arXiv subjects

Alexandra Stavrianidi

Publications and source records attributed to Alexandra Stavrianidi.

4 recordsLinked to original sources

arXiv2026

Physics-Informed Deep Learning for False Ventricular Tachycardia Alarm Reduction in the ICU

False ventricular tachycardia (VT) alarms are a leading contributor to alarm fatigue in intensive care units. We propose a deep learning framework combining a 1D SE-ResNet with ICU-realistic data augmentations and a physics-informed auxiliary reconstruction task based on the three-element Windkessel hemodynamic model, implemented as a differentiable forward simulation. By requiring the network's latent representation to produce physiologically plausible arterial pressure waveforms, artifact-driven ECG patterns are penalized while true VT remains coherent across modalities. Evaluated on the VTaC benchmark under a strict real-time protocol (10-second pre-alarm window), our method achieves a 5-point Challenge Score improvement over prior state-of-the-art. Ablation studies confirm that the physics-informed objective is the primary performance driver, providing gains in accuracy, 2x label efficiency, and more localized and clinically meaningful ECG segments.

cs.LG↗

arXiv2026

Long-time behavior for systems of Fisher-KPP type with interacting components

We study the long-time behavior of a triangular system of Fisher--KPP type with $k$ interacting components, associated with a reducible multitype branching Brownian motion with $k$ types of particles. For this cascading system, we prove convergence in shape of each component to the minimal-speed Fisher--KPP traveling wave and determine the front asymptotics up to the constant order. This yields a PDE proof of Conjecture 1.2 from [4] on the convergence in distribution of the centered maximum particle in a cascading branching Brownian motion. We also derive asymptotic front-location estimates for such systems with general Fisher--KPP nonlinearities.

math.AP↗

arXiv2026

On the density of the supremum of nonlinear SPDEs

We study the one-dimensional stochastic partial differential equation \begin{equation*} \frac{\partial u}{\partial t}(t,x) = -\kappa \frac{\partial^4 u}{\partial x^4}(t,x) + \rho \frac{\partial^2 u}{\partial x^2}(t,x) + b(u(t,x)) + \sigma(u(t,x))\, \dot W(t,x), \end{equation*} posed on a bounded spatial domain, where $u$ is understood in the random field sense and $\dot W(t,x)$ is space-time white noise. Depending on the value of $\kappa$, this equation includes the nonlinear stochastic heat equation with Dirichlet or Neumann boundary conditions, as well as the linearized stochastic Cahn-Hilliard equation with Neumann boundary conditions. We prove that the supremum of the solution admits a density with respect to Lebesgue measure. Our approach is based on Malliavin calculus, and in particular on the version of the Bouleau-Hirsch criterion for suprema developed by Nualart and Vives. One of the main difficulties lies in the analysis of the argmax set of the solution and in showing that the Malliavin derivative is almost surely nondegenerate on this set. As a byproduct of our arguments, we also establish H\"older continuity properties for the Malliavin derivative of the solution as an $L^2-$valued process in the regimes considered in this work.

math.AP↗

arXiv2025

Improving physics-informed neural network extrapolation via transfer learning and adaptive activation functions

Physics-Informed Neural Networks (PINNs) are deep learning models that incorporate the governing physical laws of a system into the learning process, making them well-suited for solving complex scientific and engineering problems. Recently, PINNs have gained widespread attention as a powerful framework for combining physical principles with data-driven modeling to improve prediction accuracy. Despite their successes, however, PINNs often exhibit poor extrapolation performance outside the training domain and are highly sensitive to the choice of activation functions (AFs). In this paper, we introduce a transfer learning (TL) method to improve the extrapolation capability of PINNs. Our approach applies transfer learning (TL) within an extended training domain, using only a small number of carefully selected collocation points. Additionally, we propose an adaptive AF that takes the form of a linear combination of standard AFs, which improves both the robustness and accuracy of the model. Through a series of experiments, we demonstrate that our method achieves an average of 40% reduction in relative L2 error and an average of 50% reduction in mean absolute error in the extrapolation domain, all without a significant increase in computational cost. The code is available at https://github.com/LiuzLab/PINN-extrapolation .

cs.LG↗