Recent Submissions

  • Tensor PDE model for biological network

    Astuto, Clarissa (2022-11-15) [Poster]
  • A Sequential Discontinuous Galerkin Scheme for Two-phase Poroelasticity

    Shen, Boqian (2022-11-15) [Poster]
    We formulate a numerical method for solving the two-phase flow poroelasticity equations. The scheme employs the interior penalty discontinuous Galerkin method and a sequential time-stepping method. The unknowns are the phase pressures and the displacement. Existence of the solution is proved. Three-dimensional numerical results show the accuracy and robustness of the proposed method.
  • Machine learning-assisted CO2 Storage Capacity Prediction in Deep Saline Aquifers: Uncertainty and Global Sensitivity Analysis

    Alqahtani, Abdulwahab (2022-11-15) [Poster]
    Geological CO2 sequestration (GCS) has been a practical approach used to mitigate global climate change. Uncertainty and sensitivity analysis of CO2 storage capacity prediction are essential aspects for large-scale CO2 sequestration. This work presents a rigorous machine learning-assisted (ML) workflow for the uncertainty and global sensitivity analysis of CO2 storage capacity prediction in deep saline aquifers. The proposed workflow comprises three main steps: 1) dataset generation we first identify the uncertainty parameters that impact CO2 storage in deep saline aquifers and then determine their corresponding ranges and distributions. We generate the required data samples by combining the Latin Hypercube Sampling (LHS) technique with high-resolution simulations. 2) ML model development a data-driven ML model is developed to map the nonlinear relationship between the input parameters and corresponding output interests from the previous step. The implementation of Bayesian optimization accelerates the tunning process of hyper-parameters instead of traditional trial-error analysis. 3) uncertainty and global sensitivity analysis Monte Carlo simulations based on the optimized surrogate are performed to explore the time-dependent uncertainty propagation of model outputs. Then the key contributors are identified by calculating the Sobol indices based on the global sensitivity analysis. The proposed workflow is accurate and efficient and could be readily implemented in field-scale CO2 sequestration in deep saline aquifers.
  • Generative Adversarial Zero-Shot Learning For Cold-start News Recommendation

    Alshehri, Manal (2022-11-15) [Poster]
    News recommendation models extremely rely on the interactive information between users and news articles to personalize the recommendation. Therefore, one of their most serious challenges is the cold-start problem (CSP). Their performance is dropped intensely for new users or new news. Zero-shot learning helps in synthesizing a virtual representation of the missing data in a variety of application tasks. Therefore, it can be a promising solution for CSP to generate virtual interaction behaviors for new users or new news articles. In this work, we utilize the generative adversarial zero-shot learning in building a framework, namely, GAZRec, which is able to address the CSP caused by purely new users or new news. GAZRec can be flexibly applied to any neural news recommendation model. According to the experimental evaluations, applying the proposed framework to various news recommendation baselines attains a significant AUC improvement of 1% - 21% in different cold start scenarios and 1.2% - 6.6% in the regular situation when both users and news have a few interactions.
  • Deep learning-based regularization of post-stack seismic inversion

    Romero, Juan (2022-11-15) [Poster]
    Seismic inversion is the prime method to estimate subsurface properties from seismic data. However, such inversion is a notoriously ill-posed inverse problem due to the band-limited and noisy nature of the data. Consequently, the data misfit term must be augmented with appropriate regularization that incorporates prior information about the sought-after solution. Conventionally, model-based regularization terms are problem-dependent and hand-crafted; this can limit the modeling capability of the inverse problem. Recently, a new framework has emerged under the name of Plug-and-Play (PnP) regularization, which suggests reinterpreting the effect of the regularizer as a denoising problem. Convolutional neural networks-based denoisers are state-of-the-art methods for image denoising: their adoption in the PnP framework has led to algorithms with improved capabilities over classical regularization in computer vision and medical imaging applications. In this work, we present a comparison between standard model-based and data-driven regularization techniques in post-stack seismic inversion and give some insights into the optimization and denoiser-related parameters tuning. The results on synthetic seismic data indicate that PnP regularization using a bias-free CNN-based denoiser with an additional noise map as input can outperform standard model-based methods.
  • GANs for 3D Porous media generation

    Corrales, Miguel (2022-11-15) [Poster]
    Linking the fluid flow at the pore scale and reservoir scale is an active area of research in projects related to CO2 storage and oil and gas recovery. A key obstacle to understanding such a process is the lack of physical samples from relevant geological areas. This issue can be addressed by generating accurate, digital representations of the rock samples available for numerical fluid flow simulations. A new promising avenue for generating realistic digital rock samples is opening up because of recent advancements in Machine Learning and Deep Generative Modeling. In particular, Generative Adversarial Networks (GANs) can learn complex distributions with high dimensions and produce high-quality samples. This study presents a Wasserstein GAN with gradient penalty (WGAN-GP) to generate high-quality porous media samples in 3D. Additionally, an evaluation metric set inspired by geometry, topology, and fluid flow properties is established to assess the generative quality.

View more