Phishing remains one of the most prevalent cybersecurity threats, endangering users’ personal data, financial assets and online privacy. Although Machine Learning Phishing Website Detectors (ML-PWDs) are an effective tool for identifying malicious webpages, recent studies have revealed that these models are vulnerable to adversarial attacks. In this study, we present a new adversarial attack strategy capable of operating in the problem space, which uses heuristic search algorithms, including Beam Search, Simulated Annealing and Monte Carlo Tree Search, to generate adversarial samples that evade state-of-the-art detectors while maintaining visual and functional fidelity. Our approach optimizes the trade-off between the number of manipulations and attack success, minimizing the distance from the original sample. Experiments on two public datasets demonstrate that our method reduces the average detection rate from 0.80 to 0.05 on Zenodo and from 0.82 to 0.03 on δ Phish, while requiring up to 70% fewer manipulations than competing attacks. Furthermore, the generated samples remain closer to the originals in the L 0 and L 2 metrics, indicating strong statistical plausibility. These results highlight the effectiveness of our approach in evading ML-PWDs and its potential for evaluating and strengthening the adversarial robustness of real-world detection systems.

Recent advances in distributed vehicle-to-vehicle communication promise to transform the user’s driving ex- perience, providing new services capable of improving safety, efficiency and quality of travelling. Due to the large amount of information exchanged, a major challenge of Vehicular Networks is the adoption of appropri- ate data dissemination protocols that ensure good performance in real-time event detection, while guarantee- ing low communication overhead. To this aim, this paper proposes an adaptive event dissemination algorithm which exploits Population Protocols (PPs) for modelling vehicle interactions as coordinated behaviors of au- tonomous agents in a distributed system. The experimental evaluation performed on realistic vehicle tracks over real-world maps demonstrates the system’s ability to efficiently disseminate information in the network in order to support reliable and distributed event detection services.

The rapid convergence of IoT technologies and artificial intelligence is reshaping university campuses into dynamic, smart environments. Faced with the challenge of managing increasingly complex and heterogeneous data streams, campus communities often struggle to benefit fully from available digital resources and personalized support. In response, this work presents a modular and scalable system designed to provide context-aware recommendations that enhance both academic and social experiences. By integrating data from physical sensors, mobile devices, and external sources, the proposed framework captures rich contextual insights to deliver adaptive, personalized services that address the diverse needs of students, faculty, and administrative staff. Developed as part of the S3 Campus project at the University of Palermo, this system represents a significant step forward in fostering innovative, intelligent campus solutions that are attuned to the evolving demands of modern educational environments.

The increasing availability of smart devices in people's daily lives is constantly driving the design of novel services aimed to support the users by leveraging data provided by sensors embedded in their devices. In this paper, we present a scenario where data generated by wearable devices, such as smartphones and smartwatches, are analyzed to perform Human Activity Recognition (HAR). Given the different nature of the devices, using a single classifier may lead to inconsistent performance, especially for tasks that are semantically complex. Conversely, a distributed approach to activity recognition, where independent classifiers are used on each device, would be more computationally demanding and challenging to maintain. To address these issues, we present a probabilistic data fusion approach to integrate measurements from multiple devices while improving the overall system accuracy. Experiments performed on real data acquired from different devices show the effectiveness of our approach, especially in the recognition of complex activities.

Recommender systems have become pivotal in modern digital platforms, guiding user choices and driving engagement. However, their widespread adoption has also made them a prime target for adversarial attacks, especially data poisoning attacks that subtly manipulate recommendations. Existing approaches often generate unrealistic fake profiles, making them vulnerable to detection by anomaly-based defenses. In this paper, we propose a novel, model-agnostic poisoning framework that combines contrastive learning and reinforcement learning with Proximal Policy Optimization (PPO) to craft highly realistic fake profiles derived from cross-domain user data. By interacting exclusively with a surrogate recommender trained on a compatible domain, our framework identifies and fine-tunes influential user profiles to maximize the impact on a black-box target system. Our experimental evaluation on real-world datasets shows that our approach successfully promotes target items across diverse recommendation models with minimal injection effort, outperforming baseline strategies.

In recent years, the widespread adoption of Machine Learning (ML) at the core of complex IT systems has driven researchers to investigate the security and reliability of ML techniques. A very specific kind of threats concerns the adversary mechanisms through which an attacker could induce a classification algorithm to provide the desired output. Such strategies, known as Adversarial Machine Learning (AML), have a twofold purpose: to calculate a perturbation to be applied to the classifier’s input such that the outcome is subverted, while maintaining the underlying intent of the original data. Although any manipulation that accomplishes these goals is theoretically acceptable, in real scenarios perturbations must correspond to a set of permissible manipulations of the input, which is rarely considered in the literature. In this article, we present AdverSPAM , an AML technique designed to fool the spam account detection system of an Online Social Network (OSN). The proposed black-box evasion attack is formulated as an optimization problem that computes the adversarial sample while maintaining two important properties of the feature space, namely statistical correlation and semantic dependency . Although being demonstrated in an OSN security scenario, such an approach might be applied in other context where the aim is to perturb data described by mutually related features. Experiments conducted on a public dataset show the effectiveness of AdverSPAM compared to five state-of-the-art competitors, even in the presence of adversarial defense mechanisms.

In recent years, Online Social Networks (OSNs) have radically changed the way people communicate. The most widely used platforms, such as Facebook, Youtube, and Instagram, claim more than one billion monthly active users each. Beyond these, news-oriented micro-blogging services, e.g., Twitter, are daily accessed by more than 120 million users sharing contents from all over the world. Unfortunately, legitimate users of the OSNs are mixed with malicious ones, which are interested in spreading unwanted, misleading, harmful, or discriminatory content. Spam detection in OSNs is generally approached by considering the characteristics of the account under analysis, its connection with the rest of the network, as well as data and metadata representing the content shared. However, obtaining all this information can be computationally expensive, or even unfeasible, on massive networks. Driven by these motivations, in this article we propose SpADe, a multi-stage Spam Account Detection algorithm with reject option, whose purpose is to exploit less costly features at the early stages, while progressively extracting more complex information only for those accounts that are difficult to classify. Experimental evaluation shows the effectiveness of the proposed algorithm compared to single-stage approaches, which are much more complex in terms of features processing and classification time.

In recent years, the widespread diffusion of smart pervasive devices able to provide AI-based services has encouraged research in the definition of new distributed learning paradigms. Federated Learning (FL) is one of the most recent approaches which allows devices to collaborate to train AI-based models, whereas guarantying privacy and lower communication costs. Although different studies on FL have been conducted, a general and modular architecture capable of performing well in different scenarios is still missing. Following this direction, this paper proposes a general FL framework whose validity is assessed by considering a distributed activity recognition scenario in which users' personal devices are employed as the basis of the sensing infrastructure. Experimental analysis was performed to evaluate the effectiveness of the architecture as compared with a centralized approach, under different settings. Results demonstrate the versatility and functionality of the proposed solution.

The main goal of Ambient Intelligence (AmI) is to support users in their daily activities by satisfying and anticipating their needs. To achieve such goal, AmI systems rely on physical infrastructures made of heterogenous sensing devices which interact in order to exchange information and perform monitoring tasks. In such a scenario, a full achievement of AmI vision would also require the capability of the system to autonomously check the status of the infrastructure and supervise its maintenance. To this aim, in this paper, we extend some previous works in order to allow the self-management of AmI devices enabling them to directly interact with maintenance service providers. In particular, the combination of smart contracts and blockchains enables AmI systems to autonomously communicate with untrusted entities and complete secure transactions without the brokering of a trusted third party. The proposed approach has been adopted to design a sample AmI application capable of managing requests from faulty devices in a Smart home.

Several issues related to Smart City development require the knowledge of accurate human mobility models, such as in the case of urban development planning or evacuation strategy definition. Nevertheless, the exploitation of real data about users' mobility results in severe threats to their privacy, since it allows to infer highly sensitive information. On the contrary, the adoption of simulation tools to handle mobility models allows to neglect privacy during the design of location-based services. In this work, we propose a simulation tool capable of generating synthetic datasets of human mobility traces; then, we exploit them to evaluate the effectiveness of algorithms which aim to detect Points of Interest visited by users of a Smart Campus. Our simulator exploits an activity-based mobility model, thus it is based on the assumption that mobility of campus users is motivated by the activities they plan to perform. It is capable of simulating the weekly repetitiveness of human behavior and to model different mobility profiles for each day of the week through a fifth-order Markov model.

Online Social Networks (OSNs) have become increasingly popular both because of their ease of use and their availability through almost any smart device. Unfortunately, these characteristics make OSNs also target of users interested in performing malicious activities, such as spreading malware and performing phishing attacks. In this paper we address the problem of spam detection on Twitter providing a novel method to support the creation of large-scale annotated datasets. More specifically, URL inspection and tweet clustering are performed in order to detect some common behaviors of spammers and legitimate users. Finally, the manual annotation effort is further reduced by grouping similar users according to some characteristics. Experimental results show the effectiveness of the proposed approach.

Being part of one of the fastest growing area in Artificial Intelligence (AI), virtual assistants are nowadays part of everyone's life being integrated in almost every smart device. Alexa, Siri, Google Assistant, and Cortana are just few examples of the most famous ones. Beyond these off-the-shelf solutions, different technologies which allow to create custom assistants are available. IBM Watson, for instance, is one of the most widely-adopted question-answering framework both because of its simplicity and accessibility through public APIs. In this work, we present a virtual assistant that exploits the Watson technology to support students and staff of a smart campus at the University of Palermo. Some in progress results show the effectiveness of the approach we propose.

Multi-agent distributed systems are characterized by autonomous entities that interact with each other to provide, and/or request, different kinds of services. In several contexts, especially when a reward is offered according to the quality of service, individual agents (or coordinated groups) may act in a selfish way. To prevent such behaviours, distributed Reputation Management Systems (RMSs) provide every agent with the capability of computing the reputation of the others according to direct past interactions, as well as indirect opinions reported by their neighbourhood. This last point introduces a weakness on gossiped information that makes RMSs vulnerable to malicious agents’ intent on disseminating false reputation values. Given the variety of application scenarios in which RMSs can be adopted, as well as the multitude of behaviours that agents can implement, designers need RMS evaluation tools that allow them to predict the robustness of the system to security attacks, before its actual deployment. To this aim, we present a simulation software for the vulnerability evaluation of RMSs and illustrate three case studies in which this tool was effectively used to model and assess state-of-the-art RMSs.

The diffusion of heterogeneous smart devices capable of capturing and analysing data about users, and/or the environment, has encouraged the growth of novel sensing methodologies. One of the most attractive scenarios in which such devices, such as smartphones, tablet computers, or activity trackers, can be exploited to infer relevant information is human activity recognition (HAR). Even though some simple HAR techniques can be directly implemented on mobile devices, in some cases, such as when complex activities need to be analysed timely, users’ smart devices can operate as part of a more complex architecture. In this article, we propose a multi-device HAR framework that exploits the fog computing paradigm to move heavy computation from the sensing layer to intermediate devices and then to the cloud. As compared to traditional cloud-based solutions, this choice allows to overcome processing and storage limitations of wearable devices while also reducing the overall bandwidth consumption. Experimental analysis aims to evaluate the performance of the entire platform in terms of accuracy of the recognition process while also highlighting the benefits it might bring in smart environments.

In distributed environments, where unknown entities cooperate to achieve complex goals, intelligent techniques for estimating agents' truthfulness are required. Distributed Reputation Management Systems (RMSs) allow to accomplish this task without the need for a central entity that may represent a bottleneck and a single point of failure. The design of a distributed RMS is a challenging task due to a multitude of factors that could impact on its performances. In order to support the researcher in evaluating the RMS robustness against security attacks since its beginning design phase, in this work we present a distributed simulation environment that allows to model both the agent's behaviors and the logic of the RMS itself. Moreover, in order to compare at simulation time the performance of the designed distributed RMS with a baseline obtained by an ideal RMS, we introduce an omniscient process called truth-holder which owns a global knowledge all involved entities. The effectiveness of our platform was proved by a set of experiments aimed at measuring the vulnerability of a RMS to a common set of security attacks.

In this short poster paper, we present an application for suggesting attractions to be visited by users, based on social signal processing techniques.

Over the last 40 years, automatic solutions to analyze text documents collection have been one of the most attractive challenges in the field of information retrieval. More recently, the focus has moved towards dynamic, distributed environments, where documents are continuously created by the users of a virtual community, i.e., the social network. In the case of Twitter, such documents, called tweets, are usually related to events which involve many people in different parts of the world. In this work we present a system for real-time Twitter data analysis which allows to follow a generic event from the user's point of view. The topic detection algorithm we propose is an improved version of the Soft Frequent Pattern Mining algorithm, designed to deal with dynamic environments. In particular, in order to obtain prompt results, the whole Twitter stream is split in dynamic windows whose size depends both on the volume of tweets and time. Moreover, the set of terms we use to query Twitter is progressively refined to include new relevant keywords which point out the emergence of new subtopics or new trends in the main topic. Tests have been performed to evaluate the performance of the framework and experimental results show the effectiveness of our solution.

In this paper, we present a method for recognizing human activities using information sensed by an RGB-D camera, namely the Microsoft Kinect. Our approach is based on the estimation of some relevant joints of the human body by means of the Kinect; three different machine learning techniques, i.e., K-means clustering, support vector machines, and hidden Markov models, are combined to detect the postures involved while performing an activity, to classify them, and to model each activity as a spatiotemporal evolution of known postures. Experiments were performed on Kinect Activity Recognition Dataset, a new dataset, and on CAD-60, a public dataset. Experimental results show that our solution outperforms four relevant works based on RGB-D image fusion, hierarchical Maximum Entropy Markov Model, Markov Random Fields, and Eigenjoints, respectively. The performance we achieved, i.e., precision/recall of 77.3% and 76.7%, and the ability to recognize the activities in real time show promise for applied use.

People make more and more use of digital image acquisition devices to capture screenshots of their everyday life. The growing number of personal pictures raise the problem of their classification. Some of the authors proposed an automatic technique for personal photo album management dealing with multiple aspects (i. e., people, time and background) in a homogenous way. In this paper we discuss a solution that allows mobile users to remotely access such technique by means of their mobile phones, almost from everywhere, in a pervasive fashion. This allows users to classify pictures they store on their devices. The whole solution is presented, with particular regard to the user interface implemented on the mobile phone, along with some experimental results.

In this paper, a new system is presented to support the user in the face annotation task. Every time a photo sequence becomes available, the system analyses it to detect and cluster faces in set corresponding to the same person. We propose to model the problem of people re-identification in photos as a data association problem. In this way, the system takes advantage from the assumption that each person can appear at most once in each photo. We propose a fully automated method for grouping facial images, the method does not require any initialization neither a priori knowledge of the number of persons that are in the photo sequence. We compare the results obtained with our method and with standard clustering methods on three personal collections and on a publicly available dataset.

The increasing number of new applications for mobile devices in pervasive environments, do not cope with changes in the wireless communications. Developers of such applications have to deal with problems arising from the available wireless connections in the given environment. A middleware is a solution that allows to overcome some of these problems. It provides to the applications a set of functions that facilitate their development. In this paper we present a Java-based communication wrapper, called SmartTraffic, which allows programmers to seamlessly use TCP or UDP protocols over Bluetooth or any IP-based wireless network. Developers can use SmartTraffic within their Java applications, thus focusing on the application goals, and leaving out details about how it should interact with the available wireless connection.