Peer-reviewed papers
Wissenschaftliche Veröffentlichungen und Beiträge aus dem Forschungsnetzwerk Anonymisierung.
Alle Publikationen
DRAMatic Speedup: Accelerating HE Operations on a Processing-in-Memory System
Autor:innen
Niklas Klinger and Jonas Sander and Peterson Yuhala and Pascal Felber and Thomas Eisenbarth
DP-Hype: Federated Differentially Private Hyperparameter Search
Johannes Liebenow and Thorsten Peinemann and Esfandiar Mohammadi
SLasH-DSA: Breaking SLH-DSA Using an Extensible End-To-End Rowhammer Framework
Jeremy Boy and Antoon Purnal and Anna Pätschke and Luca Wilke and Thomas Eisenbarth
ReDASH: Fast and Efficient Scaling in Arithmetic Garbled Circuits for Secure Outsourced Inference
Felix Maurer and Jonas Sander and Thomas Eisenbarth
Non-omniscient backdoor injection with one poison sample: Proving the one-poison hypothesis for linear regression, linear classification, and 2-layer ReLU neural networks
Thorsten Peinemann and Paula Arnold and Sebastian Berndt and Thomas Eisenbarth and Esfandiar Mohammadi
Lifted Model Construction without Normalisation: A Vectorised Approach to Exploit Symmetries in Factor Graphs
Compression Versus Accuracy: A Hierarchy of Lifted Models
StaRAI: From a Probabilistic Propositional Model to a Highly Compressed Probabilistic Relational Model (Extended Abstract)
CHAI+FCR 2025 Humanities-Centred Artificial Intelligence 2025 and Formal & Cognitive Reasoning 2025
Approximate Lifted Model Construction
Improving Statistical Privacy by Subsampling
AnonyPyx: A Python Library for Data Anonymization
Short Summary of Syntactic Privacy
Mixnets on a Tightrope: Quantifying the Leakage of Mix Networks Using a Provably Optimal Heuristic Adversary
TDXploit: Novel Techniques for Single-Stepping and Cache Attacks on Intel TDX
Prompt Pirates Need a Map: Stealing Seeds helps Stealing Prompts
Silenzio: Secure Non-Interactive Outsourced MLP Training
BarkBeetle: Stealing Decision Tree Models with Fault Injection
Attacks and Remedies for Randomness in AI: Cryptanalysis of PHILOX and THREEFRY
DP-HYPE: Distributed Differentially Private Hyperparameter Search
The AI Act and Some Implications for Developing AI-Based Systems
MammothDP: Differentially Private Boosted Decision Trees, Hyperparameter-Free and Ready for Trusted Hardware
Optimizing Systolic Array-based NTT Accelerators
Bag-Level Multiple Instance Learning for Acute Stress Detection from Video Data
Statistical Privacy
Dash: Accelerating Distributed Private Convolutional Neural Network Inference with Arithmetic Garbled Circuits
Estimating Causal Effects in Partially Directed Parametric Causal Factor Graphs
Lifting factor graphs with some unknown factors for new individuals
Differentially Private Inductive Miner
Efficient Detection of Commutative Factors in Factor Graphs
I Want To Fault My BIKE: On the Feasibility of Electromagnetic Fault Injection Attacks Against The BIKE Cryptosystem
Automated Video-Based Approach for the Diagnosis of Tourette Syndrome
TDXdown: Single-Stepping and Instruction Counting Attacks against Intel TDX
SNPGuard: Remote Attestation of SEV-SNP VMs Using Open Source Tools
Evaluating the Impact of Face Anonymization Methods on Computer Vision Tasks: A Trade-off Between Privacy and Utility
Slalom at the Carnival: Privacy-preserving Inference with Masks from Public Knowledge
S-BDT: Distributed Differentially Private Boosted Decision Trees
DPM: Clustering Sensitive Data through Separation
Towards Privacy-Preserving Relational Data Synthesis via Probabilistic Relational Models
Distributed DP-Helmet: Scalable Differentially Private Non-interactive Averaging of Single Layers
Revealing Unintentional Information Leakage in Low-Dimensional Facial Portrait Representations
Colour Passing Revisited: Lifted Model Construction with Commutative Factors
Lifted Causal Inference in Relational Domains
Efficient Detection of Exchangeable Factors in Factor Graphs
Lifting in Support of Privacy-Preserving Probabilistic Inference
Towards Privacy and Utility in Tourette Tic Detection Through Pretraining Based on Publicly Available Video Data of Healthy Subjects
Lifting Factor Graphs with Some Unknown Factors
DP-Hype: Federated Differentially Private Hyperparameter Search
BibTeX anzeigen
@article{liebenow2026dphype,
title = {DP-Hype: Federated Differentially Private Hyperparameter Search},
author = {Johannes Liebenow and Thorsten Peinemann and Esfandiar Mohammadi},
journal = {Proceedings on Privacy Enhancing Technologies},
year = {2026},
url = {https://doi.org/10.56553/popets-2026-0075},
}DRAMatic Speedup: Accelerating HE Operations on a Processing-in-Memory System
BibTeX anzeigen
@misc{klinger2026dramaticspeedupacceleratingoperations,
title = {DRAMatic Speedup: Accelerating HE Operations on a Processing-in-Memory System},
author = {Niklas Klinger and Jonas Sander and Peterson Yuhala and Pascal Felber and Thomas Eisenbarth},
year = {2026},
url = {https://arxiv.org/abs/2602.12433},
}SLasH-DSA: Breaking SLH-DSA Using an Extensible End-To-End Rowhammer Framework
BibTeX anzeigen
@inproceedings{boySLasHDSABreakingSLHDSA2026,
title = {SLasH-DSA: Breaking SLH-DSA Using an Extensible End-To-End Rowhammer Framework},
author = {Jeremy Boy and Antoon Purnal and Anna Pätschke and Luca Wilke and Thomas Eisenbarth},
year = {2026},
}ReDASH: Fast and Efficient Scaling in Arithmetic Garbled Circuits for Secure Outsourced Inference
Abstract
"ReDash extends Dash's arithmetic garbled circuits to provide a more flexible and efficient framework for secure outsourced inference. By introducing a novel garbled scaling gadget based on a generalized base extension for the residue number system, ReDash removes Dash's limitation of scaling exclusively by powers of two. This enables arbitrary scaling factors drawn from the residue number system's modular base, allowing for tailored quantization schemes and more efficient model evaluation."
BibTeX anzeigen
@inproceedings{10.1007/978-3-032-01823-6_3,
title = {ReDASH: Fast and Efficient Scaling in Arithmetic Garbled Circuits for Secure Outsourced Inference},
author = {Felix Maurer and Jonas Sander and Thomas Eisenbarth},
year = {2026},
}Non-omniscient backdoor injection with one poison sample: Proving the one-poison hypothesis for linear regression, linear classification, and 2-layer ReLU neural networks
BibTeX anzeigen
@misc{peinemann2026nonomniscientbackdoorinjectionpoison,
title = {Non-omniscient backdoor injection with one poison sample: Proving the one-poison hypothesis for linear regression, linear classification, and 2-layer ReLU neural networks},
author = {Thorsten Peinemann and Paula Arnold and Sebastian Berndt and Thomas Eisenbarth and Esfandiar Mohammadi},
year = {2026},
url = {https://arxiv.org/abs/2508.05600},
}Lifted Model Construction without Normalisation: A Vectorised Approach to Exploit Symmetries in Factor Graphs
Abstract
"Lifted probabilistic inference exploits symmetries in a probabilistic model to allow for tractable probabilistic inference with respect to domain sizes of logical variables. We found that the current state-of-the-art algorithm to construct a lifted representation in form of a parametric factor graph misses symmetries between factors that are exchangeable but scaled differently, thereby leading to a less compact representation. In this paper, we propose a generalisation of the advanced colour passing (ACP) algorithm, which is the state of the art to construct a parametric factor graph. Our proposed algorithm allows for potentials of factors to be scaled arbitrarily and efficiently detects more symmetries than the original ACP algorithm. By detecting strictly more symmetries than ACP, our algorithm significantly reduces online query times for probabilistic inference when the resulting model is applied, which we also confirm in our experiments. "
BibTeX anzeigen
@conference{pmlr-v269-luttermann25a,
title = {Lifted Model Construction without Normalisation: A Vectorised Approach to Exploit Symmetries in Factor Graphs},
author = {Malte Luttermann and Ralf Möller and Marcel Gehrke},
year = {2025},
url = {https://proceedings.mlr.press/v269/luttermann25a.html
https://raw.githubusercontent.com/mlresearch/v269/main/assets/luttermann25a/luttermann25a.pdf},
}Compression Versus Accuracy: A Hierarchy of Lifted Models
Abstract
"Proceedings of the Twenty-Eighth European Conference on Artificial Intelligence (ECAI-2025)."
BibTeX anzeigen
@incollection{Speller2025b,
title = {Compression Versus Accuracy: A Hierarchy of Lifted Models},
author = {Jan Speller and Malte Luttermann and Marcel Gehrke and Tanya Braun},
year = {2025},
url = {https://ebooks.iospress.nl/doi/10.3233/FAIA251420},
}StaRAI: From a Probabilistic Propositional Model to a Highly Compressed Probabilistic Relational Model (Extended Abstract)
BibTeX anzeigen
@inproceedings{Gehrke2025a,
title = {StaRAI: From a Probabilistic Propositional Model to a Highly Compressed Probabilistic Relational Model (Extended Abstract)},
author = {Marcel Gehrke and Malte Luttermann},
year = {2025},
url = {https://hal.science/hal-05294280},
}CHAI+FCR 2025 Humanities-Centred Artificial Intelligence 2025 and Formal & Cognitive Reasoning 2025
Abstract
"Explainability of Approximate Lifted Model Construction: A Geometric Perspective. Proceedings of the First Joint Workshop on Humanities-Centered Artificial Intelligence and Formal & Cognitive Reasoning (CHAI-2025 and FCR-2025) co-located with 48th German Conference on Artificial Intelligence (KI-2025)"
BibTeX anzeigen
@conference{Speller2025,
title = {CHAI+FCR 2025 Humanities-Centred Artificial Intelligence 2025 and Formal & Cognitive Reasoning 2025},
author = {Jan Speller and Malte Luttermann and Marcel Gehrke and Tanya Braun},
year = {2025},
url = {https://ceur-ws.org/Vol-4058/},
}Approximate Lifted Model Construction
Abstract
"Probabilistic relational models such as parametric factor graphs enable efficient (lifted) inference by exploiting the indistinguishability of objects. In lifted inference, a representative of indistinguishable objects is used for computations. To obtain a relational (i.e., lifted) representation, the Advanced Colour Passing (ACP) algorithm is the state of the art. The ACP algorithm, however, requires underlying distributions, encoded as potential-based factorisations, to exactly match to identify and exploit indistinguishabilities. Hence, ACP is unsuitable for practical applications where potentials learned from data inevitably deviate even if associated objects are indistinguishable. To mitigate this problem, we introduce the ε-Advanced Colour Passing (ε-ACP) algorithm, which allows for a deviation of potentials depending on a hyperparameter ε. ε-ACP efficiently uncovers and exploits indistinguishabilities that are not exact. We prove that the approximation error induced by ε-ACP is strictly bounded and our experiments show that the approximation error is close to zero in practice. "
BibTeX anzeigen
@conference{ijcai2025p1009,
title = {Approximate Lifted Model Construction},
author = {Malte Luttermann and Jan Speller and Marcel Gehrke and Tanya Braun and Ralf Möller and Mattis Hartwig},
year = {2025},
}Improving Statistical Privacy by Subsampling
Abstract
"Differential privacy (DP) considers a scenario, where an adversary has almost complete information about the entries of a database This worst-case assumption is likely to overestimate the privacy thread for an individual in real life. Statistical privacy (SP) denotes a setting where only the distribution of the database entries is known to an adversary, but not their exact values. In this case one has to analyze the interaction between noiseless privacy based on the entropy of distributions and privacy mechanisms that distort the answers of queries, which can be quite complex. A privacy mechanism often used is to take samples of the data for answering a query. This paper proves precise bounds how much different methods of sampling increase privacy in the statistical setting with respect to database size and sampling rate. They allow us to deduce when and how much sampling provides an improvement and how far this depends on the privacy parameter {epsilon}. To perform these investigations we develop a framework to model sampling techniques. For the DP setting tradeoff functions have been proposed as a finer measure for privacy compared to ({epsilon},{delta})-pairs. We apply these tools to statistical privacy with subsampling to get a comparable characterization "
BibTeX anzeigen
@techreport{Breutigam2025subsampling,
title = {Improving Statistical Privacy by Subsampling},
author = {Dennis Breutigam and Rüdiger Reischuk},
year = {2025},
url = {https://arxiv.org/abs/2504.11429},
}AnonyPyx: A Python Library for Data Anonymization
Abstract
"AnonyPyx is a Python library that provides traditional anonymization methods, such as k-anonymity and l-diversity. It also includes attacks on these methods and provides appropriate privacy and utility metrics. The library is a fork of the open-source AnonyPy project, with significant modifications and additions."
BibTeX anzeigen
@misc{Zapatka2023anonypyx,
title = {AnonyPyx: A Python Library for Data Anonymization},
author = {Niklas Zapatka and Taisuke Fujita},
year = {2025},
url = {https://github.com/questforwisdom/anonypyx},
}Short Summary of Syntactic Privacy
Abstract
"Balancing the trade-off between personal data processing and the right to privacy requires a measure for privacy. Such measures are provided by formal models, which also enable proofs of privacy preservation. Syntactic privacy has been praised as an intuitive family of such models. This report reviews the literature on syntactic privacy models and their shortcomings. The results reveal their inherent complexity and fragility. A short review of open-source implementations supporting syntactic privacy is also included."
BibTeX anzeigen
@techreport{Zapatka2023syntactic,
title = {Short Summary of Syntactic Privacy},
author = {Niklas Zapatka and Joshua Stock and Hannes Federrath and Jens Lindemann},
year = {2025},
url = {https://svs.informatik.uni-hamburg.de/publications/2023/working-paper-syntactic-privacy-literature-review.pdf},
}Mixnets on a Tightrope: Quantifying the Leakage of Mix Networks Using a Provably Optimal Heuristic Adversary
BibTeX anzeigen
@inproceedings{11023277,
title = { Mixnets on a Tightrope: Quantifying the Leakage of Mix Networks Using a Provably Optimal Heuristic Adversary },
author = {Sebastian Meiser and Debajyoti Das and Moritz Kirschte and Esfandiar Mohammadi and Aniket Kate},
year = {2025},
url = {https://doi.ieeecomputersociety.org/10.1109/SP61157.2025.00233},
}TDXploit: Novel Techniques for Single-Stepping and Cache Attacks on Intel TDX
BibTeX anzeigen
@inproceedings{DBLP:conf/uss/RauscherWW0G25,
title = {TDXploit: Novel Techniques for Single-Stepping and Cache Attacks on
Intel TDX},
author = {Fabian Rauscher and Luca Wilke and Hannes Weissteiner and Thomas Eisenbarth and Daniel Gruss},
year = {2025},
url = {https://www.usenix.org/conference/usenixsecurity25/presentation/rauscher},
}Prompt Pirates Need a Map: Stealing Seeds helps Stealing Prompts
BibTeX anzeigen
@misc{mächtle2025promptpiratesneedmap,
title = {Prompt Pirates Need a Map: Stealing Seeds helps Stealing Prompts},
author = {Felix Mächtle and Ashwath Shetty and Jonas Sander and Nils Loose and Sören Pirk and Thomas Eisenbarth},
year = {2025},
url = {https://arxiv.org/abs/2509.09488},
}Silenzio: Secure Non-Interactive Outsourced MLP Training
BibTeX anzeigen
@misc{sander2025silenziosecurenoninteractiveoutsourced,
title = {Silenzio: Secure Non-Interactive Outsourced MLP Training},
author = {Jonas Sander and Thomas Eisenbarth},
year = {2025},
url = {https://arxiv.org/abs/2504.17785},
}BarkBeetle: Stealing Decision Tree Models with Fault Injection
BibTeX anzeigen
@misc{wang2025barkbeetlestealingdecisiontree,
title = {BarkBeetle: Stealing Decision Tree Models with Fault Injection},
author = {Qifan Wang and Jonas Sander and Minmin Jiang and Thomas Eisenbarth and David Oswald},
year = {2025},
url = {https://arxiv.org/abs/2507.06986},
}Attacks and Remedies for Randomness in AI: Cryptanalysis of PHILOX and THREEFRY
BibTeX anzeigen
@misc{cryptoeprint:2025/2161,
title = {Attacks and Remedies for Randomness in AI: Cryptanalysis of PHILOX and THREEFRY},
author = {Jens Alich and Thomas Eisenbarth and Hossein Hadipour and Gregor Leander and Felix Mächtle and Yevhen Perehuda and Shahram Rasoolzadeh and Jonas Sander and Cihangir Tezcan},
year = {2025},
url = {https://eprint.iacr.org/2025/2161},
}DP-HYPE: Distributed Differentially Private Hyperparameter Search
BibTeX anzeigen
@misc{liebenow2025dphypedistributeddifferentiallyprivate,
title = {DP-HYPE: Distributed Differentially Private Hyperparameter Search},
author = {Johannes Liebenow and Thorsten Peinemann and Esfandiar Mohammadi},
year = {2025},
url = {https://arxiv.org/abs/2510.04902},
}The AI Act and Some Implications for Developing AI-Based Systems
Abstract
"This paper presents several challenges when developing AI-based software systems for potentially safety-critical domains in the European jurisdiction. Starting with the legal developments in the European Union, especially the so-called AI act, we recall the idea of the risk-based approach and its implications for the software engineering development process. We state the benefits of using formal methods for supporting the design and analysis steps in a corresponding development process. We identify research challenges, especially when aiming to build trustworthy AI based systems."
BibTeX anzeigen
@incollection{DBLP:conf/birthday/Leucker25,
title = {The AI Act and Some Implications for Developing AI-Based Systems},
author = {Martin Leucker},
year = {2025},
url = {https://doi.org/10.1007/978-3-031-73887-6_15},
}MammothDP: Differentially Private Boosted Decision Trees, Hyperparameter-Free and Ready for Trusted Hardware
BibTeX anzeigen
@unpublished{kirschte2025mammothdp,
title = {MammothDP: Differentially Private Boosted Decision Trees, Hyperparameter-Free and Ready for Trusted Hardware},
author = {Moritz Kirschte and Thorsten Peinemann and Kari Kostiainen and Jan Wichelmann and Thomas Eisenbarth and Esfandiar Mohammadi},
year = {2025},
url = {/uploads/2025/03/MammothDP-18.pdf},
}Optimizing Systolic Array-based NTT Accelerators
BibTeX anzeigen
@article{schultz2025optimizingb,
title = {Optimizing Systolic Array-based NTT Accelerators},
author = {Saleh Mulhem and Eike Schultz and Lukas Groth and Mladen Berekovic and Rainer Buchty},
journal = {Authorea Preprints, Accepted to IEEE Embedded Systems Letters},
year = {2025},
url = {https://www.techrxiv.org/doi/full/10.36227/techrxiv.174137763.39532014},
}Bag-Level Multiple Instance Learning for Acute Stress Detection from Video Data
BibTeX anzeigen
@inproceedings{Bruegge2025,
title = {Bag-Level Multiple Instance Learning for Acute Stress Detection from Video Data},
author = {Nele Sophie Brügge and Alexandra Korda and Stefan Borgwardt and Christina Andreou and Giorgos Giannakakis and Heinz Handels},
year = {2025},
url = {https://www.scitepress.org/Link.aspx?doi=10.5220/0013364900003911},
}Statistical Privacy
BibTeX anzeigen
@misc{breutigam2025statisticalprivacy,
title = {Statistical Privacy},
author = {Dennis Breutigam and Rüdiger Reischuk},
year = {2025},
url = {https://arxiv.org/abs/2501.12893},
}Dash: Accelerating Distributed Private Convolutional Neural Network Inference with Arithmetic Garbled Circuits
BibTeX anzeigen
@article{sander2025dash,
title = {Dash: Accelerating Distributed Private Convolutional Neural Network Inference with Arithmetic Garbled Circuits},
author = {Jonas Sander and Sebastian Berndt and Ida Bruhns and Thomas Eisenbarth},
journal = {IACR Transactions on Cryptographic Hardware and Embedded Systems},
year = {2025},
url = {https://tches.iacr.org/index.php/TCHES/article/view/11935},
}Estimating Causal Effects in Partially Directed Parametric Causal Factor Graphs
Abstract
"Lifting uses a representative of indistinguishable individuals to exploit symmetries in probabilistic relational models, denoted as parametric factor graphs, to speed up inference while maintaining exact answers. In this paper, we show how lifting can be applied to causal inference in partially directed graphs, i.e., graphs that contain both directed and undirected edges to represent causal relationships between directed and undirected edges to represent causal relationships between random variables. We present partially directed parametric causal factor graphs (PPCFGs) as a generalisation of previously introduced parametric causal factor graphs, which require a fully directed graph. We further show how causal inference can be performed on a lifted level in PPCFGs, thereby extending the applicability of lifted causal inference to a broader range of models requiring less prior knowledge about causal relationships."
BibTeX anzeigen
@inproceedings{10.1007/978-3-031-76235-2_20,
title = {Estimating Causal Effects in Partially Directed Parametric Causal Factor Graphs},
author = {Malte Luttermann and Tanya Braun and Ralf Möller and Marcel Gehrke},
year = {2025},
}Lifting factor graphs with some unknown factors for new individuals
Abstract
"Lifting exploits symmetries in probabilistic graphical models by using a representative for indistinguishable objects, allowing to carry out query answering more efficiently while maintaining exact answers. In this paper, we investigate how lifting enables us to perform probabilistic inference for factor graphs containing unknown factors, i.e., factors whose underlying function of potential mappings is unknown. We present the Lifting Factor Graphs with Some Unknown Factors (LIFAGU) algorithm to identify indistinguishable subgraphs in a factor graph containing unknown factors, thereby enabling the transfer of known potentials to unknown potentials to ensure a well-defined semantics of the model and allow for (lifted) probabilistic inference. We further extend LIFAGU to incorporate additional background knowledge about groups of factors belonging to the same individual object. By incorporating such background knowledge, LIFAGU is able to further reduce the ambiguity of possible transfers of known potentials to unknown potentials."
BibTeX anzeigen
@article{LUTTERMANN2025109371,
title = {Lifting factor graphs with some unknown factors for new individuals},
author = {Malte Luttermann and Ralf Möller and Marcel Gehrke},
journal = {International Journal of Approximate Reasoning},
year = {2025},
url = {https://www.sciencedirect.com/science/article/pii/S0888613X2500012X},
}Differentially Private Inductive Miner
Abstract
"Protecting personal data about individuals, such as event traces in process mining, is an inherently difficult task since an event trace leaks information about the path in a process model that an individual has triggered. Yet, prior anonymization methods of event traces like k-anonymity or event log sanitization struggled to protect against such leakage, in particular against adversaries with sufficient background knowledge. In this work, we provide a method that tackles the challenge of summarizing sensitive event traces by learning the underlying process tree in a privacy-preserving manner. We prove via the so-called Differential Privacy (DP) property that from the resulting summaries no useful inference can be drawn about any personal data in an event trace. On the technical side, we introduce a differentially private approximation (DPIM) of the Inductive Miner. Experimentally, we compare our DPIM with the Inductive Miner on 14 real-world event traces by evaluating well-known metrics: fitness, precision, simplicity, and generalization. The experiments show that our DPIM not only protects personal data but also generates faithful process trees that exhibit little utility loss above the Inductive Miner."
BibTeX anzeigen
@inproceedings{10680684,
title = {Differentially Private Inductive Miner},
author = {Max Schulze and Yorck Zisgen and Moritz Kirschte and Esfandiar Mohammadi and Agnes Koschmider},
year = {2024},
}Efficient Detection of Commutative Factors in Factor Graphs
Abstract
"Lifted probabilistic inference exploits symmetries in probabilistic graphical models to allow for tractable probabilistic inference with respect to domain sizes. To exploit symmetries in, e.g., factor graphs, it is crucial to identify commutative factors, i.e., factors having symmetries within themselves due to their arguments being exchangeable. The current state-of-the-art to check whether a factor is commutative with respect to a subset of its arguments iterates over all possible subsets of the factor’s arguments, i.e., O($2^n$) iterations for a factor with n arguments in the worst case. In this paper, we efficiently solve the problem of detecting commutative factors in a factor graph. In particular, we introduce the detection of commutative factors (DECOR) algorithm, which allows us to drastically reduce the computational effort for checking whether a factor is commutative in practice. We prove that DECOR efficiently identifies restrictions to drastically reduce the number of required iterations and validate the efficiency of DECOR in our empirical evaluation."
BibTeX anzeigen
@inproceedings{pmlr-v246-luttermann24a,
title = {Efficient Detection of Commutative Factors in Factor Graphs},
author = {Malte Luttermann and Johann Machemer and Marcel Gehrke},
year = {2024},
url = {https://proceedings.mlr.press/v246/luttermann24a.html},
}I Want To Fault My BIKE: On the Feasibility of Electromagnetic Fault Injection Attacks Against The BIKE Cryptosystem
BibTeX anzeigen
@techreport{boy2024bike,
title = {I Want To Fault My BIKE: On the Feasibility of Electromagnetic Fault Injection Attacks Against The BIKE Cryptosystem},
author = {Jeremy Boy and Jack Mähl and Robin Sehm},
year = {2024},
url = { https://www.its.uni-luebeck.de/fileadmin/files/documents/CS_JeremyBoy_JackMaehl_RobinSehm_I_Want_To_Fault_My_BIKE.pdf},
}Automated Video-Based Approach for the Diagnosis of Tourette Syndrome
BibTeX anzeigen
@article{Schappert2024,
title = {Automated Video-Based Approach for the Diagnosis of Tourette Syndrome},
author = {Ronja Schappert and Julius Verrel and Nele Sophie Brügge and Frédéric Li and Theresa Paulus and Leonie Becker and Tobias Bäumer and Christian Beste and Veit Roessner and Sebastian Fudickar and Alexander Münchau},
journal = {Movement disorders clinical practice},
year = {2024},
url = {https://movementdisorders.onlinelibrary.wiley.com/doi/10.1002/mdc3.14158},
}TDXdown: Single-Stepping and Instruction Counting Attacks against Intel TDX
BibTeX anzeigen
@inproceedings{wilke2024tdxdown,
title = {TDXdown: Single-Stepping and Instruction Counting Attacks against Intel TDX},
author = {Luca Wilke and Florian Sieck and Thomas Eisenbarth},
year = {2024},
url = {https://dl.acm.org/doi/10.1145/3658644.3690230},
}SNPGuard: Remote Attestation of SEV-SNP VMs Using Open Source Tools
BibTeX anzeigen
@inproceedings{wilke2024snpguard,
title = {SNPGuard: Remote Attestation of SEV-SNP VMs Using Open Source Tools},
author = {Luca Wilke and Gianluca Scopelliti},
year = {2024},
url = {https://ieeexplore.ieee.org/document/10628964/},
}Evaluating the Impact of Face Anonymization Methods on Computer Vision Tasks: A Trade-off Between Privacy and Utility
BibTeX anzeigen
@article{stenger2024evaluating,
title = {Evaluating the Impact of Face Anonymization Methods on Computer Vision Tasks: A Trade-off Between Privacy and Utility},
author = {Roland Stenger and others},
journal = {IEEE Access},
year = {2024},
url = {https://www.techrxiv.org/articles/preprint/Evaluating_the_Impact_of_Face_Anonymization_Methods_on_Computer_Vision_Tasks_A_Trade-off_Between_Privacy_and_Utility/1222287},
}Slalom at the Carnival: Privacy-preserving Inference with Masks from Public Knowledge
BibTeX anzeigen
@article{bruhns2024slalom,
title = {Slalom at the Carnival: Privacy-preserving Inference with Masks from Public Knowledge},
author = {Ida Bruhns and Sebastian Berndt and Jonas Sander and Thomas Eisenbarth},
journal = {IACR Communications in Cryptology},
year = {2024},
url = {https://cic.iacr.org/p/1/3/40},
}S-BDT: Distributed Differentially Private Boosted Decision Trees
Abstract
"We introduce S-BDT: a novel (ε,δ)-differentially private distributed gradient boosted decision tree (GBDT) learner that improves the protection of single training data points (privacy) while achieving meaningful learning goals, such as accuracy or regression error (utility). S-BDT uses less noise by relying on non-spherical multivariate Gaussian noise, for which we show tight subsampling bounds for privacy amplification and incorporate that into a Rényi filter for individual privacy accounting. We experimentally reach the same utility while saving 50% in terms of epsilon for ε ≤ 0.5 on the Abalone regression dataset (dataset size ≈ 4K), saving 30% in terms of epsilon for ε ≤ 0.08 for the Adult classification dataset (dataset size ≈ 50K), and saving 30% in terms of epsilon for ε ≤ 0.03 for the Spambase classification dataset (dataset size ≈ 5K). Moreover, we show that for situations where a GBDT is learning a stream of data that originates from different subpopulations (non-IID), S-BDT improves the saving of epsilon even further."
BibTeX anzeigen
@inproceedings{10.1145/3658644.3690301,
title = {S-BDT: Distributed Differentially Private Boosted Decision Trees},
author = {Thorsten Peinemann and Moritz Kirschte and Joshua Stock and Carlos Cotrini and Esfandiar Mohammadi},
year = {2024},
url = {https://doi.org/10.1145/3658644.3690301},
}DPM: Clustering Sensitive Data through Separation
Abstract
"Clustering is an important tool for data exploration where the goal is to subdivide a data set into disjoint clusters that fit well into the underlying data structure. When dealing with sensitive data, privacy-preserving algorithms aim to approximate the non-private baseline while minimising the leakage of sensitive information. State-of-the-art privacy-preserving clustering algorithms tend to output clusters that are good in terms of the standard metrics, inertia, silhouette score, and clustering accuracy, however, the clustering result strongly deviates from the non-private KMeans baseline. In this work, we present a privacy-preserving clustering algorithm called DPM that recursively separates a data set into clusters based on a geometrical clustering approach. In addition, DPM estimates most of the data-dependent hyper-parameters in a privacy-preserving way. We prove that DPM preserves Differential Privacy and analyse the utility guarantees of DPM. Finally, we conduct an extensive empirical evaluation for synthetic and real-life data sets. We show that DPM achieves state-of-the-art utility on the standard clustering metrics and yields a clustering result much closer to that of the popular non-private KMeans algorithm without requiring the number of classes."
BibTeX anzeigen
@inproceedings{10.1145/3658644.3690271,
title = {DPM: Clustering Sensitive Data through Separation},
author = {Johannes Liebenow and Yara Schütt and Tanya Braun and Marcel Gehrke and Florian Thaeter and Esfandiar Mohammadi},
year = {2024},
url = {https://doi.org/10.1145/3658644.3690271},
}Towards Privacy-Preserving Relational Data Synthesis via Probabilistic Relational Models
Abstract
"Probabilistic relational models provide a well-established formalism to combine first-order logic and probabilistic models, thereby allowing to represent relationships between objects in a relational domain. At the same time, the field of artificial intelligence requires increasingly large amounts of relational training data for various machine learning tasks. Collecting real-world data, however, is often challenging due to privacy concerns, data protection regulations, high costs, and so on. To mitigate these challenges, the generation of synthetic data is a promising approach. In this paper, we solve the problem of generating synthetic relational data via probabilistic relational models. In particular, we propose a fully-fledged pipeline to go from relational database to probabilistic relational model, which can then be used to sample new synthetic relational data points from its underlying probability distribution. As part of our proposed pipeline, we introduce a learning algorithm to construct a probabilistic relational model from a given relational database."
BibTeX anzeigen
@inproceedings{10.1007/978-3-031-70893-0_13,
title = {Towards Privacy-Preserving Relational Data Synthesis via Probabilistic Relational Models},
author = {Malte Luttermann and Ralf Möller and Mattis Hartwig},
year = {2024},
}Distributed DP-Helmet: Scalable Differentially Private Non-interactive Averaging of Single Layers
BibTeX anzeigen
@misc{kirschte2024distributeddphelmetscalabledifferentially,
title = {Distributed DP-Helmet: Scalable Differentially Private Non-interactive Averaging of Single Layers},
author = {Moritz Kirschte and Sebastian Meiser and Saman Ardalan and Esfandiar Mohammadi},
year = {2024},
url = {https://arxiv.org/abs/2211.02003},
}Revealing Unintentional Information Leakage in Low-Dimensional Facial Portrait Representations
BibTeX anzeigen
@inproceedings{Anderson2024,
title = {Revealing Unintentional Information Leakage in Low-Dimensional Facial Portrait Representations},
author = {Kathleen Anderson and Thomas Martinetz},
year = {2024},
}Colour Passing Revisited: Lifted Model Construction with Commutative Factors
BibTeX anzeigen
@inproceedings{Luttermann2024c,
title = {Colour Passing Revisited: Lifted Model Construction with Commutative Factors},
author = {Malte Luttermann and Tanya Braun and Ralf Möller and Marcel Gehrke},
year = {2024},
url = {https://ojs.aaai.org/index.php/AAAI/article/view/30034},
}Lifted Causal Inference in Relational Domains
BibTeX anzeigen
@inproceedings{Luttermann2024b,
title = {Lifted Causal Inference in Relational Domains},
author = {Malte Luttermann and Mattis Hartwig and Tanya Braun and Ralf Möller and Marcel Gehrke},
year = {2024},
url = {https://proceedings.mlr.press/v236/luttermann24a.html},
}Efficient Detection of Exchangeable Factors in Factor Graphs
BibTeX anzeigen
@inproceedings{Luttermann2024a,
title = {Efficient Detection of Exchangeable Factors in Factor Graphs},
author = {Malte Luttermann and Johann Machemer and Marcel Gehrke},
year = {2024},
url = {https://journals.flvc.org/FLAIRS/article/view/135518},
}Lifting in Support of Privacy-Preserving Probabilistic Inference
BibTeX anzeigen
@article{Gehrke2024,
title = {Lifting in Support of Privacy-Preserving Probabilistic Inference},
author = {Marcel Gehrke and Johannes Liebenow and Esfandiar Mohammadi and Tanya Braun},
journal = {German Journal of Artificial Intelligence},
year = {2024},
url = {https://link.springer.com/article/10.1007/s13218-024-00851-y},
}Towards Privacy and Utility in Tourette Tic Detection Through Pretraining Based on Publicly Available Video Data of Healthy Subjects
BibTeX anzeigen
@inproceedings{10095309,
title = {Towards Privacy and Utility in Tourette Tic Detection Through Pretraining Based on Publicly Available Video Data of Healthy Subjects},
author = {Nele Sophie Brügge and Esfandiar Mohammadi and Alexander Münchau and Tobias Bäumer and Christian Frings and Christian Beste and Veit Roessner and Heinz Handels},
year = {2023},
}Lifting Factor Graphs with Some Unknown Factors
BibTeX anzeigen
@inproceedings{Luttermann2023,
title = {Lifting Factor Graphs with Some Unknown Factors},
author = {Malte Luttermann and Ralf Möller and Marcel Gehrke},
year = {2023},
url = {https://link.springer.com/chapter/10.1007/978-3-031-45608-4_25},
}
