arXiv subjects

Vegard Flovik

Publications and source records attributed to Vegard Flovik.

12 recordsLinked to original sources

arXiv2026

Importance Sampling for Statistical Certification of Viable Initial Sets

We study the problem of statistically certifying viable initial sets (VISs)---sets of initial conditions whose trajectories satisfy a given control specification. While VISs can be obtained from model-based methods, these methods typically rely on simplified models. We propose a simulation-based framework to certify VISs by estimating the probability of specification violations under a high-fidelity or black-box model. Since detecting these violations may be challenging due to their scarcity, we propose a sample-efficient framework that leverages importance sampling to target high-risk regions. We derive an empirical Bernstein inequality for weighted random variables, enabling finite-sample guarantees for importance sampling estimators. We demonstrate the proposed approach on two systems and show improved convergence of the resulting bounds on an adaptive cruise control benchmark.

eess.SY↗

arXiv2026

SALVE: Sparse Autoencoder-Latent Vector Editing for Mechanistic Control of Neural Networks

Deep neural networks achieve impressive performance but remain difficult to interpret and control. We present SALVE (Sparse Autoencoder-Latent Vector Editing), a unified "discover, validate, and control" framework that bridges mechanistic interpretability and model editing. Using an $\ell_1$-regularized autoencoder, we learn a sparse, model-native feature basis without supervision. We validate these features with Grad-FAM, a feature-level saliency mapping method that visually grounds latent features in input data. Leveraging the autoencoder's structure, we perform precise and permanent weight-space interventions, enabling continuous modulation of both class-defining and cross-class features. We further derive a critical suppression threshold, $α_{crit}$, quantifying each class's reliance on its dominant feature, supporting fine-grained robustness diagnostics. Our approach is validated on both convolutional (ResNet-18) and transformer-based (ViT-B/16) models, demonstrating consistent, interpretable control over their behavior. This work contributes a principled methodology for turning feature discovery into actionable model edits, advancing the development of transparent and controllable AI systems.

cs.LG↗

arXiv2025

Gaussian Process Surrogate Models for Efficient Estimation of Structural Response Distributions and Order Statistics

Engineering disciplines often rely on extensive simulations to ensure that structures are designed to withstand harsh conditions while avoiding over-engineering for unlikely scenarios. Assessments such as Serviceability Limit State (SLS) involve evaluating weather events, including estimating loads not expected to be exceeded more than a specified number of times (e.g., 100) throughout the structure's design lifetime. Although physics-based simulations provide robust and detailed insights, they are computationally expensive, making it challenging to generate statistically valid representations of a wide range of weather conditions. To address these challenges, we propose an approach using Gaussian Process (GP) surrogate models trained on a limited set of simulation outputs to directly generate the structural response distribution. We apply this method to an SLS assessment for estimating the order statistics $Y_{100}$, representing the 100th highest response, of a structure exposed to 25 years of historical weather observations. Our results indicate that the GP surrogate models provide comparable results to full simulations but at a fraction of the computational cost.

cs.LG↗

arXiv2024

Quantifying Distribution Shifts and Uncertainties for Enhanced Model Robustness in Machine Learning Applications

Distribution shifts, where statistical properties differ between training and test datasets, present a significant challenge in real-world machine learning applications where they directly impact model generalization and robustness. In this study, we explore model adaptation and generalization by utilizing synthetic data to systematically address distributional disparities. Our investigation aims to identify the prerequisites for successful model adaptation across diverse data distributions, while quantifying the associated uncertainties. Specifically, we generate synthetic data using the Van der Waals equation for gases and employ quantitative measures such as Kullback-Leibler divergence, Jensen-Shannon distance, and Mahalanobis distance to assess data similarity. These metrics en able us to evaluate both model accuracy and quantify the associated uncertainty in predictions arising from data distribution shifts. Our findings suggest that utilizing statistical measures, such as the Mahalanobis distance, to determine whether model predictions fall within the low-error "interpolation regime" or the high-error "extrapolation regime" provides a complementary method for assessing distribution shift and model uncertainty. These insights hold significant value for enhancing model robustness and generalization, essential for the successful deployment of machine learning applications in real-world scenarios.

cs.LG↗

arXiv2023

Point Cloud Data Simulation and Modelling with Aize Workspace

This work takes a look at data models often used in digital twins and presents preliminary results specifically from surface reconstruction and semantic segmentation models trained using simulated data. This work is expected to serve as a ground work for future endeavours in data contextualisation inside a digital twin.

cs.CV↗

arXiv2017

Generation of single skyrmions by picosecond magnetic field pulses

We numerically demonstrate an ultrafast method to create $\textit{single}$ skyrmions in a $\textit{collinear}$ ferromagnetic sample by applying a picosecond (effective) magnetic field pulse in the presence of Dzyaloshinskii-Moriya interaction. For small samples the applied magnetic field pulse could be either spatially uniform or nonuniform while for large samples a nonuniform and localized field is more effective. We examine the phase diagram of pulse width and amplitude for the nucleation. Our finding could ultimately be used to design future skyrmion-based devices.

cond-mat.mes-hall↗

arXiv2016

Describing synchronization and topological excitations in arrays of magnetic spin torque oscillators through the Kuramoto model

The collective dynamics in populations of magnetic spin torque oscillators (STO) is an intensely studied topic in modern magnetism. Here, we show that arrays of STO coupled via dipolar fields can be modeled using a variant of the Kuramoto model, a well-known mathematical model in non-linear dynamics. By investigating the collective dynamics in arrays of STO we find that the synchronization in such systems is a finite size effect and show that the critical coupling-for a complete synchronized state-scales with the number of oscillators. Using realistic values of the dipolar coupling strength between STO we show that this imposes an upper limit for the maximum number of oscillators that can be synchronized. Further, we show that the lack of long range order is associated with the formation of topological defects in the phase field similar to the two-dimensional XY model of ferromagnetism. Our results shed new light on the synchronization of STO, where controlling the mutual synchronization of several oscillators is considered crucial for applications.

cond-mat.mes-hall↗

arXiv2016

Thickness and temperature dependence of the magnetodynamic damping of pulsed laser deposited $\text{La}_{0.7}\text{Sr}_{0.3}\text{MnO}_3$ on (111)-oriented SrTi$\text{O}_3$

We have investigated the magnetodynamic properties of $\text{La}_{0.7}\text{Sr}_{0.3}\text{MnO}_3$ (LSMO) films of thickness 10, 15 and 30 nm grown on (111)-oriented SrTi$\text{O}_3$ (STO) substrates by pulsed laser deposition. Ferromagnetic resonance (FMR) experiments were performed in the temperature range 100--300 K, and the magnetodynamic damping parameter $α$ was extracted as a function of both film thickness and temperature. We found that the damping is lowest for the intermediate film thickness of 15 nm with $α\approx 2 \cdot 10^{-3}$, where $α$ is relatively constant as a function of temperature well below the Curie temperature of the respective films.

cond-mat.mes-hall↗

arXiv2016

Eddy-current effects on ferromagnetic resonance: Spin wave excitations and microwave screening effects

We investigate how controlling induced eddy currents in thin film ferromagnet-normal metal (FM/NM) structures can be used to tailor the local microwave (MW) fields in ferromagnetic resonance (FMR) experiments. The MW fields produced by eddy currents will in general have a relative phase shift with respect to the applied MW field which depends on the sample geometry. The induced fields can thus partially compensate the applied MW field, effectively screening the FM in selected parts of the sample. The highly localized fields produced by eddy currents enable the excitation of spin wave modes with non-zero wave vectors, in contrast to the uniform k = 0 mode normally excited in FMR experiments. We find that the orientation of the applied MW field is one of the key parameters controlling the eddy-current effects. The induced currents are maximized when the applied MW field is oriented perpendicular to the sample plane. Increasing the magnitude of the eddy currents results in a stronger induced MW field, enabling a more effective screening of the applied MW field as well as an enhanced excitation of spin wave modes. This investigation underlines that eddy currents can be used to control the magnitude and phase of the local MW fields in thin film structures.

cond-mat.mes-hall↗

arXiv2015

Tailoring the magnetodynamic properties of nanomagnets using magnetocrystalline and shape anisotropies

Magnetodynamical properties of nanomagnets are affected by the demagnetizing fields created by the same nanoelements. In addition, magnetocrystalline anisotropy produces an effective field that also contributes to the spin dynamics. In this article we show how the dimensions of magnetic elements can be used to balance crystalline and shape anisotropies, and that this can be used to tailor the magnetodynamic properties. We study ferromagnetic ellipses patterned from a 10 nm thick epitaxial Fe film with dimensions ranging from 50 x 150 nm to 150 x 450 nm. The study combines ferromagnetic resonance (FMR) spectroscopy with analytical calculations and micromagnetic simulations, and proves that the dynamical properties can be effectively controlled by changing the size of the nanomagnets. We also show how edge defects in the samples influence the magnetization dynamics. Dynamical edge modes localized along the sample edges are strongly influenced by edge defects, and this needs to be taken into account in understanding the full FMR spectrum

cond-mat.mes-hall↗

arXiv2015

Dynamic wettability alteration in immiscible two-phase flow in porous media: Effect on transport properties and critical slowing down

The change in contact angles due to the injection of low salinity water or any other wettability altering agent in an oil-rich porous medium is modeled by a network model of disordered pores transporting two immiscible fluids. We introduce a dynamic wettability altering mechanism, where the time dependent wetting property of each pore is determined by the cumulative flow of water through it. Simulations are performed to reach steady-state for different possible alterations in the wetting angle ($θ$). We find that deviation from oil-wet conditions re-mobilizes the stuck clusters and increases the oil fractional flow. However, the rate of increase in the fractional flow depends strongly on $θ$ and as $θ\to 90^\circ$, a critical angle, the system shows critical slowing down which is characterized by two dynamic critical exponents.

physics.flu-dyn↗

arXiv2015

Eddy current interactions in a Ferromagnet-Normal metal bilayer structure, and its impact on ferromagnetic resonance lineshapes

We investigate the effect of eddy currents on ferromagnetic resonance (FMR) in ferromagnet-normal metal (FM/NM) bilayer structures. Eddy-current effects are usually neglected for NM layer thicknesses below the microwave (MW) skin depth (approx. 800 nm for Au at 10 GHz). However, we show that in much thinner NM layers (10-100 nm of Au or Cu) they induce a phase shift in the FMR excitation when the MW driving field has a component perpendicular to the sample plane. This results in a strong asymmetry of the measured absorption lines. In contrast to typical eddy-current effects, the asymmetry is larger for thinner NM layers and is tunable through changing the sample geometry and the NM layer thickness.

cond-mat.mes-hall↗