Robert Aufschläger

Mr. Robert Aufschläger

Technology Campus Vilshofen, Deggendorf Institute of Technology, Deggendorf, Bavaria, Germany
orcid
Research Assistant

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 September 2023 in Information
Verified
Journal Article
Open access
PDF file
RARobert Aufschläger
Robert Aufschläger
JFJakob Folz
Jakob Folz
EMElena März
Elena März
JGJohann Guggumos Guggumos
Johann Guggumos Guggumos
MHMichael Heigl
Michael Heigl
MSMartin Schramm
Martin Schramm
In the European Union, Data Controllers and Data Processors, who work with personal data, have to comply with the General Data Protection Regulation and other applicable laws. This affects the storing and processing of personal data. But some data processing in data mining or statistical analyses does not require any personal reference to the data. Thus, personal context can be removed. For these use cases, to comply with applicable laws, any existing personal information has to be removed by applying the so-called anonymization. However, anonymization should maintain data utility. Therefore, the concept of anonymization is a double-edged sword with an intrinsic trade-off: privacy enforcement vs. utility preservation. The former might not be entirely guaranteed when anonymized data are published as Open Data. In theory and practice, there exist diverse approaches to conduct and score anonymization. This explanatory synthesis discusses the technical perspectives on the anonymization of tabular data with a special emphasis on the European Union’s legal base. The studied methods for conducting anonymization, and scoring the anonymization procedure and the resulting anonymity are explained in unifying terminology. The examined methods and scores cover both categorical and numerical data. The examined scores involve data utility, information preservation, and privacy models. In practice-relevant examples, methods and scores are experimentally tested on records from the UCI Machine Learning Repository’s “Census Income (Adult)” dataset.

16 March 2025
Verified
Book Chapter
Open access
PDF file
RARobert Aufschläger
Robert Aufschläger
SWSebastian Wilhelm
Sebastian Wilhelm
MHMichael Heigl
Michael Heigl
MSMartin Schramm
Martin Schramm
This work introduces , an anonymization pipeline that can be used for generalization and suppression-based anonymization of nominal textual tabular data. It automatically generates value generalization hierarchies (VGHs) that, in turn, can be used to generalize attributes in quasi-identifiers. The pipeline leverages embeddings to generate semantically close value generalizations through iterative clustering. We applied KMeans and Hierarchical Agglomerative Clustering on 13 different predefined text embeddings (both open and closed-source (via APIs)). Our approach is experimentally tested on a well-known benchmark dataset for anonymization: The UCI Machine Learning Repository’s Adult dataset. supports anonymization procedures by offering more possibilities compared to using arbitrarily chosen VGHs. Experiments demonstrate that these VGHs can outperform manually constructed ones in terms of downstream efficacy (especially for small k-anonymity) and therefore can foster the quality of anonymized datasets. Our implementation is made public.

22 April 2025 in IEEE Access
Verified
Journal Article
Open access
JFJakob Folz
Jakob Folz
MVManjitha D. Vidanalage
Manjitha D. Vidanalage
RARobert Aufschläger
Robert Aufschläger
AAAmar Almaini
Amar Almaini
MHMichael Heigl
Michael Heigl
MSMartin Schramm
Martin Schramm
This study introduces the System for Calculating Open Data Re-identification Risk (SCORR), a framework for quantifying privacy risks in tabular datasets. SCORR extends conventional metrics such as k-anonymity, l-diversity, and t-closeness with novel extended metrics, including uniqueness-only risk, uniformity-only risk, correlation-only risk, and Markov Model risk, to identify a broader range of re-identification threats. It efficiently analyses event-level and person-level datasets with categorical and numerical attributes. Experimental evaluations were conducted on three publicly available datasets: OULAD, HID, and Adult, across multiple anonymisation levels. The results indicate that higher anonymisation levels do not always proportionally enhance privacy. While stronger generalisation improves k-anonymity, l-diversity and t-closeness vary significantly across datasets. Uniqueness-only and uniformity-only risk decreased with anonymisation, whereas correlation-only risk remained high. Meanwhile, Markov Model risk consistently remained high, indicating little to no improvement regardless of the anonymisation level. Scalability analysis revealed that conventional metrics and Uniqueness-only risk incurred minimal computational overhead, remaining independent of dataset size. However, correlation-only and uniformity-only risk required significantly more processing time, while Markov Model risk incurred the highest computational cost. Despite this, all metrics remained unaffected by the number of quasi-identifiers, except t-closeness, which scaled linearly beyond a certain threshold. A usability evaluation comparing SCORR with the freely available ARX Tool showed that SCORR reduced the number of user interactions required for risk analysis by 59.38%, offering a more streamlined and efficient process. These results confirm SCORR’s effectiveness in helping data custodians balance privacy protection and data utility, advancing privacy risk assessment beyond existing tools.

17 December 2024 in arXiv
Verified
Preprint
Open access
PDF file
RARobert Aufschläger
Robert Aufschläger
SWSebastian Wilhelm
Sebastian Wilhelm
MHMichael Heigl
Michael Heigl
MSMartin Schramm
Martin Schramm
This work introduces ClustEm4Ano, an anonymization pipeline that can be used for generalization and suppression-based anonymization of nominal textual tabular data. It automatically generates value generalization hierarchies (VGHs) that, in turn, can be used to generalize attributes in quasi-identifiers. The pipeline leverages embeddings to generate semantically close value generalizations through iterative clustering. We applied KMeans and Hierarchical Agglomerative Clustering on $13$ different predefined text embeddings (both open and closed-source (via APIs)). Our approach is experimentally tested on a well-known benchmark dataset for anonymization: The UCI Machine Learning Repository's Adult dataset. ClustEm4Ano supports anonymization procedures by offering more possibilities compared to using arbitrarily chosen VGHs. Experiments demonstrate that these VGHs can outperform manually constructed ones in terms of downstream efficacy (especially for small $k$-anonymity ($2 \leq k \leq 30$)) and therefore can foster the quality of anonymized datasets. Our implementation is made public.

4 February 2026 in arXiv
Verified
Preprint
Open access
PDF file
GSGautam Savaliya
Gautam Savaliya
RARobert Aufschläger
Robert Aufschläger
ASAbhishek Subedi
Abhishek Subedi
MHMichael Heigl
Michael Heigl
MSMartin Schramm
Martin Schramm
Artificial intelligence systems introduce complex privacy risks throughout their lifecycle, especially when processing sensitive or high-dimensional data. Beyond the seven traditional privacy threat categories defined by the LINDDUN framework, AI systems are also exposed to model-centric privacy attacks such as membership inference and model inversion, which LINDDUN does not cover. To address both classical LINDDUN threats and additional AI-driven privacy attacks, PriMod4AI introduces a hybrid privacy threat modeling approach that unifies two structured knowledge sources, a LINDDUN knowledge base representing the established taxonomy, and a model-centric privacy attack knowledge base capturing threats outside LINDDUN. These knowledge bases are embedded into a vector database for semantic retrieval and combined with system level metadata derived from Data Flow Diagram. PriMod4AI uses retrieval-augmented and Data Flow specific prompt generation to guide large language models (LLMs) in identifying, explaining, and categorizing privacy threats across lifecycle stages. The framework produces justified and taxonomy-grounded threat assessments that integrate both classical and AI-driven perspectives. Evaluation on two AI systems indicates that PriMod4AI provides broad coverage of classical privacy categories while additionally identifying model-centric privacy threats. The framework produces consistent, knowledge-grounded outputs across LLMs, as reflected in agreement scores in the observed range.