• The proposed SWF tokenization method, dynamically capturing emotional subtleties, noise robustness, and contextual information to robustly preserve speaker identity. • A transformative deep-self-supervised spectrogram transformer back-end, outperforming conventional approaches by effectively addressing their inherent limitations in preserving local features without the need for extensive labeled training data. • Complete elimination of dependency on additional speech enhancement, enabling seamless, efficient, and robust end-to-end learning tailored for real-world deployments. Identifying speakers in noisy and emotional conditions remains a significant challenge due to the distortion of spectral cues. This study proposes the Speech Without Filter (SWF) framework, a novel self-supervised learning paradigm that operates directly on raw spectrograms. Theoretically, this research introduces a progressive tokenization mechanism that acts as a structural inductive bias, mimicking the contracting path of a U-Net to preserve local spectro-temporal continuity. Unlike standard fixed-patch Transformers that often smooth over speaker-specific micro-textures, the SWF architecture integrates denoising and feature extraction into a single stage, challenging the traditional decoupled paradigm of speech enhancement and recognition. Using a sample of 1.58 million pre-training instances, the model was evaluated across English (RAVDESS), Arabic (ESD), and stressful (SUSAS) datasets. Results demonstrate significant improvements, with the SWF model achieving 91.01% accuracy in clean conditions and maintaining 88.5% in high-noise cocktail party environments, outperforming state-of-the-art models like WavLM and HuBERT. These findings suggest that architectural innovation in tokenization is as critical as pre-training scale for robust speech processing.

In recent years, malware attacks have become more and more sophisticated, reflecting a radical change in malware behavior. Attackers aim to create malware that, at each execution, generates a different number of independent and cooperating threads. Randomization of malware's division of labor among threads poses significant challenges to traditional detection approaches. In this paper, we demonstrate that attacks based on random division of labor among multiple threads can dramatically degrade the detection performance of five benchmark ML models, in some cases dropping their accuracy to 50% with only a few threads. Then, we propose and evaluate a novel detection technique based on polymorphic-aware training and ensemble learning with ad-hoc voting scheme (favoring minority report ). Results of experimentation carried out on real malware system call logs and assigned to threads via a Bayesian splitting accounting for inter-call dependency indicate that our ensemble has high detection capabilities (99.7% best case), and improves the baseline accuracy of a single model in detecting single-thread malware.

To facilitate the Edge AI paradigm in distributed networks, we propose novel collaborative learning methodologies for a connected network of edge nodes. Our proposed methodologies tackle the challenges in distributed learning where there are constraints on data privacy and a low degree of overlap between the classes observed by the nodes. These approaches entail sharing class distribution information between nodes, computing nodes, and class weights, training local models on each node, then aggregating the models using the determined weights. It favors nodes that have encountered unique or less common classes in their local datasets. Through a series of experiments using an activity recognition dataset, we demonstrate the effectiveness and scalability of our proposed approaches. We show the adaptive nature of the proposed approach by achieving classification accuracy above the baseline, even with little overlap between the observed classes. This study serves as a foundation for future advancements in collaborative learning on edge networks, and encourages the development of scalable solutions.

In this paper, we improve the robustness of Machine Learning (ML) classifiers against training-time attacks by linking the risk of training data being tampered with to the redundancy in the ML model's design needed to prevent it. Our defense mechanism is directly applicable to classifiers' training data, without any knowledge of the specific ML model to be hardened. First, we compute the training data proximity to class separation surfaces, identified via a reference linear model. Each data point is associated with a risk index, which is used to partition the training set by an unsupervised technique. Then, we train a learner for each partition and combine the learners' output in an ensemble. Our method treats the protected ML classifier as a black box and is inherently robust to transfer attacks. Experiments show that, for data poisoning rates between 6 and 25 percent of the training set, our method is more robust compared to benchmarks and to a monolithic version of the model trained on the whole training set. Our results make a convincing case for adopting training set partitioning and ensemble generation as a stage of ML models' development and deployment lifecycle.

In this paper, we propose Value Iteration Network for Reward Shaping (VIN-RS), a potential-based reward shaping mechanism using Convolutional Neural Network (CNN). The proposed VIN-RS embeds a CNN trained on computed labels using the message passing mechanism of the Hidden Markov Model. The CNN processes images or graphs of the environment to predict the shaping values. Recent work on reward shaping still has limitations towards training on a representation of the Markov Decision Process (MDP) and building an estimate of the transition matrix . The advantage of VIN-RS is to construct an effective potential function from an estimated MDP while automatically inferring the environment transition matrix. The proposed VIN-RS estimates the transition matrix through a self-learned convolution filter while extracting environment details from the input frames or sampled graphs. Due to (1) the previous success of using message passing for reward shaping; and (2) the CNN planning behavior, we use these messages to train the CNN of VIN-RS. Experiments are performed on tabular games, Atari 2600 and MuJoCo, for discrete and continuous action space. Our results illustrate promising improvements in the learning speed and maximum cumulative reward compared to the state-of-the-art. The improvement achieved by VIN-RS can only be observed for some of the games due to the underlying nature of some environments. In terms of the studied MuJoCo games, there is on average an increase of 30% in the maximum reward reached during early stages of learning.

Gathering the training malware traces is restive and can be a nuisance depending on the type of malware, such as behavioral polymorphism. Generative Adversarial Networks (GANs) are well suited for these issues because they can generate synthetic data that mimic actual data. This treatise sheds detailed and thorough insights into the GAN model implemented to generate a proper training mechanism for binary classification. This paper tested tabular-based and pictorial-based models in multiple trials to determine the better one for classification. Furthermore, multiple ML-based classification techniques, such as ensemble learning, Support Vector Machines (SVMs), and Linear Regression, were tested and recorded on tabular and pictorial GANs. Tabular-wise, the ∆RMSEdata collected for Random Forest Tree with the Vanilla LeakyReLU-based GAN provided the optimal classification results. Feature interactions and a biological-inspired activation function were considered for optimizing the model. However, they were only additional tests that were not considered part of the leading paper, as testing quantity was insufficient for definitive evidence of optimization.

Malware visualization analysis incorporating with Machine Learning (ML) has been proven to be a promising solution for improving security defenses on different platforms. In this work, we propose an integrated framework for addressing common problems experienced by ML utilizers in developing malware detection systems. Namely, a pictorial presentation system with extensions is designed to preserve the identities of benign/malign samples by encoding each variable into binary digits and mapping them into black and white pixels. A conditional Generative Adversarial Network based model is adopted to produce synthetic images and mitigate issues of imbalance classes. Detection models architected by Convolutional Neural Networks are for validating performances while training on datasets with and without artifactual samples. Result demonstrates accuracy rates of 98.51% and 97.26% for these two training scenarios.

Big data exchange is an emerging problem in the context of big data management and analytics. In big data exchange, multiple entities exchange big datasets beyond the common data integration or data sharing paradigms, mostly in the context of data federation architectures. How to make big data exchange while ensuring privacy preservation constraintsƒ The latter is a critical research challenge that is gaining momentum on the research community, especially due to the wide family of application scenarios where it plays a critical role (e.g., social networks, bio-informatics tools, smart cities systems and applications, and so forth). Inspired by these considerations, in this paper we provide an overview of models and issues in the context of privacy-preserving big data exchange research, along with a selection of future research directions that will play a critical role in next-generation research.

Assuring anomaly-free business process executions is a key challenge for many organizations. Traditional techniques address this challenge using prior knowledge about anomalous cases that is seldom available in real-life. In this work, we propose the usage of word2vec encoding and One-Class Classification algorithms to detect anomalies by relying on normal behavior only. We investigated 6 different types of anomalies over 38 real and synthetics event logs, comparing the predictive performance of Support Vector Machine, One-Class Support Vector Machine, and Local Outlier Factor. Results show that our technique is viable for real-life scenarios, overcoming traditional machine learning for a wide variety of settings where only the normal behavior can be labeled.

The advent of Big Data has revolutionized the way in which data are collected, analyzed, and processed, becoming a pre-requisite for each enterprise that competes in the global market. In this respect, the commodization of Big Data analytics is an essential goal to be faced in the near future. Recently, some preliminary approaches have been presented mostly focusing on distributing Big Data platforms as a service, while less has been done on cross-platform Big Data analytics. In this paper, we propose a model-based methodology for Big Data Analytics-as-a-Service that extends existing techniques by supporting cross-communication between batch and stream processing, deployment on multiple platforms, and end-to-end verification against users' requirements.

The diffusion of service-based and cloud-based systems has created a scenario where software is often made available as services, offered as commodities over corporate networks or the global net. This scenario supports the definition of business processes as composite services, which are implemented via either static or runtime composition of offerings provided by different suppliers. Fast and accurate evaluation of services’ security properties becomes then a fundamental requirement and is nowadays part of the software development process. In this article, we show how the verification of security properties of composite services can be handled by test-based security certification and built to be effective and efficient in dynamic composition scenarios. Our approach builds on existing security certification schemes for monolithic services and extends them towards service compositions. It virtually certifies composite services, starting from certificates awarded to the component services. We describe three heuristic algorithms for generating runtime test-based evidence of the composite service holding the properties. These algorithms are compared with the corresponding exhaustive algorithm to evaluate their quality and performance. We also evaluate the proposed approach in a real-world industrial scenario, which considers ENGpay online payment system of Engineering Ingegneria Informatica S.p.A. The proposed industrial evaluation presents the utility and generality of the proposed approach by showing how certification results can be used as a basis to establish compliance to Payment Card Industry Data Security Standard.

The Big Data and Data Analytics is a brand new paradigm, for the integration of Internet Technology in the human and machine context. For the first time in the history of the human mankind we are able to transforming raw data that are massively produced by humans and machines in to knowledge and wisdom capable of supporting smart decision making, innovative services, new business models, innovation, and entrepreneurship. For the Web Science research, this is a new methodological and technological spectrum of advanced methods, frameworks and functionalities never experienced in the past. At the same moment communities out of web science need to realize the potential of this new paradigm with the support of new sound business models and a critical shift in the perception of decision making. In this short visioning article, the authors are analyzing the main aspects of Big Data and Data Analytics Research and they provide their own metaphor for the next years. A number of research directions are outlined as well as a new roadmap towards the evolution of Big Data to Smart Decisions and Cognitive Computing. The authors do hope that the readers would like to react and to propose their own value propositions for the domain initiating a scientific dialogue beyond self-fulfilled expectations.

Big Data domain is one of the most promising ICT sectors with substantial expectations both on the side of market growing and design shift in the area of data storage managment and analytics. However, today, the level of complexity achieved and the lack of standardisation of Big Data management architectures represent a huge barrier towards the adoption and execution of analytics especially for those organizations and SMEs not including a sufficient amount of competences and knowledge. The full potential of Big Data Analytics (BDA) can be unleashed only through the definition of approaches that accomplish Big Data users' expectations and requirements, also when the latter are fuzzy and ambiguous. Under these premises, we propose Big Data Analytics-as-a-Service (BDAaaS) as the next-generation Big Data Analytics paradigm and we discuss issues and challenges from the BDAaaS design and development perspective.

The advent of social networks and Internet-of-Things has resulted in unprecedented capability of collecting, sharing and analyzing massive amounts of data. From a security perspective, Big Data may seriously weaken confidentiality, as techniques for improving Big Data analytics performance-including early fusion of heterogeneous data sources - increase the hidden redundancy of data representation, generating ill-protected copies. This gray area of redundancy triggers new disclosure threats that challenge traditional techniques to protect privacy and confidentiality. This position paper starts by proposing a definition of the Big Data Leak threat (as opposed to the one of data breach) and its role as a component of disclosure risk. Then, it discusses how a paradigm of Known, Detect, Contain and Recover could be used to establish Big Data security practices for containing disclosure risks connected to Big Data analytics.

Sequential, predominantly temporal nature of the vast amounts of big data released every day from many different sources could potentially be linked, aligned along the time and deliver new evidence for the next generation predictive systems or knowledge discovery engines. However, big data owners are reluctant to share their data due to legally binding privacy and identity protection concerns, thereby posing a major hurdle preventing shared exploitation of big data on a massive scale. Data anonymization is expected to solve this problem, yet the current approaches are limited predominantly to univariate time series generalized by aggregation or clustering to eliminate identifiable uniqueness of individual data points or patterns. For multivariate time series, uniqueness among of the combination of values or patterns across multiple dimensions is much harder to eliminate due the to exponentially growing number of unique configurations of point values across multiple dimensions. Our method implements linearly scalable asynchronous summarization of multivariate time series independently at every dimension. As a result the series retain only a small subset of defining points at different times along multiple dimensions effectively breaking up the multivariate time series into a collection of summarized univariate time series that are perturbed from the original series in terms of actual points and pattern shapes. Current implementation of the anonymizing summarization involves shape preserving greedy elimination and aggregation that supports parallel cluster processing for big data implementation.

A data owner outsourcing the database of a multi user application wants to prevent information leaks caused by outside attackers exploiting software vulnerabilities or by curious personnel. Query processing over encrypted data solves this problem for a single user, but provides only limited functionality in the face of access restrictions for multiple users and keys. ENKI is a system for securely executing queries over sensitive, access restricted data on an outsourced database. It introduces an encryption based access control model and techniques for query execution over encrypted, access restricted data on the database with only a few cases requiring computations on the client. A prototype of ENKI supports all queries seen in three real world use cases and executes queries from TPC-C benchmark with a modest overhead compared to the single user mode.

The Service-Oriented Architecture (SOA) paradigm is giving rise to a new generation of applications built by dynamically composing loosely coupled autonomous services. Clients (i.e., software agents acting on behalf of human users or service providers) implementing such complex applications typically search and integrate services on the basis of their functional requirements and of their trust in the service suppliers. A major issue in this scenario relates to the definition of an assurance technique allowing clients to select services on the basis of their nonfunctional requirements and increasing their confidence that the selected services will satisfy such requirements. In this article, we first present an assurance solution that focuses on security and supports a test-based security certification scheme for Web services. The certification scheme is driven by the security properties to be certified and relies upon a formal definition of the service model. The evidence supporting a certified property is computed using a model-based testing approach that, starting from the service model, automatically generates the test cases to be used in the service certification. We also define a set of indexes and metrics that evaluate the assurance level and the quality of the certification process. Finally, we present our evaluation toolkit and experimental results obtained applying our certification solution to a financial service implementing the Interactive Financial eXchange (IFX) standard.

This work presents a flexible XML selection language, FleXPath which allows the formulation of flexible constraints on both structure and content of XML documents. Some experimental results, obtained with a preliminary prototype, are described in order to show that the idea promises good results.

We present a way of building ontologies that proceeds in a bottom-up fashion, defining concepts as clusters of concrete XML objects. Our rough bottom-up ontologies are based on simple relations like association and inheritance, as well as on value restrictions, and can be used to enrich and update existing upper ontologies. Then, we show how automatically generated assertions based on our bottom-up ontologies can be associated with a flexible degree of trust by nonintrusively collecting user feedback in the form of implicit and explicit votes. Dynamic trust-based views on assertions automatically filter out imprecisions and substantially improve metadata quality in the long run

Security is a crucial concern for commercial and mission critical applications in Web-based environments. Semantic Web-style context descriptions aim at supporting widespread distribution of resources and cooperation of autonomous agents on the Web in a secure way. In this paper, context information associated with Access Control (AC) management policies is defined according to basic operators that can be represented using the Web Ontology Language (OWL). The same primitives are used, in the specification of authorizations, to compose domain scope expressions. Standard inference procedures of Description Logics (DL) can then be used to check the consistency of context information referred to by policy conditions and, moreinterestingly, to pre-process context information for grounding policy propagation and enabling conflict resolution. This work aims at extending the notion of modality conflict in the evaluation of AC policies to take into account semantic Web-style, ontology-based definitions of the entities involved.

Abstract Peer‐to‐peer (P2P) applications are rapidly gaining acceptance among users of Internet‐based services, especially because of their capability of exchanging resources while preserving the anonymity of both requesters and providers. However, concerns have been raised about the possibility that malicious users can exploit the network to spread tampered‐with resources (e.g., malicious programs and viruses). A considerable amount of research has thus focused on the development of trust and reputation models in P2P networks. In this article, we propose to use fuzzy techniques in the design of reputation systems based on collecting and aggregating peers' opinions. Fuzzy techniques are used in the evaluation and synthesis of all the opinions expressed by peers. The behavior of the proposed system is described by comparison with probabilistic approaches.

Peer-to-peer information sharing environments are increasingly gaining acceptance on the Internet as they provide an infrastructure in which the desired information can be located and downloaded while preserving the anonymity of both requestors and providers. As recent experience with P2P environments such as Gnutella shows, anonymity opens the door to possible misuses and abuses by resource providers exploiting the network as a way to spread tampered-with resources, including malicious programs, such as Trojan Horses and viruses. We propose an approach to P2P security where servants can keep track, and share with others, information about the reputation of their peers. Reputation sharing is based on a distributed polling algorithm by which resource requestors can assess the reliability of perspective providers before initiating the download. The approach complements existing P2P protocols and has a limited impact on current implementations. Furthermore, it keeps the current level of anonymity of requestors and providers, as well as that of the parties sharing their view on others' reputations.

Peer-to-peer information sharing environments are increasingly gaining acceptance on the Internet as they provide an infrastructure in which the desired information can be located and downloaded while preserving the anonymity of both requestors and providers. As recent experience with P2P environments such as Gnutella shows, anonymity opens the door to possible misuses and abuses by resource providers exploiting the network as a way to spread tampered with resources, including malicious programs, such as Trojan Horses and viruses.In this paper we propose an approach to P2P security where servents can keep track, and share with others, information about the reputation of their peers. Reputation sharing is based on a distributed polling algorithm by which resource requestors can assess the reliability of perspective providers before initiating the download. The approach nicely complements the existing P2P protocols and has a limited impact on current implementations. Furthermore, it keeps the current level of anonymity of requestors and providers, as well as that of the parties sharing their view on others' reputations.

Lightweight protocols for remote service invocation via HTTP and XML, such as SOAP, are rapidly gaining acceptance among developers of Internet-based e-services, especially because of their rewall-traversal capabilities. However, no standard technique for access control security is currently defined for either HTTP or SOAP itself. Concerns have been raised about the possibility that different SOAP applications will deal with embedded security in different ways, leading to application-dependent security holes. In this paper, we propose an approach that relies on the XML structure of SOAP requests to support fine-grained authorizations at the level of individual XML elements and attributes that compose a SOAP call. The result is a simple, yet powerful and general, technique to enforce access restrictions to SOAP invocations.

The rapid evolution of XML from a mere data exchange format to a universal syntax for encoding domain-specific information raises the need for new query languages specifically conceived to address the characteristics of XML. Such languages should be able not only to extract information from XML documents, but also to apply powerful transformation and restructuring operators, based on a well-defined semantics. Moreover, XML queries should be natural to write and understand, as nontechnical persons also are expected to access the large XML information bases supporting their businesses. This article describes XML-GL, a graphical query language for XML data. XML-GL's uniqueness is in the definition of a graph-based syntax to express a wide variety of XML queries, ranging from simple selections to expressive data transformations involving grouping, aggregation, and arithmetic calculations. XML-GL has an operational semantics based on the notion of graph matching, which serves as a guideline both for the implementation of native processors, and for the adoption of XML-GL as a front-end to any of the XML query languages that are presently under discussion as the standard paradigm for querying XML data.

In this paper we deal with the problem of querying, navigating and exchanging UML design artefacts, including components interfaces, conceptual schemas and the like. Our approach relies on a hierarchy of XML-based descriptors for UML components, representing both functional and non-functional features of artefacts. After presenting a “query-friendly” DTD for such descriptors, the paper illustrates issues related to querying and browsing design repositories. We show how our approach can also address connection-oriented services, such as HTTP-based Web sites.