ICAR - CNR (Institute for High Performance Computing and Networking of the Italian National Research Council), Via Pietro Bucci 8/9C, Rende (CS) – 87036, Italy
Dr. Riccardo Ortale
ICAR - CNR (Institute for High Performance Computing and Networking of the Italian National Research Council), Via Pietro Bucci 8/9C, Rende (CS) – 87036, Italy
A parameter-free, fully-automatic approach to clustering high-dimensional categorical data is proposed. The technique is based on a two-phase iterative procedure, which attempts to improve the overall quality of the whole partition. In the first phase, cluster assignments are given, and a new cluster is added to the partition by identifying and splitting a low-quality cluster. In the second phase, the number of clusters is fixed, and an attempt to optimize cluster assignments is done. On the basis of such features, the algorithm attempts to improve the overall quality of the whole partition and finds clusters in the data, whose number is naturally established on the basis of the inherent features of the underlying data set rather than being previously specified. Furthermore, the approach is parametric to the notion of cluster quality: Here, a cluster is defined as a set of tuples exhibiting a sort of homogeneity. We show how a suitable notion of cluster homogeneity can be defined in the context of high-dimensional categorical data, from which an effective instance of the proposed clustering scheme immediately follows. Experiments on both synthetic and real data prove that the devised algorithm scales linearly and achieves nearly optimal results in terms of compactness and separation.
We present a new probabilistic approach to modeling social interactions, that seamlessly integrates community discovery and role assignment for a deeper understanding of connectivity patterns in social networks. The devised approach is an unsupervised learning technique based on a Bayesian hierarchical model of social interactions. This model specifies an intuitive generative process, in which pairs of nodes in a social network are associated with communities as well as roles in the context of the respective communities, before that a directed interaction is possibly established between them. According to the generative semantics of the proposed model, nodes are represented as probability distributions over communities, while communities are represented as probability distributions over roles. Such distributions are unknown parameters of the proposed model, that are estimated from social-network data through approximated posterior inference and parameter estimation. A comparative evaluation over real-world social networks reveals that our approach outperforms state-of-the-art competitors in terms of link prediction.
1 January 2004 in Lecture Notes in Computer Science
The supervised classification of XML documents by structure involves learning predictive models in which certain structural regularities discriminate the individual document classes. Hitherto, research has focused on the adoption of prespecified substructures. This is detrimental for classification effectiveness, since the a priori chosen substructures may not accord with the structural properties of the XML documents. Therein, an unexplored question is how to choose the type of structural regularity that best adapts to the structures of the available XML documents. We tackle this problem through X-Class, an approach that handles all types of tree-like substructures and allows for choosing the most discriminatory one. Algorithms are designed to learn compact rule-based classifiers in which the chosen substructures discriminate the classes of XML documents. X-Class is studied across various domains and types of substructures. Its classification performance is compared against several rule-based and SVM-based competitors. Empirical evidence reveals that the classifiers induced by X-Class are compact, scalable, and at least as effective as the established competitors. In particular, certain substructures allow the induction of very compact classifiers that generally outperform the rule-based competitors in terms of effectiveness over all chosen corpora of XML data. Furthermore, such classifiers are substantially as effective as the SVM-based competitor, with the additional advantage of a high-degree of interpretability.
26 October 2023 in IEEE Transactions on Services Computing
The behavioral analysis of smart devices plays a key role in enforcing security for IoT environments. In particular, anomalous patterns can be targeted in the behavior of smart devices as potential IoT cybersecurity threats. In this paper, an explainable machine-learning approach is proposed for dealing with behavioral anomalies. Essentially, a rule-based classifier is inferred from the observed behavior of smart devices, to detect and explain patterns of behavioral anomalies. Predictive association modeling is adopted in the formulation of the classifier, to achieve superior effectiveness in detecting behavioral patterns and ensuring clear explanations of both these latter and their classifications. Moreover, the specifically-conceived design of the classifier reduces the number of tunable parameters to one. An extensive empirical evaluation is comparatively carried out on real-world benchmark data. The experimental results reveal the effectiveness, robustness, and scalability of the proposed approach