Index Model

Recent technological developments in the area of mobile devices in combination with the ubiquity of mobile connectivity have turned handheld devices into rich mobile web clients. On the other hand, we witness an increased number of web resources that are no longer only available in HTML but also in RDF format. In order to access relevant information in a mobile setting, we often want to query a subset of this large amount of heterogeneous RDF resources based on various context factors. We present a semantic technology-based client-side solution for efficient querying of large sets of distributed RDF sources without query endpoints. Our solution continuously extracts metadata from distributed RDF sources and manages this metadata in a local Source Index Model (SIM). When a new query has to be processed, the SIM is consulted to identify, assemble and query only potentially relevant resources. We do not plan to replace existing mobile query engines, but rather build on them to efficiently and transparently query large sets of distributed online RDF sources. The evaluation of our enhanced query handler reveals significant improvements in the overall query execution time based on the proposed SIM.

Client-side cross-site scripting protection

Web applications are becoming the dominant way to provide access to online services. At the same time, web application vulnerabilities are being discovered and disclosed at an alarming rate. Web applications often make use of JavaScript code that is embedded into web pages to support dynamic client-side behavior. This script code is executed in the context of the user's web browser. To protect the user's environment from malicious JavaScript code, browsers use a sand-boxing mechanism that limits a script to access only resources associated with its origin site. Unfortunately, these security mechanisms fail if a user can be lured into downloading malicious JavaScript code from an intermediate, trusted site. In this case, the malicious script is granted full access to all resources (e.g., authentication tokens and cookies) that belong to the trusted site. Such attacks are called cross-site scripting (XSS) attacks. In general, XSS attacks are easy to execute, but difficult to detect and prevent. One reason is the high flexibility of HTML encoding schemes, offering the attacker many possibilities for circumventing server-side input filters that should prevent malicious scripts from being injected into trusted sites. Also, devising a client-side solution is not easy because of the difficulty of identifying JavaScript code as being malicious. This paper presents Noxes, which is, to the best of our knowledge, the first client-side solution to mitigate cross-site scripting attacks. Noxes acts as a web proxy and uses both manual and automatically generated rules to mitigate possible cross-site scripting attempts. Noxes effectively protects against information leakage from the user's environment while requiring minimal user interaction and customization effort.

A New Technology for Interactive Online Mapping with Vector Markup and XML

As Internet cartography matures from static map images to interactive and animated maps, and embraces extensive GIS functionality, the limitations of presenting Web maps as image files become obvious. In this paper, a new technology for Internet cartography is demonstrated that uses direct vector rendering in a browser to create highly interactive virtual maps from distributed sources of geographic data. This technology is made possible by the advent of XML (eXtensible Markup Language) and XML applications for 2D vector rendering such as VML (Vector Markup Language) and SVG (Scalable Vector Graphics). AXIOMAP – Application of XML for Interactive Online Mapping – is a Web map publishing kit and a customizable virtual map interface that allows for the display and manipulation of multiple point, line and area layers, database query, choropleth mapping, hyperlinking, map labeling and annotation. To render maps in a Web browser (Internet Explorer 5, in the current version), AXIOMAP generates VML shapes “on the fly” from XML-encoded geographic data that can physically reside on different servers. A thin client-side solution, AXIOMAP provides for better interactivity than traditional map serverbased approaches. The paper explains the functionality of AXIOMAP, the technology behind it, and presents several applications. A free version of the software can be downloaded from www.elzaresearch.com/landv/.