arXiv subjects

Marzieh Alireza Mirhoseini

Publications and source records attributed to Marzieh Alireza Mirhoseini.

5 recordsLinked to original sources

arXiv2026

Observation-Guided Neural Surrogate Learning for Scientific Simulation Emulation: A Single-Gauge Flood-Inundation Proof of Concept

We present an observation-guided neural surrogate-learning framework for scientific simulation emulation, demonstrated on urban flood-inundation mapping. The framework combines LISFLOOD-FP hydrodynamic simulations with a real Gauge L stage record that is mapped to the simulation grid and converted to a datum-consistent local water-depth target before being used as single-site supervision. Focusing on a 256 x 256 crop around Gauge L in the Chicago metropolitan area, the method first constructs an ensemble-approximated Gaussian-process/local analogue surrogate (EnsCGP) to obtain a coarse flood-depth estimate and an uncertainty proxy. A U-Net-ASPP neural corrector then refines the coarse map using only simulation-derived and geospatial inputs: EnsCGP depth, the uncertainty proxy, rainfall, and spatial coordinates. The converted gauge-derived local depth is used only as a pointwise training target at the mapped gauge pixel; simulation-based losses are evaluated away from that pixel. Across temporally held-out events from 2013-2019, the emulator closely reproduces LISFLOOD-FP simulation targets outside the gauge-constrained pixel, with R^2 approximately 0.99 and mean absolute error below 0.01 m, and shows strong pointwise consistency with the converted Gauge L local depth target under the stated rolling-year protocol. We interpret these results as strong simulator-emulation agreement with pointwise observation-guided correction, not as independent validation of real-world inundation accuracy or as a complete operational flood-forecasting system.

physics.ao-ph↗

arXiv2025

Oja's plasticity rule overcomes several challenges of training neural networks under biological constraints

Deep neural networks have achieved impressive performance through carefully engineered training strategies. Nonetheless, such methods lack parallels in biological neural circuits, relying heavily on non-local credit assignment, precise initialization, normalization layers, batch processing, and large datasets. Biologically plausible plasticity rules, such as random feedback alignment, often suffer from instability and unbounded weight growth without these engineered methods, while Hebbian-type schemes fail to provide goal-oriented credit. In this study, we demonstrate that incorporating Oja's plasticity rule into error-driven training yields stable, efficient learning in feedforward and recurrent architectures, obviating the need for carefully engineered tricks. Our results show that Oja's rule preserves richer activation subspaces, mitigates exploding or vanishing signals, and improves short-term memory in recurrent networks. Notably, meta-learned local plasticity rules incorporating Oja's principle not only match but surpass standard backpropagation in data-scarce regimes. These findings reveal a biologically grounded pathway bridging engineered deep networks and plausible synaptic mechanisms.

q-bio.NC↗

arXiv2024

Learning Surrogate Rainfall-driven Inundation Models with Few Data

Flood hazard assessment demands fast and accurate predictions. Hydrodynamic models are detailed but computationally intensive, making them impractical for quantifying uncertainty or identifying extremes. In contrast, machine learning surrogates can be rapid, but training on scarce simulated or observed extreme data can also be ineffective. This work demonstrates the development of an effective surrogate model for flood hazard prediction by initializing deep learning (ResNet-18) with ensemble-approximated Conditional Gaussian Processes (EnsCGP) and finalizing it with a bias correction. The proposed methodology couples EnsCGP with a ResNet-18 architecture to estimate flood depth and uses ensemble optimal estimation for bias correction. The surrogate model was trained and evaluated using rainfall data from Daymet and hydrodynamic simulations from LISFLOOD-FP, spanning the period from 1981 to 2019. The training involved using data up to a certain year and testing on the subsequent year, iteratively progressing through the dataset. This process required approximately 100 training iterations and extensive data. Inundation depths are estimated rapidly at runtime (approximately 0.006 seconds per event). Results over multiple years in the current climate over Chicago demonstrate an average R-squared greater than 0.96, with median relative errors in flood depth estimates of about 1 percent.

physics.ao-ph↗

arXiv2023

Accelerated solutions of convection-dominated partial differential equations using implicit feature tracking and empirical quadrature

This work introduces an empirical quadrature-based hyperreduction procedure and greedy training algorithm to effectively reduce the computational cost of solving convection-dominated problems with limited training. The proposed approach circumvents the slowly decaying $n$-width limitation of linear model reduction techniques applied to convection-dominated problems by using a nonlinear approximation manifold systematically defined by composing a low-dimensional affine space with bijections of the underlying domain. The reduced-order model is defined as the solution of a residual minimization problem over the nonlinear manifold. An online-efficient method is obtained by using empirical quadrature to approximate the optimality system such that it can be solved with mesh-independent operations. The proposed reduced-order model is trained using a greedy procedure to systematically sample the parameter domain. The effectiveness of the proposed approach is demonstrated on two shock-dominated computational fluid dynamics benchmarks.

math.NA↗

arXiv2021

Model reduction of convection-dominated partial differential equations via optimization-based implicit feature tracking

This work introduces a new approach to reduce the computational cost of solving partial differential equations (PDEs) with convection-dominated solutions: model reduction with implicit feature tracking. Traditional model reduction techniques use an affine subspace to reduce the dimensionality of the solution manifold and, as a result, yield limited reduction and require extensive training due to the slowly decaying Kolmogorov $n$-width of convection-dominated problems. The proposed approach circumvents the slowly decaying $n$-width limitation by using a nonlinear approximation manifold systematically defined by composing a low-dimensional affine space with a space of bijections of the underlying domain. Central to the implicit feature tracking approach is a residual minimization problem over the reduced nonlinear manifold that simultaneously determines the reduced coordinates in the affine space and the domain mapping that minimize the residual of the unreduced PDE discretization. The nonlinear trial manifold is constructed by using the proposed residual minimization formulation to determine domain mappings that cause parametrized features to align in a reference domain for a set of training parameters. Because the feature is stationary in the reference domain, i.e., the convective nature of solution removed, the snapshots are effectively compressed to define an affine subspace. The space of domain mappings, originally constructed using high-order finite elements, are also compressed in a way that ensures the boundaries of the original domain are maintained. Several numerical experiments are provided, including transonic and supersonic, inviscid, compressible flows, to demonstrate the potential of the method to yield accurate approximations to convection-dominated problems with limited training.

math.NA↗