Antonio Maratea

A Bound for the Accuracy of Sensors Acquiring Compositional Data

Among the many challenges that the Internet of Things poses, the accuracy of the sensor network and relative data flow is of the foremost importance: sensors monitor the surrounding environment of an object and give information on its position, situation or context, and an error in the acquired data can lead to inappropriate decisions and uncontrolled consequences. Given a sensor network that gathers relative data - that is data for which ratios of parts are more important than absolute values - acquired data have a compositional nature and all values need to be scaled. To analyze these data a common practice is to map bijectively compositions into the ordinary euclidean space through a suitable transformation, so that standard multivariate analysis techniques can be used. In this paper an error bound on the commonly used asymmetric log-ratio transformation is found in the Simplex. The purpose is to highlight areas of the Simplex where the transformation is ill conditioned and to isolate values for which the additive log-ratio transform cannot be accurately computed. Results show that the conditioning of the transformation is strongly affected by the closeness of the transformed values and that not negligible distortions can be generated due to the unbounded propagation of the errors. An explicit formula for the accuracy of the sensors given the maximum allowed tolerance has been derived, and the critical values in the Simplex where the transformation is component-wise ill conditioned have been isolated.

Generation of description metadata for video files

Automatic Metadata Generation in the context of e-learning standards is usually referred to algorithms able to process and annotate semi structured documents in plain text. As most of the information available on the web nowadays is unstructured and in the form of multimedia files, the need for more general approaches arises. We propose an automatic metadata generation procedure that allows to label specific unstructured data (video lectures) with metadata compliant to the Learning Object Metadata standard. After preprocessing, three different summarization algorithms are tested and used to obtain a synthetic description of video content, both in terms of Description and Title. Results show that, in the provided context, the obtained Description has a good agreement with the lesson abstract written by its author.

Concordance indices for comparing fuzzy, possibilistic, rough and grey partitions

Many indices have been proposed in literature for the comparison of two crisp data partitions, as resulting from two different classifications attempts, two different clustering solutions or the comparison of a predicted vs. a true labelling. Crisp partitions however cannot model ambiguity, vagueness or uncertainty in class definition and thus are not suitable to model all cases where information lacks, terms definitions are intrinsically imprecise or the classification results from a human expert knowledge representation. In presence of vagueness, it is not obvious how to quantify overlap or agreement of two different partitions of the same data and many facets of vagueness have emerged in literature through complimentary theories. The aim of the paper is to give simple numerical indices to quantify partitions agreement in the fuzzy, possibilistic, rough and grey frameworks. We propose a method based on pseudo counts, intuitive in the meaning and simple to implement that is very general and allows comparing fuzzy, possibilistic, rough and grey partitions, even with a different number of classes. The proposed method has just one free parameter used to model sensitivity to higher values of membership.

Numerical Stability Analysis of the Centered Log-Ratio Transformation

Data have a compositional nature when the information content to be extracted and analyzed is conveyed into the ratio of parts, instead of the absolute amount. When the data are compositional, they need to be scaled so that subsequent analysis are scale-invariant, and geometrically this means to force them into the open Simplex. A common practice to analyze compositional data is to map bijectively compositions into the ordinary euclidean space through a suitable transformation, so that standard multivariate analysis techniques can be used. In this paper, the stability analysis of the Centered Log-Ratio (clr) transformation is performed. The purpose is to isolate areas of the Simplex where the clr transformation is ill conditioned and to highlight values for which the clr transformation cannot be accurately computed. Results show that the mapping accuracy is strongly affected by the closeness of the values to their geometric mean, and that in the worst case the clr can amplify the errors by an unbounded factor.

Semantic Maps of Twitter Conversations

Twitter is an irreplaceable source of data for opinion mining, emergency communications, or fact sharing, whose readability is severely limited by the sheer volume of tweets published every day. A method to represent and synthesize the information content of conversations on Twitter in form of semantic maps, from which the main topics and the main orientations of tweeters may easily be read, is proposed hereafter. After a preliminary grouping of tweets in conversations, relevant keywords and Named Entities are extracted, disambiguated and clustered. Annotations are made using extensive knowledge bases and state-of-the-art techniques from Natural Language Processing and Machine Learning. The results are in form of coloured graphs, to be easily interpretable. Several experiments confirm the high understandability and the good adherence to tackled topics of the mapped conversations.

A resampling strategy based on bootstrap to reduce the effect of large blunders in GPS absolute positioning

In the absence of obstacles, a GPS device is generally able to provide continuous and accurate estimates of position, while in urban scenarios buildings can generate multipath and echo-only phenomena that severely affect the continuity and the accuracy of the provided estimates. Receiver autonomous integrity monitoring (RAIM) techniques are able to reduce the negative consequences of large blunders in urban scenarios, but require both a good redundancy and a low contamination to be effective. In this paper a resampling strategy based on bootstrap is proposed as an alternative to RAIM, in order to estimate accurately position in case of low redundancy and multiple blunders: starting with the pseudorange measurement model, at each epoch the available measurements are bootstrapped—that is random sampled with replacement—and the generated a posteriori empirical distribution is exploited to derive the final position. Compared to standard bootstrap, in this paper the sampling probabilities are not uniform, but vary according to an indicator of the measurement quality. The proposed method has been compared with two different RAIM techniques on a data set collected in critical conditions, resulting in a clear improvement on all considered figures of merit.

Integrating Rough Set Principles in the Graded Possibilistic Clustering

Abstract Applied to fuzzy clustering, the graded possibilistic model allows the soft transition from probabilistic to possibilistic memberships, constraining the memberships in a region that is narrower the closer to probabilistic the memberships are. The integration of rough sets principles in the graded possibilistic clustering aims to improve the flexibility and the performance of the graded possibilistic model, providing a further option for uncertainty modeling. Through the novel concept of the Rough Feasible Region, the proposed approach differentiates the projection of memberships in the core and in the boundary of each cluster, exploiting the indiscernibility relation typical of rough sets and allowing a more robust and efficient estimation of centroids. Tests on real data confirm its viability.

Decoy clustering through graded possibilistic c-medoids

Modern methods for ab initio prediction of protein structures typically explore multiple simulated conformations, called decoys, to find the best native-like conformations. To limit the search space, clustering algorithms are routinely used to group similar decoys, based on the hypothesis that the largest group of similar decoys will be the closest to the native state. In this paper a novel clustering algorithm, called Graded Possibilistic c-medoids, is proposed and applied to a decoy selection problem. As it will be shown, the added flexibility of the graded possibilistic framework allows an effective selection of the best decoys with respect to similar methods based on medoids — that is on the most central points belonging to each cluster. The proposed algorithm has been compared with other c-medoids algorithms and also with SPICKER on real data, the large majority of times outperforming both.

Extended Graph Backbone for Motif Analysis

Local interaction patterns in complex networks, called motifs, explain many of the network properties but are challenging to extract due to the large search space. In this paper first an approximate representation of a complex network in terms of an extended backbone is proposed, then a reduced sampling space that speeds up the motif search in different kinds of networks is explored based on this representation. It will be shown using several real datasets that the proposed method is effective in reducing the sampling space, extracts the same relevant patterns, and hence preservs the network local structural information.

Mapping the reliability of the additive log-ratio transformation

Data for which ratios of parts are more important than absolute values have a compositional nature and should be analysed in the (D - 1)-dimensional simplex, but in order to use standard multivariate analysis techniques they are often mapped bijectively from the simplex into the ordinary Euclidean space. The additive log-ratio (alr) is one popular ad hoc transformation, that has already been shown to amplify the relative errors in data, in some cases by an unbounded factor. In this paper an explicit mapping of the reliability of the alr transformed data is derived, given the desired accuracy and the relative error of the sensors acquiring them. To this purpose, first the relative condition number for the alr transformation has been defined and plotted in the 3-simplex, then a generalised characterisation of well conditioned compositions (with a condition number less than or equal to an arbitrary threshold) has been derived and finally isoconditioning surfaces have been defined, analytically derived and plotted in the 3-simplex.