arXiv subjects

Lukas Schmidt

Publications and source records attributed to Lukas Schmidt.

10 recordsLinked to original sources

arXiv2026

What Does It Take to Detect an AI Agent? Minimal Feature Sets for Behavioral Detection under Browser Automation

Bot detectors deployed at scale treat traffic as binary: human or bot. This assumption breaks when AI agents browse the web through browser automation, a traffic class that is neither and that binary classifiers structurally cannot represent. We present a three-class detection framework distinguishing humans, bots, and AI agents, and show that the binary-vs-agent confusion is architectural: a binary human-vs-bot detector misroutes agent sessions because its label space lacks an agent class. On our controlled benchmark, an MLP binary classifier misclassifies 39.1% of real AI agents as human and a SAINT binary transformer misclassifies 34.5%; adding an explicit agent class yields per-class agent F1 = 1.000 in all 30 runs (3 model families $\times$ 10 seeds). To measure evasion resistance, we construct a five-level evasion ladder spanning passive observation, GAN-generated trajectories, and replay of real human cursor data ($n = 2299$ evasion sessions). Across 10 seeds and 3 model families we observe zero agent misses in 22990 per-seed predictions. The discriminative signal is a browser-automation artifact, not evidence of agent reasoning: Playwright does not emit the raw pointer-move and wheel-delta streams a physical input device produces, and this absence signature survives trajectory manipulation. Exhaustive search over all feature subsets of size 1-5 (9401 GBMs) shows that two behavioral features (mouse_event_rate, teleport_click_ratio) give 100% observed agent recall at every evasion level with agent precision 0.994; five features lift macro-F1 to 0.991. The signal is redundantly encoded: removing teleport_click_ratio leaves agent detection at 100%. The single-feature regime is degenerate, flagging every agent only by collapsing the classifier to always predict "agent". Two features robustly isolate agents; five separate all three traffic classes at macro-F1 $\geq 0.99$.

cs.AI↗

arXiv2026

Measuring Healthcare Data Leaks and Security Flaws at Internet Scale

Systems that process medical data should be meticulously secured. Yet, network services in healthcare environments often fail to implement basic security measures. For example, previous studies showed that network segmentation flaws led to DICOM systems leaking millions of patient records. In addition to DICOM, healthcare facilities rely heavily on the HL7 and FHIR protocols to transmit data. For nine months, we operated a low-interaction honeypot for medical protocols. We found it was regularly scanned for DICOM but never for HL7 or FHIR, indicating that despite their widespread use and importance for patient data security, the security of these services remains underexplored. In this paper, we present the first large-scale study on HL7 and FHIR services and expand previous work on DICOM. Our large-scale Internet scans, covering the three major healthcare protocols across IPv4 and IPv6 address spaces, identify healthcare systems and uncover data leaks due to authentication flaws. Additionally, we scanned for deficiencies in TLS configurations of these services and known insecure healthcare software. In total, we found 2,841 healthcare services with authentication flaws. 94.4% of all exposed systems do not support transport encryption, and 1,373 systems have known software vulnerabilities, including those with potential for system takeover and CVSS scores up to 9.8. Overall, our study reveals an alarming state of cybersecurity in healthcare deployments, for which we discuss potential reasons and countermeasures. Finally, we report on the coordinated disclosure campaign we initiated to improve the security of patient data.

cs.CR↗

arXiv2026

Integrating Causal Machine Learning into Clinical Decision Support Systems: Insights from Literature and Practice

Current clinical decision support systems (CDSSs) typically base their predictions on correlation, not causation. In recent years, causal machine learning (ML) has emerged as a promising way to improve decision-making with CDSSs by offering interpretable, treatment-specific reasoning. However, existing research often emphasizes model development rather than designing clinician-facing interfaces. To address this gap, we investigated how CDSSs based on causal ML should be designed to effectively support collaborative clinical decision-making. Using a design science research methodology, we conducted a structured literature review and interviewed experienced physicians. From these, we derived eight empirically grounded design requirements, developed seven design principles, and proposed nine practical design features. Our results establish guidance for designing CDSSs that deliver causal insights, integrate seamlessly into clinical workflows, and support trust, usability, and human-AI collaboration. We also reveal tensions around automation, responsibility, and regulation, highlighting the need for an adaptive certification process for ML-based medical products.

cs.HC↗

arXiv2025

Exact nematic and mixed magnetic phases driven by competing orders on the pyrochlore lattice

Pyrochlore magnets are a paradigmatic example of three-dimensional frustrated systems and provide an excellent platform for studying a variety of exotic many-body phenomena, including spin liquids, nematic phases, fragmentation, and order by disorder. In recent years, increasing attention has been devoted to bilinear spin models on this lattice, where multiple magnetic phases can be degenerate in energy, often stabilizing unconventional magnetic states. In this work, we focus on one such model, parametrized by the interaction coupling $J_{z\pm}$, which defines a line in parameter space corresponding to the phase boundary between three distinct magnetic phases. Using a combination of analytical and numerical methods, we show that this model exhibits an order-by-disorder mechanism at low temperatures, giving rise to a \emph{mixed} magnetic phase. This represents the first realization of a $\mathbf{q}=0$ long-range-ordered phase in a pyrochlore magnet characterized by two distinct order parameters, which we denote as the $A_2 \oplus \psi_2$ phase. Furthermore, at $J_{z\pm} = 1/\sqrt{2}$, the model acquires a subextensive number of discrete symmetries, which preclude the stabilization of conventional long-range order and instead lead to the emergence of a novel nematic phase. We characterize this nematic phase, describe how its ground-state configurations are constructed, and analyze its stability at higher temperatures and under small deviations from $J_{z\pm} = 1/\sqrt{2}$.

cond-mat.str-el↗

arXiv2024

Optimization of 3-D flight trajectory of variable trim kites for airborne wind energy production

Skysails Power GmbH is the leading manufacturer of light and efficient power kites that harness the wind's untapped supplies at high altitudes, aiming at profoundly altering wind energy's impact in achieving the global energy transition. Novel, variable trim kites have been developed that allow to modulate the aerodynamic coefficients of the airborne system, significantly improving the overall system efficiency. The flight control of variable trim kites is much more complex than that of previous kite generations and its mastering is a challenge and one of the keys to a successful operation. Numerical optimization is applied to find a set of flight trajectories in order to maximize the energy production while satisfying several constraints on the system operating in a wide range of conditions. This industry abstract provides a general introduction of the trajectory optimization problem with variable trim kites. We also briefly introduce the state-of-the-art optimization setup. This is followed by demonstration of high-quality example results of the optimization. Finally, we discuss the results and their applications.

math.OC↗

arXiv2022

Don't Get Me Wrong: How to Apply Deep Visual Interpretations to Time Series

The correct interpretation of convolutional models is a hard problem for time series data. While saliency methods promise visual validation of predictions for image and language processing, they fall short when applied to time series. These tend to be less intuitive and represent highly diverse data, such as the tool-use time series dataset. Furthermore, saliency methods often generate varied, conflicting explanations, complicating the reliability of these methods. Consequently, a rigorous objective assessment is necessary to establish trust in them. This paper investigates saliency methods on time series data to formulate recommendations for interpreting convolutional models and implements them on the tool-use time series problem. To achieve this, we first employ nine gradient-, propagation-, or perturbation-based post-hoc saliency methods across six varied and complex real-world datasets. Next, we evaluate these methods using five independent metrics to generate recommendations. Subsequently, we implement a case study focusing on tool-use time series using convolutional classification models. Our results validate our recommendations that indicate that none of the saliency methods consistently outperforms others on all metrics, while some are sometimes ahead. Our insights and step-by-step guidelines allow experts to choose suitable saliency methods for a given model and dataset.

cs.CV↗

arXiv2022

Which model features matter? An experimental approach to evaluate power market modeling choices

A novel experimental approach of inter- and intramodel comparisons is conducted with five power market models to give recommendations for modelers working on decarbonization pathways of Europe until 2050. The experiments investigate the impact of model type (optimization vs. simulation), planning horizon (intertemporal vs. myopic), temporal resolution (8760 vs. 384 hours), and spatial resolution (28 countries vs. 12 mega-regions). The model type fundamentally determines the evolution of capacity expansion. Planning horizon (assumed foresight of firms) plays a minor role for scenarios with high carbon prices. For low carbon prices in turn, results from myopic models deviate considerably from those of intertemporal models. Lower temporal and spatial resolutions foster wind power via storage and via neglected transmission boundaries, respectively. Using simulation instead of optimization frameworks, a shorter planning horizon of firms, or lower temporal and spatial resolutions might be necessary to reduce the computational complexity. This paper delivers recommendations on how to limit the discrepancies in such cases.

physics.soc-ph↗

arXiv2017

Inhomogeneous preferential concentration of inertial particles in turbulent channel flow

Turbophoresis leading to preferential concentration of inertial particles in regions of low turbulent diffusivity is a unique feature of inhomogeneous turbulent flows, such as free shear flows or wall-bounded flows. In this work, the theory for clustering of weakly inertial particles in homogeneous turbulence is extended to the inhomogeneous case of a turbulent channel flow. The inhomogeneity contributes to the cluster formation in addition to clustering in homogeneous turbulence. A space-dependent rate for the creation of inhomogeneous particle concentration is derived in terms of local statistics of turbulence. We provide the formula for the pair-correlation function of concentration that factorizes in product of time and space-dependent average concentrations and time-independent factor of clustering that obeys a power-law in the distance between the points. This power-law characterizes inhomogeneous multifractality of the particle distribution. A unique demonstration and quantification of the combined effects of turbophoresis and fractal clustering in a direct numerical simulation of particle motion in a turbulent channel flow is performed according to the presented theory. The strongest contribution to clustering coming from the inhomogeneity of the flow occurs in the transitional region between viscous sublayer and the buffer layer. Further the ratio of homogeneous and inhomogeneous term depends on the wall distance. The inhomogeneous terms may significantly increase the preferential concentration of inertial particles, thus the overall degree of clustering in inhomogeneous turbulence is potentially stronger compared to particles with the same inertia in purely homogeneous turbulence.

physics.flu-dyn↗

arXiv2016

Inertial particles distribute in turbulence as Poissonian points with random intensity inducing clustering and supervoiding

This work considers the distribution of inertial particles in turbulence using the point-particle approximation. We demonstrate that the random point process formed by the positions of particles in space is a Poisson point process with log-normal random intensity ("log Gaussian Cox process" or LGCP). The probability of having a finite number of particles in a small volume is given in terms of the characteristic function of a log-normal distribution. Corrections due to discreteness of the number of particles to the previously derived statistics of particle concentration in the continuum limit are provided. These are relevant for dealing with experimental or numerical data. The probability of having regions without particles, i.e. voids, is larger for inertial particles than for tracer particles where voids are distributed according to Poisson processes. Further, the probability of having large voids decays only log-normally with size. This shows that particles cluster, leaving voids behind. At scales where there is no clustering there can still be an increase of the void probability so that turbulent voiding is stronger than clustering. The demonstrated double stochasticity of the distribution originates in the two-step formation of fluctuations. First, turbulence brings the particles randomly close together which happens with Poisson-type probability. Then, turbulence compresses the particles' volume in the observation volume. We confirm the theory of the statistics of the number of particles in small volumes by numerical observations of inertial particle motion in a chaotic ABC flow. The improved understanding of clustering processes can be applied to predict the long-time survival probability of reacting particles. Our work implies that the particle distribution in weakly compressible flow with finite time correlations is a LGCP, independently of the details of the flow statistics.

physics.flu-dyn↗

arXiv2016

On phoretic clustering of particles in turbulence

We demonstrate that diffusiophoretic, thermophoretic and chemotactic phenomena in turbulence lead to clustering of particles on multi-fractal sets that can be described using one single framework, valid when the particle size is much smaller than the smallest length scale of turbulence $l_0$. To quantify the clustering, we derive positive pair correlations and fractal dimensions that hold for scales smaller than $l_0$. Statistics of the number of particles in a small volume are non-Poissonian manifesting deviations from the case of uncorrelated particles. For scales larger than $l_0$ we predict a stretched exponential decay to 1 of the pair correlation function. For the case of inhomogeneous turbulence we find that the fractal dimension depends on the inhomogeneous direction. By performing experiments of clustering of diffusiophoretic particles induced by salinity gradients in a turbulent gravity current we demonstrate clustering in conformity to the theory. The particle size in the experiment is comparable to $l_0$, outside the strict validity region of the theory, suggesting that the theoretical predictions transfer to this practically relevant regime. This clustering mechanism can provide the key to the understanding of a multitude of processes such as formation of marine snow in the ocean and population dynamics of chemotactic bacteria.

physics.flu-dyn↗