Distance concentration is a phantom menace for the labeling of high dimensional data by distance-... more Distance concentration is a phantom menace for the labeling of high dimensional data by distance-based classifiers. Filter methods reduce data dimensionality, but they also add their ranking bias indirectly into the classification procedure. In this study, we examine the filtering problem from another perspective, in which multiple filters are aggregated according to classifiers’ constraints. Our approach, named S-Filter, is designed as a top-k skyline (k-skyband) search over multiple rankings by relying on the concept of \(\mathcal {F}\)–dominance for weighted and monotone linear functions. Unlike existing approaches, S-Filter provides a deterministic strategy for joining multiple filters and avoids the semantic problem of breaking top-k ties. S-Filter’s first stage uses labeling-driven measures, e.g., F1-Score, for assessing the quality of each filter with regards to a particular classifier, whereas range-tolerance intervals around the initial quality measures define the partial search weights. Next, S-Filter applies the FSA instance-optimal algorithm for selecting all the dimensions that can be among the top-k for a weight within the range-tolerance intervals. Experiments on high dimensional datasets show that S-Filter outperforms state-of-the-art filters in two scenarios: (i) exploratory analysis on varying k and range-tolerance intervals, and (ii) data reduction to its intrinsic dimensionality.
The problem of aggregating scores, so as to provide a ranking of objects in a dataset according t... more The problem of aggregating scores, so as to provide a ranking of objects in a dataset according to different evaluation criteria, is central to many modern data-intensive applications. Although efficient (instance optimal) algorithms exist to this purpose (such as the Threshold Algorithm TA and its variants) none of them is able to deal with scenarios in which the function used to aggregate scores is only partially specified. This is the typical case when the function is a weighted sum, and the user is unable to provide precise values for the weights. In this paper, we consider the problem of processing multi-source top-k queries, when only constraints, rather than precise values, are available for the weights. After observing that the so-called Fagin's Algorithm (FA) can be adapted to solve the problem, yet only when no constraints at all are present (a case in which our queries will return the k-skyband of the dataset), we introduce the novel FSA algorithm, which we prove to be instance optimal for any set of constraints on the weights. We also propose several optimizations to the basic FSA logic so as to improve execution times. Experimental analysis on both real and synthetic datasets shows that our optimizations are indeed highly effective and that the increased flexibility provided by FSA introduces little overhead with respect to the case of classical top-k queries.
With the increasing popularity of multimedia databases, k-nearest-neighbor searches in high dimen... more With the increasing popularity of multimedia databases, k-nearest-neighbor searches in high dimensional feature spaces have become one of more common requirements in multimedia information retrieval. Many indexing structures and retrieval methods have been developed. However, their performance is far from satisfactory. In most cases, they are no better in practice than a linear search. Due to limitations of exact nearest-neighbor search techniques in high-dimensional spaces, there have been proposals for approximate nearest-neighbor searches, which are also known as similarity searches. The basic principle is to speed up the termination of searches by giving a limit to the acceptable worst-case error. This thesis focuses on providing an infrastructure and techniques for efficient similarity searches for multimedia databases. First, the similarity search is redefined to make it more practical and usable for multimedia databases. The redefinition includes reformulation of the goals and performance measures for similarity searches. Building on this formulation, a hierarchical structure for supporting efficient similarity searches is proposed. The structure maintains data locality using a novel adaptive k-means algorithm. The use of the k-means algorithm makes it possible to build the structure efficiently. This is in sharp contrast to other clustering-based structures previously used in the information retrieval area that utilize graph-based clustering methods with complexity O(n2), where n is the number of objects in the database. Previously, k-means was deemed as unsuitable since the algorithm requires the number of clusters to be known. The adaptive k-means algorithm solves this problem. The method can be also used for other applications requiring an efficient and automatic clustering method. The hierarchical structure is complemented by a probabilistic search which utilizes the “branch-and-bound” strategy. A probabilistic distance function based on the Chebyshev theorem is used as the branch selection criterion. An experimental framework to evaluate similarity search performance is proposed. The measured values will give an indication of the recall/precision of the similarity search results and the quality of the approximate solution. The proposed scheme has cut the search cost by up to 95% compared with exact k-nearest-neighbor searches using other tree-based methods. It has cut the search cost by up to 75% compared with exact k-nearest-neighbor searches on recently proposed signature-based method. It can serve as a general framework for content-based visual information retrieval in multimedia databases and the Internet.
This paper reports on the 28th Italian Symposium on Advanced Database Systems (SEBD 2020), held o... more This paper reports on the 28th Italian Symposium on Advanced Database Systems (SEBD 2020), held online as a virtual conference from the 21st to the 24th of June 2020. The topics that were addressed in this edition of the conference were organized in the sessions: ontologies and data integration, anomaly detection and dependencies, text analysis and search, deep learning, noSQL data, trajectories and diffusion, health and medicine, context and ranking, social and knowledge graphs, multimedia content analysis, security issues, and data mining.
Preferences about objects of interest are often expressed at different levels of granularity, not... more Preferences about objects of interest are often expressed at different levels of granularity, not always matching the level of detail of stored data. For instance, we prefer rock to pop music, yet scheduled concerts only cite the name of the performer, with no reference to the musical genre. In this paper we address this common mismatch by leveraging the vast amounts of data organized in taxonomies (such as those found in electronic catalogs and classification systems) for propagating preferences from more generic to more specific concepts. This will help users to locate their preferred objects. In spite of its apparent simplicity, this problem requires special care in order to avoid some undesirable effects, e.g., when conflicting preferences at different levels have to be combined (although, generally, we prefer rock to pop music, we would never miss a performance by Madonna). We present a formal model to represent preferences and state the desirable properties of preference propagation, such as the fact that more specific preferences always prevail over more generic ones. We then propose a method for propagating preferences along taxonomies, complying with the stated properties, and show how preferred objects can thereby be efficiently determined.
With the amount of heterogeneous data that Next Generation Sequencing (NGS) is producing, many in... more With the amount of heterogeneous data that Next Generation Sequencing (NGS) is producing, many interesting computational problems are emerging and call for urgent solutions. Genome Browsers (e.g., IGB) are tools to visually compare and browse through multiple genomic feature samples aligned to the same genome reference and laid out on different genome browser tracks. They allow the visual inspection and identification of interesting “patterns” on multiple tracks, i.e. sets of genomic regions/peaks at given distances from each other in different genome browser tracks. Nevertheless, once such patterns are visually identified in a genome section, the search of their occurrences along the whole genome is a complex computational task currently not supported, although their discovery along the whole genome is very important for the biological interpretation of NGS experimental results and comprehension of biomolecular phenomena. To overcome such limitation, we present an optimized “similarity”-based pattern-search algorithm able to efficiently find, within a large set of genomic data, genomic region sets which are similar to a given pattern. We implemented our algorithm within an IGB plugin, which allows intuitive user interaction in both the visual selection of an interesting pattern on the loaded IGB tracks and the visualization within the IGB of the occurrences along the entire genome of the region sets similar to a selected pattern found by our algorithm. This demonstrates the efficiency and the accuracy of the proposed method
In a scenario with “traditional” and “multimedia” data sources, this position paper discusses the... more In a scenario with “traditional” and “multimedia” data sources, this position paper discusses the following question: “How can a multimedia local source (e.g., Windsurf) supporting ranking queries be integrated into a mediator system without such capabilities (e.g., MOMIS)?” More precisely, “How to support ranking queries coming from a multimedia local source within a mediator system with a “traditional” query processor based on an SQL-engine?” We first describe a naive approach for the execution of range and Top-K global queries where the MOMIS query processing method remains substantially unchanged, but, in the case of Top-K queries, it does not guarantee to obtain K results. We then discuss two alternative modalities for allowing MOMIS to return the Top-K best results of a global query.
Motivation Next Generation Sequencing (NGS), with the high amount of heterogeneous data that it i... more Motivation Next Generation Sequencing (NGS), with the high amount of heterogeneous data that it is generating, is opening many interesting practical and theoretical computational problems. Genome browsers, e.g. UCSC Genome Browser (Kuhn et al., 2013) or Integrated Genome Browser (IGB) (Nicol et al., 2009), allow visual inspection and identification of interesting patterns on multiple genome browser tracks, i.e. of sets of (epi)genomic regions/peaks at given distances from each other in different tracks. For example, such patterns can describe gene expression regulatory DNA areas including heterogeneous (epi)genomic features (e.g. histone modification and/or different transcription factor binding regions). Yet, once such patterns are visually identified in a genome section, the search of their occurrences along the whole genome is a complex computational task that is currently not supported, despite their discovery along the whole genome is very important for the biological interpret...
The number of datasets available to legal practitioners, policy makers, scientists, and many othe... more The number of datasets available to legal practitioners, policy makers, scientists, and many other categories of citizens is growing at an unprecedented rate. Ethics-aware data processing has become a pressing need, considering that data are often used within critical decision processes (e.g., staff evaluation, college admission, criminal sentencing). The goal of this paper is to propose a vision for the injection of ethical principles (fairness, non-discrimination, transparency, data protection, diversity, and human interpretability of results) into the data analysis lifecycle (source selection, data integration, and knowledge extraction) so as to make them first-class requirements. In our vision, a comprehensive checklist of ethical desiderata for data protection and processing needs to be developed, along with methods and techniques to ensure and verify that these ethically motivated requirements and related legal norms are fulfilled throughout the data selection and exploration ...
Distance concentration is a phantom menace for the labeling of high dimensional data by distance-... more Distance concentration is a phantom menace for the labeling of high dimensional data by distance-based classifiers. Filter methods reduce data dimensionality, but they also add their ranking bias indirectly into the classification procedure. In this study, we examine the filtering problem from another perspective, in which multiple filters are aggregated according to classifiers’ constraints. Our approach, named S-Filter, is designed as a top-k skyline (k-skyband) search over multiple rankings by relying on the concept of \(\mathcal {F}\)–dominance for weighted and monotone linear functions. Unlike existing approaches, S-Filter provides a deterministic strategy for joining multiple filters and avoids the semantic problem of breaking top-k ties. S-Filter’s first stage uses labeling-driven measures, e.g., F1-Score, for assessing the quality of each filter with regards to a particular classifier, whereas range-tolerance intervals around the initial quality measures define the partial search weights. Next, S-Filter applies the FSA instance-optimal algorithm for selecting all the dimensions that can be among the top-k for a weight within the range-tolerance intervals. Experiments on high dimensional datasets show that S-Filter outperforms state-of-the-art filters in two scenarios: (i) exploratory analysis on varying k and range-tolerance intervals, and (ii) data reduction to its intrinsic dimensionality.
The problem of aggregating scores, so as to provide a ranking of objects in a dataset according t... more The problem of aggregating scores, so as to provide a ranking of objects in a dataset according to different evaluation criteria, is central to many modern data-intensive applications. Although efficient (instance optimal) algorithms exist to this purpose (such as the Threshold Algorithm TA and its variants) none of them is able to deal with scenarios in which the function used to aggregate scores is only partially specified. This is the typical case when the function is a weighted sum, and the user is unable to provide precise values for the weights. In this paper, we consider the problem of processing multi-source top-k queries, when only constraints, rather than precise values, are available for the weights. After observing that the so-called Fagin's Algorithm (FA) can be adapted to solve the problem, yet only when no constraints at all are present (a case in which our queries will return the k-skyband of the dataset), we introduce the novel FSA algorithm, which we prove to be instance optimal for any set of constraints on the weights. We also propose several optimizations to the basic FSA logic so as to improve execution times. Experimental analysis on both real and synthetic datasets shows that our optimizations are indeed highly effective and that the increased flexibility provided by FSA introduces little overhead with respect to the case of classical top-k queries.
With the increasing popularity of multimedia databases, k-nearest-neighbor searches in high dimen... more With the increasing popularity of multimedia databases, k-nearest-neighbor searches in high dimensional feature spaces have become one of more common requirements in multimedia information retrieval. Many indexing structures and retrieval methods have been developed. However, their performance is far from satisfactory. In most cases, they are no better in practice than a linear search. Due to limitations of exact nearest-neighbor search techniques in high-dimensional spaces, there have been proposals for approximate nearest-neighbor searches, which are also known as similarity searches. The basic principle is to speed up the termination of searches by giving a limit to the acceptable worst-case error. This thesis focuses on providing an infrastructure and techniques for efficient similarity searches for multimedia databases. First, the similarity search is redefined to make it more practical and usable for multimedia databases. The redefinition includes reformulation of the goals and performance measures for similarity searches. Building on this formulation, a hierarchical structure for supporting efficient similarity searches is proposed. The structure maintains data locality using a novel adaptive k-means algorithm. The use of the k-means algorithm makes it possible to build the structure efficiently. This is in sharp contrast to other clustering-based structures previously used in the information retrieval area that utilize graph-based clustering methods with complexity O(n2), where n is the number of objects in the database. Previously, k-means was deemed as unsuitable since the algorithm requires the number of clusters to be known. The adaptive k-means algorithm solves this problem. The method can be also used for other applications requiring an efficient and automatic clustering method. The hierarchical structure is complemented by a probabilistic search which utilizes the “branch-and-bound” strategy. A probabilistic distance function based on the Chebyshev theorem is used as the branch selection criterion. An experimental framework to evaluate similarity search performance is proposed. The measured values will give an indication of the recall/precision of the similarity search results and the quality of the approximate solution. The proposed scheme has cut the search cost by up to 95% compared with exact k-nearest-neighbor searches using other tree-based methods. It has cut the search cost by up to 75% compared with exact k-nearest-neighbor searches on recently proposed signature-based method. It can serve as a general framework for content-based visual information retrieval in multimedia databases and the Internet.
This paper reports on the 28th Italian Symposium on Advanced Database Systems (SEBD 2020), held o... more This paper reports on the 28th Italian Symposium on Advanced Database Systems (SEBD 2020), held online as a virtual conference from the 21st to the 24th of June 2020. The topics that were addressed in this edition of the conference were organized in the sessions: ontologies and data integration, anomaly detection and dependencies, text analysis and search, deep learning, noSQL data, trajectories and diffusion, health and medicine, context and ranking, social and knowledge graphs, multimedia content analysis, security issues, and data mining.
Preferences about objects of interest are often expressed at different levels of granularity, not... more Preferences about objects of interest are often expressed at different levels of granularity, not always matching the level of detail of stored data. For instance, we prefer rock to pop music, yet scheduled concerts only cite the name of the performer, with no reference to the musical genre. In this paper we address this common mismatch by leveraging the vast amounts of data organized in taxonomies (such as those found in electronic catalogs and classification systems) for propagating preferences from more generic to more specific concepts. This will help users to locate their preferred objects. In spite of its apparent simplicity, this problem requires special care in order to avoid some undesirable effects, e.g., when conflicting preferences at different levels have to be combined (although, generally, we prefer rock to pop music, we would never miss a performance by Madonna). We present a formal model to represent preferences and state the desirable properties of preference propagation, such as the fact that more specific preferences always prevail over more generic ones. We then propose a method for propagating preferences along taxonomies, complying with the stated properties, and show how preferred objects can thereby be efficiently determined.
With the amount of heterogeneous data that Next Generation Sequencing (NGS) is producing, many in... more With the amount of heterogeneous data that Next Generation Sequencing (NGS) is producing, many interesting computational problems are emerging and call for urgent solutions. Genome Browsers (e.g., IGB) are tools to visually compare and browse through multiple genomic feature samples aligned to the same genome reference and laid out on different genome browser tracks. They allow the visual inspection and identification of interesting “patterns” on multiple tracks, i.e. sets of genomic regions/peaks at given distances from each other in different genome browser tracks. Nevertheless, once such patterns are visually identified in a genome section, the search of their occurrences along the whole genome is a complex computational task currently not supported, although their discovery along the whole genome is very important for the biological interpretation of NGS experimental results and comprehension of biomolecular phenomena. To overcome such limitation, we present an optimized “similarity”-based pattern-search algorithm able to efficiently find, within a large set of genomic data, genomic region sets which are similar to a given pattern. We implemented our algorithm within an IGB plugin, which allows intuitive user interaction in both the visual selection of an interesting pattern on the loaded IGB tracks and the visualization within the IGB of the occurrences along the entire genome of the region sets similar to a selected pattern found by our algorithm. This demonstrates the efficiency and the accuracy of the proposed method
In a scenario with “traditional” and “multimedia” data sources, this position paper discusses the... more In a scenario with “traditional” and “multimedia” data sources, this position paper discusses the following question: “How can a multimedia local source (e.g., Windsurf) supporting ranking queries be integrated into a mediator system without such capabilities (e.g., MOMIS)?” More precisely, “How to support ranking queries coming from a multimedia local source within a mediator system with a “traditional” query processor based on an SQL-engine?” We first describe a naive approach for the execution of range and Top-K global queries where the MOMIS query processing method remains substantially unchanged, but, in the case of Top-K queries, it does not guarantee to obtain K results. We then discuss two alternative modalities for allowing MOMIS to return the Top-K best results of a global query.
Motivation Next Generation Sequencing (NGS), with the high amount of heterogeneous data that it i... more Motivation Next Generation Sequencing (NGS), with the high amount of heterogeneous data that it is generating, is opening many interesting practical and theoretical computational problems. Genome browsers, e.g. UCSC Genome Browser (Kuhn et al., 2013) or Integrated Genome Browser (IGB) (Nicol et al., 2009), allow visual inspection and identification of interesting patterns on multiple genome browser tracks, i.e. of sets of (epi)genomic regions/peaks at given distances from each other in different tracks. For example, such patterns can describe gene expression regulatory DNA areas including heterogeneous (epi)genomic features (e.g. histone modification and/or different transcription factor binding regions). Yet, once such patterns are visually identified in a genome section, the search of their occurrences along the whole genome is a complex computational task that is currently not supported, despite their discovery along the whole genome is very important for the biological interpret...
The number of datasets available to legal practitioners, policy makers, scientists, and many othe... more The number of datasets available to legal practitioners, policy makers, scientists, and many other categories of citizens is growing at an unprecedented rate. Ethics-aware data processing has become a pressing need, considering that data are often used within critical decision processes (e.g., staff evaluation, college admission, criminal sentencing). The goal of this paper is to propose a vision for the injection of ethical principles (fairness, non-discrimination, transparency, data protection, diversity, and human interpretability of results) into the data analysis lifecycle (source selection, data integration, and knowledge extraction) so as to make them first-class requirements. In our vision, a comprehensive checklist of ethical desiderata for data protection and processing needs to be developed, along with methods and techniques to ensure and verify that these ethically motivated requirements and related legal norms are fulfilled throughout the data selection and exploration ...
Uploads