Search

Machine Learning (ML) is crucial in many sectors, including computer vision. However, ML models trained on sensitive data face security challenges, as they can be attacked and leak information. Privacy-Preserving Machine Learning (PPML) addresses this by using Differential Privacy (DP) to balance utility and privacy. This study identifies image dataset characteristics that affect the utility and vulnerability of private and non-private Convolutional Neural Network (CNN) models. Through analyzing multiple datasets and privacy budgets, we find that imbalanced datasets increase vulnerability in minority classes, but DP mitigates this issue. Datasets with fewer classes improve both model utility and privacy, while high entropy or low Fisher Discriminant Ratio (FDR) datasets deteriorate the utility-privacy trade-off. These insights offer valuable guidance for practitioners and researchers in estimating and optimizing the utility-privacy trade-off in image datasets, helping to inform data and privacy modifications for better outcomes based on dataset characteristics.

Cluster repair methods aim to determine errors in clusters and modify them so that each cluster consists of records representing the same entity. Current cluster repair methodologies primarily assume duplicate-free data sources, where each record from one source corresponds to a unique record from another. However, real-world data often deviates from this assumption due to quality issues. Recent approaches apply clustering methods in combination with link categorization methods so they can be applied to data sources with duplicates. Nevertheless, the results do not show a clear picture since the quality highly varies depending on the configuration and dataset. In this study, we introduce a novel approach for cluster repair that utilizes graph metrics derived from the underlying similarity graphs. These metrics are pivotal in constructing a classification model to distinguish between correct and incorrect edges. To address the challenge of limited training data, we integrate an active learning mechanism tailored to cluster-specific attributes. The evaluation shows that the method outperforms existing cluster repair methods without distinguishing between duplicate-free or dirty data sources. Notably, our modified active learning strategy exhibits enhanced performance when dealing with datasets containing duplicates, showcasing its effectiveness in such scenarios.

Smartwatch health sensor data are increasingly utilized in smart health applications and patient monitoring, including stress detection. However, such medical data often comprise sensitive personal information and are resource-intensive to acquire for research purposes. In response to this challenge, we introduce the privacy-aware synthetization of multi-sensor smartwatch health readings related to moments of stress, employing Generative Adversarial Networks (GANs) and Differential Privacy (DP) safeguards. Our method not only protects patient information but also enhances data availability for research. To ensure its usefulness, we test synthetic data from multiple GANs and employ different data enhancement strategies on an actual stress detection task. Our GAN-based augmentation methods demonstrate significant improvements in model performance, with private DP training scenarios observing an 11.90-15.48% increase in F1-score, while non-private training scenarios still see a 0.45% boost. These results underline the potential of differentially private synthetic data in optimizing utility-privacy trade-offs, especially with the limited availability of real training samples. Through rigorous quality assessments, we confirm the integrity and plausibility of our synthetic data, which, however, are significantly impacted when increasing privacy requirements., Comment: Published in the MDPI Sensors Journal

Wearables are widely used for health data collection due to their availability and advanced sensors, enabling smart health applications like stress detection. However, the sensitivity of personal health data raises significant privacy concerns. While user de-identification by removing direct identifiers such as names and addresses is commonly employed to protect privacy, the data itself can still be exploited to re-identify individuals. We introduce a novel framework for similarity-based Dynamic Time Warping (DTW) re-identification attacks on time series health data. Using the WESAD dataset and two larger synthetic datasets, we demonstrate that even short segments of sensor data can achieve perfect re-identification with our Slicing-DTW-Attack. Our attack is independent of training data and computes similarity rankings in about 2 minutes for 10,000 subjects on a single CPU core. These findings highlight that de-identification alone is insufficient to protect privacy. As a defense, we show that adding random noise to the signals significantly reduces re-identification risk while only moderately affecting usability in stress detection tasks, offering a promising approach to balancing privacy and utility.

With knowledge graphs (KGs) at the center of numerous applications such as recommender systems and question answering, the need for generalized pipelines to construct and continuously update such KGs is increasing. While the individual steps that are necessary to create KGs from unstructured (e.g. text) and structured data sources (e.g. databases) are mostly well-researched for their one-shot execution, their adoption for incremental KG updates and the interplay of the individual steps have hardly been investigated in a systematic manner so far. In this work, we first discuss the main graph models for KGs and introduce the major requirement for future KG construction pipelines. Next, we provide an overview of the necessary steps to build high-quality KGs, including cross-cutting topics such as metadata management, ontology development, and quality assurance. We then evaluate the state of the art of KG construction w.r.t the introduced requirements for specific popular KGs as well as some recent tools and strategies for KG construction. Finally, we identify areas in need of further research and improvement., Comment: 51 pages, 5 figures, 4 tables, 328 references

Machine learning (ML) can help fight pandemics like COVID-19 by enabling rapid screening of large volumes of images. To perform data analysis while maintaining patient privacy, we create ML models that satisfy Differential Privacy (DP). Previous works exploring private COVID-19 models are in part based on small datasets, provide weaker or unclear privacy guarantees, and do not investigate practical privacy. We suggest improvements to address these open gaps. We account for inherent class imbalances and evaluate the utility-privacy trade-off more extensively and over stricter privacy budgets. Our evaluation is supported by empirically estimating practical privacy through black-box Membership Inference Attacks (MIAs). The introduced DP should help limit leakage threats posed by MIAs, and our practical analysis is the first to test this hypothesis on the COVID-19 classification task. Our results indicate that needed privacy levels might differ based on the task-dependent practical threat from MIAs. The results further suggest that with increasing DP guarantees, empirical privacy leakage only improves marginally, and DP therefore appears to have a limited impact on practical MIA defense. Our findings identify possibilities for better utility-privacy trade-offs, and we believe that empirical attack-specific privacy estimation can play a vital role in tuning for practical privacy., Comment: Extended version of the paper accepted at the 20th International Conference on Security and Cryptography SECRYPT 2023. This version is more detailed and includes additional content: a longer results chapter and an appendix containing a proof

This work is a summarized view on the results of a one-year cooperation between Oracle Corp. and the University of Leipzig. The goal was to research the organization of relationships within multi-dimensional time-series data, such as sensor data from the IoT area. We showed in this project that temporal property graphs with some extensions are a prime candidate for this organizational task that combines the strengths of both data models (graph and time-series). The outcome of the cooperation includes four achievements: (1) a bitemporal property graph model, (2) a temporal graph query language, (3) a conception of continuous event detection, and (4) a prototype of a bitemporal graph database that supports the model, language and event detection., Comment: 21 pages

Entity Resolution (ER) is a constitutional part for integrating different knowledge graphs in order to identify entities referring to the same real-world object. A promising approach is the use of graph embeddings for ER in order to determine the similarity of entities based on the similarity of their graph neighborhood. The similarity computations for such embeddings translates to calculating the distance between them in the embedding space which is comparatively simple. However, previous work has shown that the use of graph embeddings alone is not sufficient to achieve high ER quality. We therefore propose a more comprehensive ER approach for knowledge graphs called EAGER (Embedding-Assisted Knowledge Graph Entity Resolution) to flexibly utilize both the similarity of graph embeddings and attribute values within a supervised machine learning approach. We evaluate our approach on 23 benchmark datasets with differently sized and structured knowledge graphs and use hypothesis tests to ensure statistical significance of our results. Furthermore we compare our approach with state-of-the-art ER solutions, where our approach yields competitive results for table-oriented ER problems and shallow knowledge graphs but much better results for deeper knowledge graphs., Comment: 10 pages, 7 figures

Entity resolution targets at identifying records that represent the same real-world entity from one or more datasets. A major challenge in learning-based entity resolution is how to reduce the label cost for training. Due to the quadratic nature of record pair comparison, labeling is a costly task that often requires a significant effort from human experts. Inspired by recent advances of generative adversarial network (GAN), we propose a novel deep learning method, called ErGAN, to address the challenge. ErGAN consists of two key components: a label generator and a discriminator which are optimized alternatively through adversarial learning. To alleviate the issues of overfitting and highly imbalanced distribution, we design two novel modules for diversity and propagation, which can greatly improve the model generalization power. We have conducted extensive experiments to empirically verify the labeling and learning efficiency of ErGAN. The experimental results show that ErGAN beats the state-of-the-art baselines, including unsupervised, semi-supervised, and unsupervised learning methods.

Data integration tasks such as the creation and extension of knowledge graphs involve the fusion of heterogeneous entities from many sources. Matching and fusion of such entities require to also match and combine their properties (attributes). However, previous schema matching approaches mostly focus on two sources only and often rely on simple similarity measurements. They thus face problems in challenging use cases such as the integration of heterogeneous product entities from many sources. We therefore present a new machine learning-based property matching approach called LEAPME (LEArning-based Property Matching with Embeddings) that utilizes numerous features of both property names and instance values. The approach heavily makes use of word embeddings to better utilize the domain-specific semantics of both property names and instance values. The use of supervised machine learning helps exploit the predictive power of word embeddings. Our comparative evaluation against five baselines for several multi-source datasets with real-world data shows the high effectiveness of LEAPME. We also show that our approach is even effective when training data from another domain (transfer learning) is used.

Privacy-Preserving Record Linkage (PPRL) supports the integration of sensitive information from multiple datasets, in particular the privacy-preserving matching of records referring to the same entity. PPRL has gained much attention in many application areas, with the most prominent ones in the healthcare domain. PPRL techniques tackle this problem by conducting linkage on masked (encoded) values. Employing PPRL on records from multiple (more than two) parties/sources (multi-party PPRL, MP-PPRL) is an increasingly important but challenging problem that so far has not been sufficiently solved. Existing MP-PPRL approaches are limited to finding only those entities that are present in all parties thereby missing entities that match only in a subset of parties. Furthermore, previous MP-PPRL approaches face substantial scalability limitations due to the need of a large number of comparisons between masked records. We thus propose and evaluate new MP-PPRL approaches that find matches in any subset of parties and still scale to many parties. Our approaches maintain all matches within clusters, where these clusters are incrementally extended or refined by considering records from one party after the other. An empirical evaluation using multiple real datasets ranging from 3 to 26 parties each containing up to $5$ million records validates that our protocols are efficient, and significantly outperform existing MP-PPRL approaches in terms of linkage quality and scalability.

Given a large graph, a graph sample determines a subgraph with similar characteristics for certain metrics of the original graph. The samples are much smaller thereby accelerating and simplifying the analysis and visualization of large graphs. We focus on the implementation of distributed graph sampling for Big Data frameworks and in-memory dataflow systems such as Apache Spark or Apache Flink. We evaluate the scalability of the new implementations and analyze to what degree the sampling approaches preserve certain graph metrics compared to the original graph. The latter analysis also uses comparative graph visualizations. The presented methods will be open source and be integrated into Gradoop, a system for distributed graph analytics.

Link discovery is an active field of research to support data integration in the Web of Data. Due to the huge size and number of available data sources, efficient and effective link discovery is a very challenging task. Common pairwise link discovery approaches do not scale to many sources with very large entity sets. We here propose a distributed holistic approach to link many data sources based on a clustering of entities that represent the same real-world object. Our clustering approach provides a compact and fused representation of entities, and can identify errors in existing links as well as many new links. We support a distributed execution of the clustering approach to achieve faster execution times and scalability for large real-world data sets. We provide a novel gold standard for multi-source clustering, and evaluate our methods with respect to effectiveness and efficiency for large data sets from the geographic and music domains.

Transactional frequent subgraph mining identifies frequent subgraphs in a collection of graphs. This research problem has wide applicability and increasingly requires higher scalability over single machine solutions to address the needs of Big Data use cases. We introduce DIMSpan, an advanced approach to frequent subgraph mining that utilizes the features provided by distributed in-memory dataflow systems such as Apache Spark or Apache Flink. It determines the complete set of frequent subgraphs from arbitrary string-labeled directed multigraphs as they occur in social, business and knowledge networks. DIMSpan is optimized to runtime and minimal network traffic but memory-aware. An extensive performance evaluation on large graph collections shows the scalability of DIMSpan and the effectiveness of its pruning and optimization techniques.

Privacy-preserving record linkage (PPRL) aims at integrating sensitive information from multiple disparate databases of different organizations. PPRL approaches are increasingly required in real-world application areas such as healthcare, national security, and business. Previous approaches have mostly focused on linking only two databases as well as the use of a dedicated linkage unit. Scaling PPRL to more databases (multi-party PPRL) is an open challenge since privacy threats as well as the computation and communication costs for record linkage increase significantly with the number of databases. We thus propose the use of a new encoding method of sensitive data based on Counting Bloom Filters (CBF) to improve privacy for multi-party PPRL. We also investigate optimizations to reduce communication and computation costs for CBF-based multi-party PPRL with and without the use of a dedicated linkage unit. Empirical evaluations conducted with real datasets show the viability of the proposed approaches and demonstrate their scalability, linkage quality, and privacy protection., Comment: This is an extended version of an article published in IEEE ICDM International Workshop on Privacy and Discrimination in Data Mining (PDDM) 2016 - Scalable privacy-preserving linking of multiple databases using counting Bloom filters

Many Big Data applications in business and science require the management and analysis of huge amounts of graph data. Previous approaches for graph analytics such as graph databases and parallel graph processing systems (e.g., Pregel) either lack sufficient scalability or flexibility and expressiveness. We are therefore developing a new end-to-end approach for graph data management and analysis based on the Hadoop ecosystem, called Gradoop (Graph analytics on Hadoop). Gradoop is designed around the so-called Extended Property Graph Data Model (EPGM) supporting semantically rich, schema-free graph data within many distinct graphs. A set of high-level operators is provided for analyzing both single graphs and collections of graphs. Based on these operators, we propose a domain-specific language to define analytical workflows. The Gradoop graph store is currently utilizing HBase for distributed storage of graph data in Hadoop clusters. An initial version of Gradoop has been used to analyze graph data for business intelligence and social network analysis., Comment: Technical Report

Product counterfeiting is a serious problem causing the industry estimated losses of billions of dollars every year. With the increasing spread of e-commerce, the number of counterfeit products sold online increased substantially. We propose the adoption of a semi-automatic workflow to identify likely counterfeit offers in online platforms and to present these offers to a domain expert for manual verification. The workflow includes steps to generate search queries for relevant product offers, to match and cluster similar product offers, and to assess the counterfeit suspiciousness based on different criteria. The goal is to support the periodic identification of many counterfeit offers with a limited amount of manual effort. We explain how the proposed approach can be realized. We also present a preliminary evaluation of its most important steps on a case study using the eBay platform.

Mappings between related ontologies are increasingly used to support data integration and analysis tasks. Changes in the ontologies also require the adaptation of ontology mappings. So far the evolution of ontology mappings has received little attention albeit ontologies change continuously especially in the life sciences. We therefore analyze how mappings between popular life science ontologies evolve for different match algorithms. We also evaluate which semantic ontology changes primarily affect the mappings. We further investigate alternatives to predict or estimate the degree of future mapping changes based on previous ontology and mapping transitions., Comment: Keywords: mapping evolution, ontology matching, ontology evolution

Mapping complex metadata structures is crucial in a number of domains such as data integration, ontology alignment or model management. To speed up that process automatic matching systems were developed to compute mapping suggestions that can be corrected by a user. However, constructing and tuning match strategies still requires a high manual effort by matching experts as well as correct mappings to evaluate generated mappings. We therefore propose a self-configuring schema matching system that is able to automatically adapt to the given mapping problem at hand. Our approach is based on analyzing the input schemas as well as intermediate matching results. A variety of matching rules use the analysis results to automatically construct and adapt an underlying matching process for a given match task. We comprehensively evaluate our approach on different mapping problems from the schema, ontology and model management domains. The evaluation shows that our system is able to robustly return good quality mappings across different mapping problems and domains., Comment: 10 Pages

The effectiveness and scalability of MapReduce-based implementations of complex data-intensive tasks depend on an even redistribution of data between map and reduce tasks. In the presence of skewed data, sophisticated redistribution approaches thus become necessary to achieve load balancing among all reduce tasks to be executed in parallel. For the complex problem of entity resolution, we propose and evaluate two approaches for such skew handling and load balancing. The approaches support blocking techniques to reduce the search space of entity resolution, utilize a preprocessing MapReduce job to analyze the data distribution, and distribute the entities of large blocks among multiple reduce tasks. The evaluation on a real cloud infrastructure shows the value and effectiveness of the proposed load balancing approaches.

The proliferation of ontologies and taxonomies in many domains increasingly demands the integration of multiple such ontologies. The goal of ontology integration is to merge two or more given ontologies in order to provide a unified view on the input ontologies while maintaining all information coming from them. We propose a new taxonomy merging algorithm that, given as input two taxonomies and an equivalence matching between them, can generate an integrated taxonomy in a fully automatic manner. The approach is target-driven, i.e. we merge a source taxonomy into the target taxonomy and preserve the structure of the target ontology as much as possible. We also discuss how to extend the merge algorithm providing auxiliary information, like additional relationships between source and target concepts, in order to semantically improve the final result. The algorithm was implemented in a working prototype and evaluated using synthetic and real-world scenarios.

Cloud infrastructures enable the efficient parallel execution of data-intensive tasks such as entity resolution on large datasets. We investigate challenges and possible solutions of using the MapReduce programming model for parallel entity resolution. In particular, we propose and evaluate two MapReduce-based implementations for Sorted Neighborhood blocking that either use multiple MapReduce jobs or apply a tailored data replication.