arXiv subjects

Christian J. Cyron

Publications and source records attributed to Christian J. Cyron.

At least 19 recordsLinked to original sources

arXiv2026

Molecular Déjà Vu: Digit-Level Retrieval of Published Values in Frontier Language Models

Large language models (LLMs) are increasingly evaluated on molecular property benchmarks, but accuracy cannot distinguish a model that predicts a property from one that retrieves a published number. We audit 22 frontier models on 12 regression benchmarks for verbatim retrieval and find that it is widespread but relatively benchmark-specific: on five datasets more than $50\%$ of the LLMs show verbatim retrieval, while on the remaining datasets it appears only in isolated cells. We run our experiments at two reasoning levels and find that reasoning changes retrieval. The same experiments, on the same molecules and with the same prompt, are flagged $89\%$ more often at the higher reasoning level than at the lowest one. Finally, we test a way to interrupt retrieval in our most contaminated cases, and find that the strongest models in some cases still recognise a combination of transformed SMILES strings and original labels. Furthermore, suppressing retrieval moves the prediction errors of the different models closer together in relative terms, while their differing use of verbatim retrieval spreads them apart. This indicates that the general predictive capability of an LLM is not determined solely by the amount of memorised values. This work provides an overview of the amount and depth of verbatim retrieval in molecular regression benchmarks using LLMs.

cs.AI↗

arXiv2026

In-Context Molecular Property Prediction with LLMs: A Blinding Study on Memorization and Knowledge Conflicts

The capabilities of large language models (LLMs) have expanded beyond natural language processing to scientific prediction tasks, including molecular property prediction. However, their effectiveness in in-context learning remains ambiguous, particularly given the potential for training data contamination in widely used benchmarks. This paper investigates whether LLMs perform genuine in-context regression on molecular properties or instead rely on verbatim retrieval of memorized target values. Furthermore, we analyze the interplay between pre-trained knowledge and in-context information through a series of progressively blinded experiments. We evaluate nine LLM variants across three families (GPT-4.1, GPT-5, Gemini 2.5) on three MoleculeNet datasets (Delaney solubility, Lipophilicity, QM7 atomization energy) using a systematic blinding approach that iteratively reduces available information, complemented by 0-, 60-, and 1000-shot in-context sample sizes as an additional control for information access. To validate the memorization analysis and the blinding experiments, we add a positive and a negative control for the memorization experiments and structural reference baselines for the multi-shot experiments as well as bootstrap confidence intervals for all results. We find no evidence of verbatim retrieval on the legacy benchmarks and show that blinding exposes conflicts between pre-trained knowledge and in-context information. This work provides a principled framework for evaluating molecular property prediction under controlled information access.

cs.LG↗

arXiv2026

Inelastic Constitutive Kolmogorov-Arnold Networks: A generalized framework for automated discovery of interpretable inelastic material models

A key problem of solid mechanics is the identification of the constitutive law of a material, that is, the relation between strain history and stress. Machine learning has lead to considerable advances in this field lately. Here we introduce inelastic Constitutive Kolmogorov-Arnold Networks (iCKANs). This novel artificial neural network architecture can discover in an automated manner symbolic constitutive laws describing both the elastic and inelastic behavior of materials. That is, it can translate data from material testing into corresponding elastic and inelastic potential functions in closed mathematical form. We demonstrate the advantages of iCKANs using both synthetic data and experimental data of the viscoelastic polymer materials VHB 4910 and VHB 4905. The results demonstrate that iCKANs accurately capture complex viscoelastic behavior while preserving physical interpretability. It is a particular strength of iCKANs that they can process not only mechanical data but also arbitrary additional information available about a material (e.g., about temperature-dependent behavior). This makes iCKANs a powerful tool to discover in the future also how specific processing or service conditions affect the properties of materials.

cond-mat.mtrl-sci↗

arXiv2025

Electromechanical computational model of the human stomach

The stomach plays a central role in digestion through coordinated muscle contractions, known as gastric peristalsis, driven by slow-wave electrophysiology. Understanding this process is critical for treating motility disorders such as gastroparesis, dyspepsia, and gastroesophageal reflux disease. Computer simulations can be a valuable tool to deepen our understanding of these disorders and help to develop new therapies. However, existing approaches often neglect spatial heterogeneity, fail to capture large anisotropic deformations, or rely on computationally expensive three-dimensional formulations. We present here a computational framework of human gastric electromechanics, that combines a nonlinear, rotation-free shell formulation with a constrained mixture material model. The formulation incorporates active-strain, constituent-specific prestress, and spatially non-uniform parameter fields. Numerical examples demonstrate that the framework can reproduce characteristic features of gastric motility, including slow-wave entrainment, conduction velocity gradients, and large peristaltic contractions with physiologically realistic amplitudes. The proposed framework enables robust electromechanical simulations of the whole stomach at the organ scale. It thus provides a promising basis for future in silico studies of both physiological function and pathological motility disorders.

cs.CE↗

arXiv2025

Machine-learned domain partitioning for computationally efficient coupling of continuum and particle simulations of membrane fabrication

All simulation approaches eventually face limits in computational scalability when applied to large spatiotemporal domains. This challenge becomes especially apparent in molecular-level particle simulations, where high spatial and temporal resolution leads to rapidly increasing computational demands. To overcome these limitations, hybrid methods that combine simulations with different levels of resolution offer a promising solution. In this context, we present a machine learning-based decision model that dynamically selects between simulation methods at runtime. The model is built around a Multilayer perceptron (MLP) that predicts the expected discrepancy between particle and continuum simulation results, enabling the localized use of high-fidelity particle simulations only where they are expected to add value. This concurrent approach is applied to the simulation of membrane fabrication processes, where a particle simulation is coupled with a continuum model. This article describes the architecture of the decision model and its integration into the simulation workflow, enabling efficient, scalable, and adaptive multiscale simulations.

physics.comp-ph↗

arXiv2025

Unifying local and nonlocal corrosion frameworks: A convergent nonlocal extension of the KKS phase-field model

We introduce a nonlocal extension of the Kim-Kim-Suzuki (KKS) phase-field corrosion model aimed at bridging local and nonlocal corrosion modeling approaches, such as phase-field and peridynamic frameworks. In this formulation, classical gradient operators are replaced with integral operators defined over a finite interaction horizon, naturally embedding an intrinsic length scale that aligns with nonlocal theories like peridynamics. Under precise assumptions on function spaces and kernel functions, we define a nonlocal free energy that integrates a standard bulk free energy density with a nonlocal interaction term. Through differentiation in an appropriate Hilbert space, we derive evolution equations, yielding a nonlocal Allen--Cahn equation for the phase-field and a nonlocal Cahn--Hilliard-type equation for the concentration. The latter is expressed as a gradient flow in a metric induced by the inverse of the nonlocal operator, mirroring the classical $H^{-1}$ metric for conserved dynamics. We establish the well-posedness of these equations using Galerkin approximations, uniform energy estimates, and compactness arguments. Furthermore, we prove the convergence of the nonlocal model to its local KKS counterpart as the interaction horizon approaches zero, effectively unifying local and nonlocal perspectives. Numerical experiments, implemented via finite difference spatial discretization and explicit time-stepping, demonstrate the effects of nonlocality and confirm the theoretical convergence, reinforcing the connection between the two modeling paradigms.

math.AP↗

arXiv2025

A 3D-1D-0D Multiscale Model of the Neuro-Glial-Vascular Unit for Synaptic and Vascular Dynamics in the Dorsal Vagal Complex

Cerebral blood flow regulation is critical for brain function, and its disruption is implicated in various neurological disorders. Many existing models do not fully capture the complex, multiscale interactions among neuronal activity, astrocytic signaling, and vascular dynamics--especially in key brainstem regions. In this work, we present a 3D-1D-0D multiscale computational framework for modeling the neuro-glial-vascular unit (NGVU) in the dorsal vagal complex (DVC). Our approach integrates a quadripartite synapse model--which represents the interplay among excitatory and inhibitory neurons, astrocytes, and vascular smooth muscle cells--with a hierarchical description of vascular dynamics that couples a three-dimensional microcirculatory network with a one-dimensional macrocirculatory representation and a zero-dimensional synaptic component. By linking neuronal spiking, astrocytic calcium and gliotransmitter signaling, and vascular tone regulation, our model reproduces key features of functional hyperemia and elucidates the feedback loops that help maintain cerebral blood flow. Simulation results demonstrate that neurotransmitter release triggers astrocytic responses that modulate vessel radius to optimize oxygen and nutrient delivery. This integrated framework, to our knowledge the first model to combine these elements for the NGVU in the DVC, provides a robust and modular platform for future investigations into the pathophysiology of cerebral blood flow regulation and its role in autonomic control, including the regulation of stomach function.

q-bio.TO↗

arXiv2025

Constitutive Kolmogorov-Arnold Networks (CKANs): Combining Accuracy and Interpretability in Data-Driven Material Modeling

Hybrid constitutive modeling integrates two complementary approaches for describing and predicting a material's mechanical behavior: purely data-driven black-box methods and physically constrained, theory-based models. While black-box methods offer high accuracy, they often lack interpretability and extrapolability. Conversely, physics-based models provide theoretical insight and generalizability but may not capture complex behaviors with the same accuracy. Traditionally, hybrid modeling has required a trade-off between these aspects. In this paper, we show how recent advances in symbolic machine learning, specifically Kolmogorov-Arnold Networks (KANs), help to overcome this limitation. We introduce Constitutive Kolmogorov-Arnold Networks (CKANs) as a new class of hybrid constitutive models. By incorporating a post-processing symbolification step, CKANs combine the predictive accuracy of data-driven models with the interpretability and extrapolation capabilities of symbolic expressions, bridging the gap between machine learning and physical modeling.

physics.comp-ph↗

arXiv2024

Active partitioning: inverting the paradigm of active learning

Datasets often incorporate various functional patterns related to different aspects or regimes, which are typically not equally present throughout the dataset. We propose a novel, general-purpose partitioning algorithm that utilizes competition between models to detect and separate these functional patterns. This competition is induced by multiple models iteratively submitting their predictions for the dataset, with the best prediction for each data point being rewarded with training on that data point. This reward mechanism amplifies each model's strengths and encourages specialization in different patterns. The specializations can then be translated into a partitioning scheme. The amplification of each model's strengths inverts the active learning paradigm: while active learning typically focuses the training of models on their weaknesses to minimize the number of required training data points, our concept reinforces the strengths of each model, thus specializing them. We validate our concept -- called active partitioning -- with various datasets with clearly distinct functional patterns, such as mechanical stress and strain data in a porous structure. The active partitioning algorithm produces valuable insights into the datasets' structure, which can serve various further applications. As a demonstration of one exemplary usage, we set up modular models consisting of multiple expert models, each learning a single partition, and compare their performance on more than twenty popular regression problems with single models learning all partitions simultaneously. Our results show significant improvements, with up to 54% loss reduction, confirming our partitioning algorithm's utility.

cs.LG↗

arXiv2024

A Calibration Approach for Elasticity Estimation with Medical Tools

Soft tissue elasticity is directly related to different stages of diseases and can be used for tissue identification during minimally invasive procedures. By palpating a tissue with a robot in a minimally invasive fashion force-displacement curves can be acquired. However, force-displacement curves strongly depend on the tool geometry which is often complex in the case of medical tools. Hence, a tool calibration procedure is desired to directly map force-displacement curves to the corresponding tissue elasticity.We present an experimental setup for calibrating medical tools with a robot. First, we propose to estimate the elasticity of gelatin phantoms by spherical indentation with a state-of-the-art contact model. We estimate force-displacement curves for different gelatin elasticities and temperatures. Our experiments demonstrate that gelatin elasticity is highly dependent on temperature, which can lead to an elasticity offset if not considered. Second, we propose to use a more complex material model, e.g., a neural network, that can be trained with the determined elasticities. Considering the temperature of the gelatin sample we can represent different elasticities per phantom and thereby increase our training data.We report elasticity values ranging from 10 to 40 kPa for a 10% gelatin phantom, depending on temperature.

q-bio.TO↗

arXiv2024

ECToNAS: Evolutionary Cross-Topology Neural Architecture Search

We present ECToNAS, a cost-efficient evolutionary cross-topology neural architecture search algorithm that does not require any pre-trained meta controllers. Our framework is able to select suitable network architectures for different tasks and hyperparameter settings, independently performing cross-topology optimisation where required. It is a hybrid approach that fuses training and topology optimisation together into one lightweight, resource-friendly process. We demonstrate the validity and power of this approach with six standard data sets (CIFAR-10, CIFAR-100, EuroSAT, Fashion MNIST, MNIST, SVHN), showcasing the algorithm's ability to not only optimise the topology within an architectural type, but also to dynamically add and remove convolutional cells when and where required, thus crossing boundaries between different network types. This enables researchers without a background in machine learning to make use of appropriate model types and topologies and to apply machine learning methods in their domains, with a computationally cheap, easy-to-use cross-topology neural architecture search framework that fully encapsulates the topology optimisation within the training process.

cs.LG↗

arXiv2023

A homogenized constrained mixture model of cardiac growth and remodeling: Analyzing mechanobiological stability and reversal

Cardiac growth and remodeling (G&R) patterns change ventricular size, shape, and function both globally and locally. Biomechanical, neurohormonal, and genetic stimuli drive these patterns through changes in myocyte dimension and fibrosis. We propose a novel microstructure-motivated model that predicts organ-scale G&R in the heart based on the homogenized constrained mixture theory. Previous models, based on the kinematic growth theory, reproduced consequences of G&R in bulk myocardial tissue by prescribing the direction and extent of growth but neglected underlying cellular mechanisms. In our model, the direction and extent of G&R emerge naturally from intra- and extra cellular turnover processes in myocardial tissue constituents and their preferred homeostatic stretch state. We additionally propose a method to obtain a mechanobiologically equilibrated reference configuration. We test our model on an idealized 3D left ventricular geometry and demonstrate that our model aims to maintain tensional homeostasis in hypertension conditions. In a stability map, we identify regions of stable and unstable G&R from an identical parameter set with varying systolic pressures and growth factors. Furthermore, we show the extent of G&R reversal after returning the systolic pressure to baseline following stage 1 and 2 hypertension. A realistic model of organ-scale cardiac G&R has the potential to identify patients at risk of heart failure, enable personalized cardiac therapies, and facilitate the optimal design of medical devices.

q-bio.TO↗

arXiv2023

Viscoelastic Constitutive Artificial Neural Networks (vCANNs) $-$ a framework for data-driven anisotropic nonlinear finite viscoelasticity

The constitutive behavior of polymeric materials is often modeled by finite linear viscoelastic (FLV) or quasi-linear viscoelastic (QLV) models. These popular models are simplifications that typically cannot accurately capture the nonlinear viscoelastic behavior of materials. For example, the success of attempts to capture strain rate-dependent behavior has been limited so far. To overcome this problem, we introduce viscoelastic Constitutive Artificial Neural Networks (vCANNs), a novel physics-informed machine learning framework for anisotropic nonlinear viscoelasticity at finite strains. vCANNs rely on the concept of generalized Maxwell models enhanced with nonlinear strain (rate)-dependent properties represented by neural networks. The flexibility of vCANNs enables them to automatically identify accurate and sparse constitutive models of a broad range of materials. To test vCANNs, we trained them on stress-strain data from Polyvinyl Butyral, the electro-active polymers VHB 4910 and 4905, and a biological tissue, the rectus abdominis muscle. Different loading conditions were considered, including relaxation tests, cyclic tension-compression tests, and blast loads. We demonstrate that vCANNs can learn to capture the behavior of all these materials accurately and computationally efficiently without human guidance.

cond-mat.mtrl-sci↗

arXiv2022

Ultrasound Shear Wave Elasticity Imaging with Spatio-Temporal Deep Learning

Ultrasound shear wave elasticity imaging is a valuable tool for quantifying the elastic properties of tissue. Typically, the shear wave velocity is derived and mapped to an elasticity value, which neglects information such as the shape of the propagating shear wave or push sequence characteristics. We present 3D spatio-temporal CNNs for fast local elasticity estimation from ultrasound data. This approach is based on retrieving elastic properties from shear wave propagation within small local regions. A large training data set is acquired with a robot from homogeneous gelatin phantoms ranging from 17.42 kPa to 126.05 kPa with various push locations. The results show that our approach can estimate elastic properties on a pixelwise basis with a mean absolute error of 5.01+-4.37 kPa. Furthermore, we estimate local elasticity independent of the push location and can even perform accurate estimates inside the push region. For phantoms with embedded inclusions, we report a 53.93% lower MAE (7.50 kPa) and on the background of 85.24% (1.64 kPa) compared to a conventional shear wave method. Overall, our method offers fast local estimations of elastic properties with small spatio-temporal window sizes.

eess.IV↗

arXiv2022

A Versatile SPH Modeling Framework for Coupled Microfluid-Powder Dynamics in Additive Manufacturing: Binder Jetting, Material Jetting, Directed Energy Deposition and Powder Bed Fusion

Many additive manufacturing (AM) technologies rely on powder feedstock, which is fused to form the final part either by melting or by chemical binding with subsequent sintering. In both cases, process stability and resulting part quality depend on dynamic interactions between powder particles and a fluid phase, i.e., molten metal or liquid binder. The present work proposes a versatile computational modeling framework for simulating such coupled microfluid-powder dynamics problems involving thermo-capillary flow and reversible phase transitions. In particular, a liquid and a gas phase are interacting with a solid phase that consists of a substrate and mobile powder particles while simultaneously considering temperature-dependent surface tension and wetting effects. In case of laser-metal interactions, the effect of rapid evaporation is incorporated through additional mechanical and thermal interface fluxes. All phase domains are spatially discretized using smoothed particle hydrodynamics. The method's Lagrangian nature is beneficial in the context of dynamically changing interface topologies. Special care is taken in the formulation of phase transitions, which is crucial for the robustness of the computational scheme. While the underlying model equations are of a very general nature, the proposed framework is especially suitable for the mesoscale modeling of various AM processes. To this end, the generality and robustness of the computational modeling framework is demonstrated by several application-motivated examples representing the specific AM processes binder jetting, material jetting, directed energy deposition, and powder bed fusion. Among others, it is shown how the dynamic impact of droplets in binder jetting or the evaporation-induced recoil pressure in powder bed fusion leads to powder motion, distortion of the powder packing structure, and powder particle ejection.

cs.CE↗

arXiv2021

A novel smoothed particle hydrodynamics and finite element coupling scheme for fluid-structure interaction: the sliding boundary particle approach

A novel numerical formulation for solving fluid-structure interaction (FSI) problems is proposed where the fluid field is spatially discretized using smoothed particle hydrodynamics (SPH) and the structural field using the finite element method (FEM). As compared to fully mesh- or grid-based FSI frameworks, due to the Lagrangian nature of SPH this framework can be easily extended to account for more complex fluids consisting of multiple phases and dynamic phase transitions. Moreover, this approach facilitates the handling of large deformations of the fluid domain respectively the fluid-structure interface without additional methodological and computational efforts. In particular, to achieve an accurate representation of interaction forces between fluid particles and structural elements also for strongly curved interface geometries, the novel sliding boundary particle approach is proposed to ensure full support of SPH particles close to the interface. The coupling of the fluid and the structural field is based on a Dirichlet-Neumann partitioned approach, where the fluid field is the Dirichlet partition with prescribed interface displacements and the structural field is the Neumann partition subject to interface forces. To overcome instabilities inherent to weakly coupled schemes an iterative fixed-point coupling scheme is employed. Several numerical examples in form of well-known benchmark tests are considered to validate the accuracy, stability, and robustness of the proposed formulation. Finally, the filling process of a highly flexible thin-walled balloon-like container is studied, representing a model problem close to potential application scenarios of the proposed scheme in the field of biomechanics.

cs.CE↗

arXiv2021

What do cells regulate in soft tissues on short time scales?

Cells within living soft biological tissues seem to promote the maintenance of a mechanical state within a defined range near a so-called set-point. This mechanobiological process is often referred to as mechanical homeostasis. During this process, cells intimately interact with the fibers of the surrounding extracellular matrix (ECM). It remains poorly understood, however, what individual cells actually regulate during these interactions, and how these micromechanical regulations are translated to tissue level to lead to what we macroscopically call mechanical homeostasis. Herein, we examine this question by a combination of experiments, theoretical analysis and computational modeling. We demonstrate that on short time scales (hours) - during which deposition and degradation of ECM fibers can largely be neglected - cells appear to regulate neither the stress / strain in the ECM nor their own shape, but rather only the contractile forces that they exert on the surrounding ECM.

q-bio.CB↗

arXiv2021

A computational framework for modeling cell-matrix interactions in soft biological tissues

Living soft tissues appear to promote the development and maintenance of a preferred mechanical state within a defined tolerance around a so-called set-point. This phenomenon is often referred to as mechanical homeostasis. In contradiction to the prominent role of mechanical homeostasis in various (patho)physiological processes, its underlying micromechanical mechanisms acting on the level of individual cells and fibers remain poorly understood, especially, how these mechanisms on the microscale lead to what we macroscopically call mechanical homeostasis. Here, we present a novel finite element based computational framework that is constructed bottom up, that is, it models key mechanobiological mechanisms such as actin cytoskeleton contraction and molecular clutch behavior of individual cells interacting with a reconstructed three-dimensional extracellular fiber matrix. The framework reproduces many experimental observations regarding mechanical homeostasis on short time scales (hours), in which the deposition and degradation of extracellular matrix can largely be neglected. This model can serve as a systematic tool for future in silico studies of the origin of the numerous still unexplained experimental observations about mechanical homeostasis.

cs.CE↗