Search

Fraud detection and prevention has received a lot of attention from the research community due to its high impact on financial institutions' revenues and reputation. The increased use of the web and the provision of online services open up the pathway for exposing these systems to numerous threats and jeopardizing their effective functioning. Naturally, financial frauds are increased in number and form imposing various requirements for their efficient and immediate detection. These requirements are related to the performance of the adopted models as well as the timely response of the decision-making mechanism. Machine learning and data mining are two research domains that can provide a number of techniques/algorithms for fraud detection and setup the road for mitigation actions. However, these methods still need to be improved with respect to the detection of unknown fraud patterns and the incorporation of big data processing mechanisms. This paper presents our attempt to build a hybrid system, i.e., a sequential scheme for combining two deep learning models and efficiently detecting potential financial frauds. We elaborate on the combination of an autoencoder and a Long Short-Term Memory Recurrent Neural Network trained upon datasets which are processed through the use of an oversampling technique. Oversampling is adopted to handle heavily imbalanced datasets which is the 'natural' scenario due to the limited number of frauds compared to the humongous volumes of transactions. The proposed approach tends to capture much more fraud events in comparison with other conventional ML techniques. Our experimental evaluation exposes that our model exhibits a good performance in terms of recall and precision. [ABSTRACT FROM AUTHOR]

The internet of things (IoT) offers a vast infrastructure where numerous devices interact to collect data or perform simple processing activities. These devices are, usually, equipped with sensors, software and storage capabilities being able to process the collected data and exchange knowledge with their peers. Data and knowledge may be transferred, in an upwards mode, to the edge infrastructure and Cloud to be the subject of more complex processing. Advanced processing may be also realized upon clusters of IoT devices or edge nodes, i.e., a set of interconnected devices/nodes.Such group of nodes may work to achieve a common goal, thus, we may desire to rely on their collected data to support activities like federated learning.A training process, when necessary, may be delivered across the cluster of nodes to conclude the view of the group. In this paper, we focus on the development of a monitoring process upon the collected data and a grouping method upon multiple datasets formulated by IoT devices (hosted by edge nodes) in order to deliver the most similar ones. Our vision is to continuously identify similar datasets to support knowledge extraction through the adoption of techniques like the aforementioned federated learning. We utilize a correlation detection method combined with a probabilistic model to conclude the most similar datasets. [ABSTRACT FROM AUTHOR]

The Internet of Things offers a vast infrastructure where numerous devices interact to collect data or perform processing activities (tasks). These devices are, usually, equipped with sensors, software and storage capabilities being able to process the collected data. Task scheduling in edge computing environments has gained considerable attention lately due to the fact that the Edge computing provides lower latency compared to the Cloud. The main challenge is to find a way to maximize the utilization of limited resources available in the edge compared to the Cloud and minimize response time. Many research efforts have been published in order to overcome this challenge. The main limitation of these efforts is the fact that they do not account for task requirements or task correlation that originates from these requirements. In this paper, we focus on the development of a mechanism that utilizes correlation between tasks and takes task requirements into consideration in order to provide efficient task scheduling. Our vision is to minimize task failures and maximize resource utilization with great benefits for the efficient management of the limited resources. [ABSTRACT FROM AUTHOR]

Pervasive Computing (PC) opens up the room for the adoption of devices very close to end users that gives the opportunity to interact with them and execute various applications to facilitate their every day activities. Pervasive applications are supported by the evolution of the Internet of Things (IoT) as well as the Edge Computing (EC) that offer vast infrastructures where data can be collected and processed. EC acts as the mediator between the IoT and the Cloud becoming the middle point where data are transferred before they become the subject of processing by Cloud services. IoT devices can assist in the collection of data and EC nodes could play the role of intermediate processing points executing the desired tasks requested by applications. Any processing can be affected by the presence of missing values that may jeopardize the quality of the outcomes. In this paper, we propose a setting where EC nodes play the aforementioned processing role for data reported by IoT devices and adopt an ensemble scheme for data imputation in the case where missing values are present. Our model relies on the local view of the IoT devices reporting a data vector with a missing value, the view of the group that consists of the IoT nodes with a high similarity in the reported data and a probabilistic approach that reveals the statistics of data as realized in the group of similar reports. The proposed scheme continuously detects the correlation between the incoming data streams and efficiently combines the available data vectors before it is in a position to suggest replacements for missing values. The envisioned aggregation mechanism is capable of resulting the appropriate replacements aligned with the aforementioned views on the collected data. Our ensemble model relies on a number of similarity metrics and statistics to derive the final outcome. The paper reports on the description of the proposed model and elaborates on its validation based on various evaluation scenarios. [ABSTRACT FROM AUTHOR]

The banking sector is on the eve of a serious transformation and the thrust behind it is artificial intelligence (AI). Novel AI applications have been already proposed to deal with challenges in the areas of credit scoring, risk assessment, client experience and portfolio management. One of the most critical challenges in the aforementioned sector is fraud detection upon streams of transactions. Recently, deep learning models have been introduced to deal with the specific problem in terms of detecting and forecasting possible fraudulent events. The aim is to estimate the unknown distribution of normal/fraudulent transactions and then to identify deviations that may indicate a potential fraud. In this paper, we elaborate on a novel multistage deep learning model that targets to efficiently manage the incoming streams of transactions and detect the fraudulent ones. We propose the use of two autoencoders to perform feature selection and learn the latent data space representation based on a nonlinear optimization model. On the delivered significant features, we subsequently apply a deep convolutional neural network to detect frauds, thus combining two different processing blocks. The adopted combination has the goal of detecting frauds over the exposed latent data representation and not over the initial data. [ABSTRACT FROM AUTHOR]

Context-aware data management becomes the focus of several research efforts, which can be placed at the intersection between the Internet of Things (IoT) and Edge Computing (EC). Huge volumes of data captured by IoT devices are processed in EC environments. Even if edge nodes undertake the responsibility of data management tasks, they are characterized by limited storage and computational resources compared to Cloud. Apparently, this mobilises the introduction of intelligent data selection methods capable of deciding which of the collected data should be kept locally based on end users/applications requests. In this paper, we devise a mechanism where edge nodes learn their own data selection filters, and decide the distributed allocation of newly collected data to their peers and/or Cloud once these data are not conformed with the local data filters. Our mechanism intents to postpone final decisions on data transfer to Cloud (e.g., data centers) to pervasively keep relevant data as close and as long to end users/applications as possible. The proposed mechanism derives a data-selection map across edge nodes by learning specific data sub-spaces, which facilitate the placement of processing tasks (e.g., analytics queries). This is very critical when we target to support near real time decision making and would like to minimize all parts of the tasks allocation procedure. We evaluate and compare our approach against baselines and schemes found in the literature showcasing its applicability in pervasive edge computing environments. [ABSTRACT FROM AUTHOR]

Internet of Things offers the infrastructure for smooth functioning of autonomous context-aware devices being connected towards the Cloud. Edge Computing (EC) relies between the IoT and Cloud providing significant advantages. One advantage is to perform local data processing (limited latency, bandwidth preservation) with real time communication among IoT devices, while multiple nodes become hosts of the collected data (reported by IoT devices). In this work, we provide a mechanism for the exchange of data synopses (summaries of extracted knowledge) among EC nodes that are necessary to give the knowledge on the data present in EC environments. The overarching aim is to intelligently decide on when nodes should exchange data synopses in light of efficient execution of tasks. We enhance such a decision with a stochastic optimization model based on the Theory of Optimal Stopping. We provide the fundamentals of our model and the relevant formulations on the optimal time to disseminate data synopses to network edge nodes. We report a comprehensive experimental evaluation and comparative assessment related to the optimality achieved by our model and the positive effects on EC. [ABSTRACT FROM AUTHOR]

Data management at the edge of the network can increase the performance of applications as the processing is realized close to end users limiting the observed latency in the provision of responses. A typical data processing involves the execution of queries/tasks defined by users or applications asking for responses in the form of analytics. Query/task execution can be realized at the edge nodes that can undertake the responsibility of delivering the desired analytics to the interested users or applications. In this paper, we deal with the problem of allocating queries to a number of edge nodes. The aim is to support the goal of eliminating further the latency by allocating queries to nodes that exhibit a low load and high processing speed, thus, they can respond in the minimum time. Before any allocation, we propose a method for estimating the computational burden that a query/task will add to a node and, afterwards, we proceed with the final assignment. The allocation is concluded by the assistance of an ensemble similarity scheme responsible to deliver the complexity class for each query/task and a probabilistic decision making model. The proposed scheme matches the characteristics of the incoming queries and edge nodes trying to conclude the optimal allocation. We discuss our mechanism and through a large set of simulations and the adoption of benchmarking queries, we reveal the potentials of the proposed model supported by numerical results. [ABSTRACT FROM AUTHOR]

Nowadays, the increased amount of users’ devices produce huge volumes of data that should be efficiently managed by modern applications. Streams are adopted to deliver data that, usually, are stored into a number of partitions. Splitting the data offers a lot of advantages as applications can process them in parallel, thus, they increase the speed of processing. Progressive analytics are also adopted to deliver partial responses, during processing, thus, saving time in the execution of applications. Data exploration and analytics queries are very significant for future applications. Usually, such queries demand for an ordered set of objects as a response and require intelligent predictive schemes to deliver the responses on top of the partial results retrieved by the distributed data partitions. A finite set of query processors are adopted to produce these partial results. Processors are placed in front of each partition and report progressive analytics to a central entity. In this paper, we envision the query controller (QC) as the central entity that collects progressive analytics and return the final response to users/applications. The QC receives partial ordered sets of objects and aggregates them to derive the final outcome. We focus on a QC that applies time-optimized techniques and aggregation operators to deliver every response, i.e., ordered sets, over streams of partial ordered subsets. We perform a comprehensive performance assessment with synthetic data and report on the performance of the QC. Our experimental evaluation reveals the pros and cons of the proposed model and a comparison assessment places this paper in the respective literature. [ABSTRACT FROM AUTHOR]

The involvement of numerous devices and data sources in the current form of Web leads to the collection of vast volumes of data. The advent of the Internet of Things (IoT) enhances the devices to act autonomously and transforms them into information and knowledge producers. The vast infrastructure of the Web/IoT becomes the basis for producing data either in a structured or in an unstructured way. In this paper, we focus on a distributed scheme for securing the quality of data as collected and stored in multiple partitions. A high quality is achieved through the adoption of a model that identifies any change in the accuracy of the collected data. The proposed scheme determines if the incoming data negatively affect the accuracy of the already present datasets and when this is the case, it excludes them from further processing. We are based on a scheme that also identifies the appropriate partition where the incoming data should be allocated. We describe the proposed scheme and present simulation and comparison results that give insights on the pros and cons of our solution. [ABSTRACT FROM AUTHOR]

Abstract The Internet of Things (IoT) offers a vast infrastructure of numerous interconnected devices capable of communicating and exchanging data. Pervasive computing applications can be formulated on top of the IoT involving nodes that can interact with their environment and perform various processing tasks. Any task is part of intelligent services executed in nodes or the back end infrastructure for supporting end users' applications. In this setting, one can identify the need for applying updates in the software/firmware of the autonomous nodes. Updates are extensions or patches significant for the efficient functioning of nodes. Legacy methodologies deal with centralized approaches where complex protocols are adopted to support the distribution of the updates in the entire network. In this paper, we depart from the relevant literature and propose a distributed model where each node is responsible to, independently, initiate and conclude the update process. Nodes monitor a set of metrics related to their load and the performance of the network and through a time-optimized scheme identify the appropriate time to conclude the update process. We report on an infinite horizon optimal stopping model on top of the collected performance data. The aim is to make nodes capable of identifying when their performance and the performance of the network are of high quality to efficiently conclude the update process. We provide specific formulations and the analysis of the problem while extensive simulations and a comparison assessment reveal the advantages of the proposed solution. Highlights • We propose an time optimized, performance-aware scheme for software updates management. • The mechanism decides when an update process should be activated. • We offer a method for deriving the optimal time for initiating the update process. • With the proposed mechanism, IoT nodes enjoy the best possible performance. • We offer extensive simulations and an analysis of the results. [ABSTRACT FROM AUTHOR]

Data stream monitoring provides the basis for building intelligent context-aware applications over contextual data streams. A number of wireless sensors could be spread in a specific area and monitor contextual parameters for identifying various phenomena, e.g., fire or flood. A back-end system receives measurements and derives decisions for possible abnormalities related to negative effects. We propose a mechanism which, based on multivariate sensors data streams, provides real-time identification of phenomena. The proposed framework performs contextual information fusion over consensus theory for the efficient measurements aggregation while time-series prediction is adopted to result future insights on the aggregated values. The unanimous fused and predicted pieces of context are fed into a type-2 fuzzy inference system to derive highly accurate identification of events. The type-2 inference process offers reasoning capabilities under the uncertainty of the phenomena identification. We provide comprehensive experimental evaluation over real contextual data and report on the advantages and disadvantages of the proposed mechanism. Our mechanism is further compared with type-1 fuzzy inference and other mechanisms to demonstrate its false alarms minimization capability. [ABSTRACT FROM PUBLISHER]

Cooperative problem solving has attracted great interest in the research community. A number of nodes can cooperate to reach a common goal which is the solution to a specific problem. Through a team effort, nodes try to achieve the best possible result. Each node undertakes the responsibility to fulfill a specific task (part of a complex plan towards the solution of the problem) and return the final outcome to a coordinator. The coordinator assigns tasks to nodes and, accordingly, collects the results. In this paper, we propose the adoption of fuzzy logic (FL) in the decision making process of each node. Nodes adopt the provided (by the coordinator) FL knowledge base that indicates the appropriate actions during the problem solving process. This knowledge base is updated during the execution of the assigned task to be fully aligned with the environment characteristics. Partial knowledge bases experienced by the nodes are sent back to the coordinator and are aggregated to generate a 'global' knowledge base that incorporates the experience retrieved by the team. We describe the aggregation process and propose two models: the first adopts the immediate distribution of the aggregated FL rule base to the active nodes and the second indicates the future use of the aggregated FL rule base. Our evaluation involves the realization of the proposed framework in a specific research domain as well as numerical results retrieved by a large number of simulations. Our results show that there is a trade-off between the two proposed models concerning the quality of the final solution and the time required to retrieve the final outcome. [ABSTRACT FROM AUTHOR]

The rapid growth of the Web means that humans become increasingly incapable of searching among millions of resources to find and purchase items. Autonomous entities such as agents could help in these situations. Electronic markets (EMs) are virtual sites where these autonomous entities can interact to exchange items and obtain specific returns. In this study, we consider the interactions between buyers and sellers in EMs, where we focus specifically on the buyer side. These interactions can be modeled as finite horizon negotiations. However, the buyer cannot be certain of the characteristics of the seller during negotiations (incomplete knowledge). Thus, to address this uncertainty, we propose a fuzzy logic (FL) system that is responsible for determining the appropriate actions of the buyer during every negotiation round. We also propose an adaptation technique that updates the FL rule base and system membership functions as necessary. Using this approach, the system can respond to even the complex strategies followed by a seller. A seller strategy estimation method is also adopted by the system, which employs the known kernel density estimator (KDE). We provide results for a large number of negotiations and compare our system with previous research in this area. Our results show that the proposed system exhibits good performance in many negotiation scenarios. [ABSTRACT FROM AUTHOR]

Research society has developed a number of models and tools to support emergency management. The proposed models are mainly designed for indoor applications oriented to provide guidance directly to people in danger. Only a few of them deal with outdoor scenarios as well as with providing directions to field commanders or rescue teams. Additionally, load balancing techniques for the optimal allocation of a number of entities into a number of resources are understudied creating a gap in the corresponding research. In this paper, we propose a load balancing model oriented to assist field commanders and rescue teams in a post-emergency scenario. The proposed system could be applied either for indoor or outdoor applications. The module builds on top of the solution provided for the known Santa Fe Bar Problem (SFBP). It consists of an intelligent technique aiming to distribute a number of entities into a finite number of resources. A set of predictors undertake the responsibility of estimating the load of each resource. These predictors are adopted to select the appropriate resource for each entity. A case study deals with the distribution of injured persons into a number of hospitals and presents the functionality of the proposed module. Finally, numerical results reveal computational and time requirements of our system. [ABSTRACT FROM AUTHOR]

A Recommendation System (RS) aims to deliver meaningful recommendations to users for items (e.g., music and books), which are of high interest to them. We consider an RS which directly communicates with a set of providers in order to access the information of the items (e.g., descriptions), rate them according to the user's preferences, and deliver an Item List (IL). The RS is enhanced with a mechanism, which sequentially observes the rating information (e.g., similarity degree) of the items and decides when to deliver the IL to the user, without exhausting the entire set of providers. Hence, the RS saves time and resources. We propose two mechanisms based on the theory of optimal stopping. Both mechanisms deliver an IL, which sufficiently matches to the user's needs having examined a partial set of items. That is, the number of items in the delivered IL is optimal, producing a high level of user satisfaction, i.e., Quality of Recommendation (QoR). Our simulations reveal the efficiency of the mechanisms and quantify the benefits stemming from their adoption. [ABSTRACT FROM AUTHOR]

Electronic markets are virtual frameworks where entities not known in advance have the opportunity to interact for the trading of products or services. Usually, a negotiation is necessary for the conclusion of the transaction. The conclusion is either positive (agreement) or negative (conflict). An efficient reasoning mechanism is necessary for players participating in negotiations. In this paper, we focus on the buyer side and propose two decision models based on the optimal stopping theory (OST). OST is proved to be very efficient in cases where an entity tries to find the time to stop a process with the aim of maximizing her utility. The outcome of the proposed decision method indicates whether the buyer stops a negotiation either by accepting the offer or continuing in the negotiation by rejecting it. In our models, we assume zero knowledge on the players' characteristics. Our proposed decision models do not require any complex modeling or any information provided by experts. Experimental results reveal the efficiency of each model and provide a comparison assessment with other research efforts. [ABSTRACT FROM AUTHOR]

Abstract: The current form of Web provides numerous product resources available to users. Users can rely on intelligent agents for purchase actions. These actions are taken in specific environments such as Electronic Markets (EMs). In this paper, we study the interaction process between buyers and sellers and focus on the buyer side. Each buyer has the opportunity to interact with a number of sellers trying to buy the most appropriate products. This interaction can be modeled as a finite horizon Bargaining Game (BG). In this game, players have opposite goals concerning the product price. We adopt a number of techniques in the buyer side trying to give the appropriate level of efficiency in the buyer decision process. The buyer uses a prediction mechanism in combination with the use of Fuzzy Logic (FL) theory in order to be able to predict the upcoming seller proposal and, thus, understand the seller pricing policy. Based on this, he/she can adapt his/her behavior when trying to purchase products. The buyer adaptation mechanism produces the belief that the buyer has about the seller pricing policy and a parameter that indicates his/her own pricing policy which yields the buyer offers in the upcoming rounds. Moreover, the buyer is based on FL system that derives the appropriate actions at every round of the BG. Our results show that the combination of Fuzzy Logic (FL) with the above-mentioned techniques provides an efficient decision mechanism in the buyer side that in specific scenarios outperforms an optimal stopping model. [Copyright &y& Elsevier]

The multi-layered negative effects caused by pollutants released into the atmosphere as a result of fires served as the stimulus for the development of a system that protects the health of firefighters operating in the affected area. A collaborative network comprising mobile and stationary Internet of Things (IoT) devices that are furnished with gas sensors, along with a remote server, constructs a resilient framework that monitors the concentrations of harmful emissions, characterizes the ambient air quality of the vicinity where the fire transpires, adopting European Air Quality levels, and communicates the outcomes via suitable applications (RESTful APIs and visualizations) to the stakeholders responsible for fire management decision making. Different experimental evaluations adopting separate contexts illustrate the operation of the infrastructure. [ABSTRACT FROM AUTHOR]

Nowadays, we are witnessing the advent of the Internet of Things (IoT) with numerous devices performing interactions between them or with their environment. The huge number of devices leads to huge volumes of data that demand the appropriate processing. The "legacy" approach is to rely on Cloud where increased computational resources can realize any desired processing. However, the need for supporting real-time applications requires a reduced latency in the provision of outcomes. Edge Computing (EC) comes as the "solver" of the latency problem. Various processing activities can be performed at EC nodes having direct connection with IoT devices. A number of challenges should be met before we conclude a fully automated ecosystem where nodes can cooperate or understand their status to efficiently serve applications. In this article, we perform a survey of the relevant research activities towards the vision of Edge Mesh (EM), i.e., a "cover" of intelligence upon the EC. We present the necessary hardware and discuss research outcomes in every aspect of EC/EM nodes functioning. We present technologies and theories adopted for data, tasks, and resource management while discussing how machine learning and optimization can be adopted in the domain. [ABSTRACT FROM AUTHOR]

Convolution Neural Networks (CNNs) are gaining ground in deep learning and Artificial Intelligence (AI) domains, and they can benefit from rapid prototyping in order to produce efficient and low-power hardware designs. The inference process of a Deep Neural Network (DNN) is considered a computationally intensive process that requires hardware accelerators to operate in real-world scenarios due to the low latency requirements of real-time applications. As a result, High-Level Synthesis (HLS) tools are gaining popularity since they provide attractive ways to reduce design time complexity directly in register transfer level (RTL). In this paper, we implement a MobileNetV2 model using a state-of-the-art HLS tool in order to conduct a design space exploration and to provide insights on complex hardware designs which are tailored for DNN inference. Our goal is to combine design methodologies with sparsification techniques to produce hardware accelerators that achieve comparable error metrics within the same order of magnitude with the corresponding state-of-the-art systems while also significantly reducing the inference latency and resource utilization. Toward this end, we apply sparse matrix techniques on a MobileNetV2 model for efficient data representation, and we evaluate our designs in two different weight pruning approaches. Experimental results are evaluated with respect to the CIFAR-10 data set using several different design methodologies in order to fully explore their effects on the performance of the model under examination. [ABSTRACT FROM AUTHOR]