arXiv subjects

Michael Heigl

Publications and source records attributed to Michael Heigl.

14 recordsLinked to original sources

arXiv2026

Towards Context-Aware Image Anonymization with Multi-Agent Reasoning

Street-level imagery contains personally identifiable information (PII), some of which is context-dependent. Existing anonymization methods either over-process images or miss subtle identifiers, while API-based solutions compromise data sovereignty. We present an agentic framework CAIAMAR (\underline{C}ontext-\underline{A}ware \underline{I}mage \underline{A}nonymization with \underline{M}ulti-\underline{A}gent \underline{R}easoning) for context-aware PII segmentation with diffusion-based anonymization, combining pre-defined processing for high-confidence cases with multi-agent reasoning for indirect identifiers. Three specialized agents coordinate via round-robin speaker selection in a Plan-Do-Check-Act (PDCA) cycle, enabling large vision-language models to classify PII based on spatial context (private vs. public property) rather than rigid category rules. The agents implement spatially-filtered coarse-to-fine detection where a scout-and-zoom strategy identifies candidates, open-vocabulary segmentation processes localized crops, and $IoU$-based deduplication ($30\%$ threshold) prevents redundant processing. Modal-specific diffusion guidance with appearance decorrelation substantially reduces re-identification (Re-ID) risks. On CUHK03-NP, our method reduces person Re-ID risk by $73\%$ ($R1$: $16.9\%$ vs. $62.4\%$ baseline). For image quality preservation on CityScapes, we achieve KID: $0.001$, and FID: $9.1$, significantly outperforming existing anonymization. The agentic workflow detects non-direct PII instances across object categories, and downstream semantic segmentation is preserved. Operating entirely on-premise with open-source models, the framework generates human-interpretable audit trails supporting EU's GDPR transparency requirements while flagging failed cases for human review.

cs.CV↗

arXiv2026

PriMod4AI: Lifecycle-Aware Privacy Threat Modeling for AI Systems using LLM

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.

cs.CR↗

arXiv2025

Following the Clues: Experiments on Person Re-ID using Cross-Modal Intelligence

The collection and release of street-level recordings as Open Data play a vital role in advancing autonomous driving systems and AI research. However, these datasets pose significant privacy risks, particularly for pedestrians, due to the presence of Personally Identifiable Information (PII) that extends beyond biometric traits such as faces. In this paper, we present cRID, a novel cross-modal framework combining Large Vision-Language Models, Graph Attention Networks, and representation learning to detect textual describable clues of PII and enhance person re-identification (Re-ID). Our approach focuses on identifying and leveraging interpretable features, enabling the detection of semantically meaningful PII beyond low-level appearance cues. We conduct a systematic evaluation of PII presence in person image datasets. Our experiments show improved performance in practical cross-dataset Re-ID scenarios, notably from Market-1501 to CUHK03-np (detected), highlighting the framework's practical utility. Code is available at https://github.com/RAufschlaeger/cRID.

cs.CV↗

arXiv2024

ClustEm4Ano: Clustering Text Embeddings of Nominal Textual Attributes for Microdata Anonymization

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.

cs.CL↗

arXiv2023

Nonreciprocal Spin Waves in Nanoscale Domain Walls Detected by Scanning X-ray Microscopy in Perpendicular Magnetic Anisotropic Fe/Gd Multilayers

Spin wave nonreciprocity in domain walls (DWs) allows for unidirectional signal processing in reconfigurable magnonic circuits. Using scanning transmission x-ray microscopy (STXM), we examined coherently-excited magnons propagating in Bloch-like DWs in amorphous Fe/Gd multilayers with perpendicular magnetic anisotropy (PMA). Near 1 GHz we detected magnons with short wavelengths down to $\lambda = 281$ nm in DWs whose minimum width amounted to $\delta_{\rm DW} = 52$ nm. Consistent with micromagnetic simulations, the STXM data reveal their nonreciprocal magnon band structures. We identified Bloch points which disrupted the phase evolution of magnons and induced different $\lambda$ adjacent to the topological defects. Our observations provide direct evidence of nonreciprocal spin waves within Bloch-like DWs, serving as programmable waveguides in magnonic devices with directed information flow.

cond-mat.mes-hall↗

arXiv2022

Generation and annihilation of skyrmions and antiskyrmions in magnetic heterostructures

We demonstrate the controlled generation and annihilation of (anti)skyrmions with tunable chirality in magnetic heterostructures by means of micromagnetic simulations. By making use of magnetic (anti)vortices in patterned ferromagnetic layer, we stabilize full lattices of (anti)skyrmions in an underlying skyrmionic thin film in a reproducible manner. The stability of the (anti)skyrmion depends on the polarization of the (anti)vortex, whereas their chirality is given by those of the (anti)vortices. Furthermore, we demonstrate that the core coupling between the (anti)vortices and (anti)skyrmions allows to annihilate the spin-objects in a controlled fashion by applying short pulses of in-plane external magnetic fields, representing a new key paradigm in skyrmionic devices.

cond-mat.mes-hall↗

arXiv2021

Dipolar-stabilized first and second-order antiskyrmions in ferrimagnetic multilayers

Skyrmions and antiskyrmions are topologically protected spin structures with opposite topological charge. Particularly in coexisting phases, these two types of magnetic quasi-particles may show fascinating physics and potential for spintronic devices. While skyrmions are observed in a wide range of materials, until now antiskyrmions were exclusive to materials with D2d symmetry. In this work, we show first and second-order antiskyrmions stabilized by magnetic dipole-dipole interaction in Fe/Gd-based multilayers. We modify the magnetic properties of the multilayers by Ir insertion layers. Using Lorentz transmission electron microscopy imaging, we observe coexisting antiskyrmions, Bloch skyrmions, and type-2 bubbles and determine the range of material properties and magnetic fields where the different spin objects form and dissipate. We perform micromagnetic simulations to obtain more insight into the studied system and conclude that the reduction of saturation magnetization and uniaxial anisotropy leads to the existence of this zoo of different spin objects and that they are primarily stabilized by dipolar interaction.

cond-mat.mtrl-sci↗

arXiv2020

Nonreciprocal Dzyaloshinskii-Moriya magnetoacoustic waves

We study the interaction of surface acoustic waves with spin waves in ultra-thin CoFeB/Pt bilayers. Due to the interfacial Dzyaloshinskii-Moriya interaction (DMI), the spin wave dispersion is non-degenerate for oppositely propagating spin waves in CoFeB/Pt. In combination with the additional nonreciprocity of the magnetoacoustic coupling itself, highly nonreciprocal acoustic wave transmission through the magnetic film is observed. We systematically characterize the magnetoacoustic wave propagation in a thickness series of CoFeB($d$)/Pt samples as a function of magnetic field magnitude and direction, and at frequencies up to 7 GHz. We quantitatively model our results to extract the strength of the DMI and magnetoacoustic driving fields.

cond-mat.mes-hall↗

arXiv2020

Ultrafast high-harmonic nanoscopy of magnetization dynamics

Light-induced magnetization changes, such as all-optical switching, skyrmion nucleation, and intersite spin transfer, unfold on temporal and spatial scales down to femtoseconds and nanometers, respectively. Pump-probe spectroscopy and diffraction studies indicate that spatio-temporal dynamics may drastically affect the non-equilibrium magnetic evolution. Yet, direct real-space magnetic imaging on the relevant timescale has remained challenging. Here, we demonstrate ultrafast high-harmonic nanoscopy employing circularly polarized high-harmonic radiation for real-space imaging of femtosecond magnetization dynamics. We observe the reversible and irreversible evolution of nanoscale spin textures following femtosecond laser excitation. Specifically, we map quenched magnetic domains and localized spin structures in Co/Pd multilayers with a sub-wavelength spatial resolution down to 16 nm, and strobosocopically trace the local magnetization dynamics with 40 fs temporal resolution. Our approach enables the highest spatio-temporal resolution of magneto-optical imaging to date. Facilitating ultrafast imaging with an extreme sensitivity to various microscopic degrees of freedom expressed in chiral and linear dichroism, we envisage a wide range of applications spanning magnetism, phase transitions, and carrier dynamics.

physics.optics↗

arXiv2020

Microscopic origin of magnetization reversal in exchange-coupled ferro-/ferrimagnetic bilayers

In this study, the magnetic reversal process of exchange-coupled bilayer systems, consisting of a ferrimagnetic TbFeCo alloy layer and a ferromagnetic [Co/Ni/Pt]N multilayer, was investigated. In particular, minor loop studies, probing solely the reversal characteristics of the softer ferromagnetic layer, reveal two distinct reversal mechanisms, which depend strongly on the thickness of the ferromagnetic layer. For thick layers, irreversible switching of the macroscopic minor loop is observed. The underlying microscopic origin of this reversal process was studied in detail by high-resolution magnetic force microscopy, showing that the reversal is triggered by in-plane domain walls propagating through the ferromagnetic layer. In contrast, thin ferromagnetic layers show a hysteresis-free reversal, which is nucleation-dominated due to grain-to-grain variations in magnetic anisotropy of the Co/Ni/Pt multilayer and an inhomogeneous exchange coupling with the magnetically hard TbFeCo layer, as confirmed by micromagnetic simulations.

cond-mat.mtrl-sci↗

arXiv2020

Hysteresis-free magnetization reversal of exchange-coupled bilayers with finite magnetic anisotropy

Exchange-coupled structures consisting of ferromagnetic and ferrimagnetic layers become technologically more and more important. We show experimentally the occurrence of completely reversible, hysteresis-free minor loops of [Co(0.2 nm)/Ni(0.4 nm)/Pt(0.6 nm)]$_N$ multilayers exchange-coupled to a 20 nm thick ferrimagnetic Tb$_{28}$Co$_{14}$Fe$_{58}$ layer, acting as hard magnetic pinning layer. Furthermore, we present detailed theoretical investigations by means of micromagnetic simulations and most important a purely analytical derivation for the condition of the occurrence of full reversibility in magnetization reversal. Hysteresis-free loops always occur if a domain wall is formed during the reversal of the ferromagnetic layer and generates an intrinsic hard-axis bias field that overcomes the magnetic anisotropy field of the ferromagnetic layer. The derived condition further reveals that the magnetic anisotropy and the bulk exchange of both layers, as well as the exchange coupling strength and the thickness of the ferromagnetic layer play an important role for its reversibility.

cond-mat.mtrl-sci↗

arXiv2020

Magnetic properties of Co/Ni-based multilayers with Pd and Pt insertion layers

In this study, the influence of Pd and Pt insertion layers in Co/Ni multilayers (MLs) on their magnetic properties, e.g. magnetic anisotropies, saturation magnetization, coercivity, magnetic domain size, and Curie temperature, is investigated. We compare three series of [Co/Ni/X]N ML systems (X = Pd, Pt, no insertion layer), varying the individual Co layer thickness as well as the repetition number N. All three systems behave very similarly for the different Co layer thicknesses. For all systems, a maximum effective magnetic anisotropy was achieved for MLs with a Co layer thickness between 0.15 nm and 0.25 nm. The transition from an out-of-plane to an in-plane system occurs at about 0.4 nm of Co. While [Co(0.2 nm)/Ni(0.4 nm)]N MLs change their preferred easy magnetization axis from out-of-plane to in-plane after 6 bilayer repetitions, insertion of Pd and Pt results in an extension of this transition beyond 15 repetitions. The maximum effective magnetic anisotropy was more than doubled from 105 kJ/m3 for [Co/Ni]3 to 275 and 186 kJ/m3 for Pt and Pd, respectively. Furthermore, the insertion layers strongly reduce the initial saturation magnetization of 1100 kA/m of Co/Ni MLs and lower the Curie temperature from 720 to around 500 K

cond-mat.mtrl-sci↗

arXiv2014

Dynamic Acoustic Control of Individual Optically Active Quantum Dot-like Emission Centers in Heterostructure Nanowires

We probe and control the optical properties of emission centers forming in radial het- erostructure GaAs-Al0.3Ga0.7As nanowires and show that these emitters, located in Al0.3Ga0.7As layers, can exhibit quantum-dot like characteristics. We employ a radio frequency surface acoustic wave to dynamically control their emission energy and occupancy state on a nanosec- ond timescale. In the spectral oscillations we identify unambiguous signatures arising from both the mechanical and electrical component of the surface acoustic wave. In addition, differ- ent emission lines of a single quantum dot exhibit pronounced anti-correlated intensity oscilla- tions during the acoustic cycle. These arise from a dynamically triggered carrier extraction out of the quantum dot to a continuum in the radial heterostructure. Using finite element modeling and Wentzel-Kramers-Brillouin theory we identify quantum tunneling as the underlying mech- anism. These simulation results quantitatively reproduce the observed switching and show that in our systems these quantum dots are spatially separated from the continuum by > 10.5 nm.

cond-mat.mes-hall↗

arXiv2014

Radio frequency occupancy state control of a single nanowire quantum dot

The excitonic occupancy state of a single, nanowire-based, heterostructure quantum dot is dynamically programmed by a surface acoustic wave. The quantum dot is formed by an interface or thickness fluctuation of a GaAs QW embedded in a AlGaAs shell of a GaAs-AlGaAs core-shell nanowire. As we tune the time at which carriers are photogenerated during the acoustic cycle, we find pronounced intensity oscillations of neutral and negatively charged excitons. At high acoustic power levels these oscillations become anticorrelated which enables direct acoustic programming of the dot's charge configuration, emission intensity and emission wavelength. Numerical simulations confirm that the observed modulations arise from acoustically controlled modulations of the electron and electron-hole-pair concentrations at the position of the quantum dot.

cond-mat.mes-hall↗