arXiv subjects

Mina Karimi

Publications and source records attributed to Mina Karimi.

9 recordsLinked to original sources

arXiv2026

LMM Modality Transfer: A Pre-requisite for Autonomous GIS Agents

AI models are becoming increasingly adept at understanding and processing spatial information, thereby facilitating agentic problem-solving in spatial tasks and workflows. However, most of the research on their spatial capabilities (e.g., spatial reasoning) has focused on the textual modality as input and output. This contrasts with the human approach to GIS workflows, where text and visual modalities are often used together, interchangeably, and in a complementary manner. Thus, to truly achieve an automated GIS analysis pipeline or carry out human-designed GIS workflows, AI models --- Large Multimodal Models (LMMs) in particular --- need to be able to seamlessly transition between image- and text-based modalities that are traditionally used in such workflows. We present a modality transfer task that (1) asks an LMM to first describe an input image of colored squares in a regular grid, and (2) asks a new LMM instance to re-generate an image of the original spatial scene using the textual description output by the former model. This task quantifies the ability of LMMs to transfer spatial information between image and text modalities. Ultimately, by examining the modality transfer capability of LMMs through the lens of spatial information theory, this work highlights a critical bottleneck: achieving strong and robust geospatial understanding in LMMs requires rigorous, multi-modal alignment. Our results indicate that recent LMMs (here from OpenAI) still struggle with modality transfer, when tasked with re-generating an image of a simple spatial grid of color squares.

cs.AI↗

arXiv2026

Assessing the Geographic Diversity of AI's Platial Representations in Image Generation

(Gen)AI diversity is not merely an ethical issue. From the perspective of geographic information science (GIScience), it could be interpreted as a function of uncertainty and as a form of cognitive bias, embedded in AI outputs. Recent work has sought to develop information-theoretic diversity measures and apply them to evaluate AI-chatbot outputs in a geographic context. As the AI ecosystem to which we are exposed on a daily basis becomes rapidly multimodal, we believe it is important to examine geographic diversity across various modalities. Focusing on images, this paper aims to fill this research gap. First, we select the GPT and DALL-E models as state-of-the-art examples and point out how assessing their geographic diversity involves various stages, including prompt revision and image generation. Then, taking inspiration from species diversity measures in ecological research, we incorporate similarity weighting into the measurement of geographic diversity. Next, we demonstrate how to evaluate geographic diversity in image generation through a case study. Our analysis reveals several counterintuitive findings. For instance, older models can exhibit greater geographic diversity despite producing lower-quality images, and prompt revision yields greater geographic diversity than image generation. At the same time, we observe explicit model homogeneity underlying the lack of geographic diversity, as the selected models consistently depict the same prototypical geo-specific feature or similar features. This is concerning, as it risks producing stereotypical representations of places.

cs.CY↗

arXiv2025

Impact of gas/liquid phase change of CO$_2$ during injection for sequestration

CO$_2$ sequestration in deep saline formations is an effective and important process to control the rapid rise in CO$_2$ emissions. The process of injecting CO$_2$ requires reliable predictions of the stress in the formation and the fluid pressure distributions -- particularly since monitoring of the CO$_2$ migration is difficult -- to mitigate leakage, prevent induced seismicity, and analyze wellbore stability. A key aspect of CO$_2$ is the gas-liquid phase transition at the temperatures and pressures of relevance to leakage and sequestration, which has been recognized as being critical for accurate predictions but has been challenging to model without \textit{ad hoc} empiricisms. This paper presents a robust multiphase thermodynamics-based poromechanics model to capture the complex phase transition behavior of CO$_2$ and predict the stress and pressure distribution under super- and sub- critical conditions during the injection process. A finite element implementation of the model is applied to analyze the behavior of a multiphase porous system with CO$_2$ as it displaces the fluid brine phase. We find that if CO$_2$ undergoes a phase transition in the geologic reservoir, the spatial variation of the density is significantly affected, and the migration mobility of CO$_2$ decreases in the reservoir. A key feature of our approach is that we do not \textit{a priori} assume the location of the CO$_2$ gas/liquid interface -- or even if it occurs at all -- but rather, this is a prediction of the model, along with the spatial variation of the phase of CO$_2$ and the change of the saturation profile due to the phase change.

physics.geo-ph↗

arXiv2023

Accelerated Computational Micromechanics for Reactive Flow in Porous Media

Reactive transport in permeable porous media is relevant for a variety of applications, but poses a significant challenge due to the range of length and time scales. Multiscale methods that aim to link microstructure with the macroscopic response of geo-materials have been developed, but require the repeated solution of the small-scale problem and provide the motivation for this work. We present an efficient computational method to study fluid flow and solute transport problems in periodic porous media. Fluid flow is governed by the Stokes equation, and the solute transport is governed by the advection-diffusion equation. We follow the accelerated computational micromechanics approach that leads to an iterative computational method where each step is either local or the solution of a Poisson's equation. This enables us to implement these methods on accelerators like graphics processing units (GPUs) and exploit their massively parallel architecture. We verify the approach by comparing the results against established computational methods and then demonstrate the accuracy, efficacy, and performance by studying various examples. This method efficiently calculates the effective transport properties for complex pore geometries.

math.NA↗

arXiv2023

A learning-based multiscale model for reactive flow in porous media

We study solute-laden flow through permeable geological formations with a focus on advection-dominated transport and volume reactions. As the fluid flows through the permeable medium, it reacts with the medium, thereby changing the morphology and properties of the medium; this in turn, affects the flow conditions and chemistry. These phenomena occur at various lengths and time scales, and makes the problem extremely complex. Multiscale modeling addresses this complexity by dividing the problem into those at individual scales, and systematically passing information from one scale to another. However, accurate implementation of these multiscale methods are still prohibitively expensive. We present a methodology to overcome this challenge that is computationally efficient and quantitatively accurate. We introduce a surrogate for the solution operator of the lower scale problem in the form of a recurrent neural operator, train it using one-time off-line data generated by repeated solutions of the lower scale problem, and then use this surrogate in application-scale calculations. The result is the accuracy of concurrent multiscale methods, at a cost comparable to those of classical models. We study various examples, and show the efficacy of this method in understanding the evolution of the morphology, properties and flow conditions over time in geological formations.

math.NA↗

arXiv2023

Deformation Decomposition versus Energy Decomposition for Chemo- and Poro- Mechanics

We briefly compare the structure of two classes of popular models used to describe poro- and chemo- mechanics wherein a fluid phase is transported within a solid phase. The multiplicative deformation decomposition has been successfully used to model permanent inelastic shape change in plasticity, solid-solid phase transformation, and thermal expansion, which has motivated its application to poro- and chemo- mechanics. However, the energetic decomposition provides a more transparent structure and advantages, such as to couple to phase-field fracture, for models of poro- and chemo- mechanics.

cond-mat.soft↗

arXiv2023

Hessian-informed Hamiltonian Monte Carlo for high-dimensional problems

We investigate the effect of using local and non-local second derivative information on the performance of Hamiltonian Monte Carlo (HMC) sampling methods, for high-dimension non-Gaussian distributions, with application to Bayesian inference and nonlinear inverse problems. The Riemannian Manifold Hamiltonian Monte Carlo (RMHMC) method uses second and third derivative information to improve the performance of the HMC approach. We propose using the local Hessian information at the start of each iteration, instead of re-calculating the higher order derivatives in all sub-steps of the leapfrog updating algorithm. We compare the result of Hessian-informed HMC method using the local and nonlocal Hessian information, in a test bed of a high-dimensional log-normal distribution, related to a problem of inferring soil properties.

stat.CO↗

arXiv2023

High-dimensional Nonlinear Bayesian Inference of Poroelastic Fields from Pressure Data

We investigate solution methods for large-scale inverse problems governed by partial differential equations (PDEs) via Bayesian inference. The Bayesian framework provides a statistical setting to infer uncertain parameters from noisy measurements. To quantify posterior uncertainty, we adopt Markov Chain Monte Carlo (MCMC) approaches for generating samples. To increase the efficiency of these approaches in high-dimension, we make use of local information about gradient and Hessian of the target potential, also via Hamiltonian Monte Carlo (HMC). Our target application is inferring the field of soil permeability processing observations of pore pressure, using a nonlinear PDE poromechanics model for predicting pressure from permeability. We compare the performance of different sampling approaches in this and other settings. We also investigate the effect of dimensionality and non-gaussianity of distributions on the performance of different sampling methods.

stat.AP↗

arXiv2021

Energetic Formulation of Large-Deformation Poroelasticity

The modeling of coupled fluid transport and deformation in a porous medium is essential to predict the various geomechanical process such as CO2 sequestration, hydraulic fracturing, and so on. Current applications of interest, for instance, that include fracturing or damage of the solid phase, require a nonlinear description of the large deformations that can occur. This paper presents a variational energy-based continuum mechanics framework to model large-deformation poroelasticity. The approach begins from the total free energy density that is additively composed of the free energy of the components. A variational procedure then provides the balance of momentum, fluid transport balance, and pressure relations. A numerical approach based on finite elements is applied to analyze the behavior of saturated and unsaturated porous media using a nonlinear constitutive model for the solid skeleton. Examples studied include the Terzaghi and Mandel problems; a gas-liquid phase-changing fluid; multiple immiscible gases; and unsaturated systems where we model injection of fluid into soil. The proposed variational approach can potentially have advantages for numerical methods as well as for combining with data-driven models in a Bayesian framework.

math.AP↗