MSMrinal Kanti Sen

Mrinal Kanti Sen

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Mrinal Kanti Sen has a total of 243 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Structure‐Aware Bayesian Full Waveform Inversion via Diffusion‐Regularized Stein Variational Gradient Descent

    by Chao Li, Mrinal K. Sen, Sergey Fomel, Yangkang Chen

    Full waveform inversion (FWI) is a powerful tool for high‐resolution subsurface model reconstruction, but it remains highly nonlinear and ill‐posed, especially when the initial model is inaccurate or the acquired data are noisy and incomplete. Annealed Stein variational gradient descent (ASVGD) provides a particle‐based Bayesian framework for FWI by approximating the posterior distribution with an ensemble of velocity models, thereby enabling uncertainty quantification beyond a single deterministic solution. However, conventional ASVGD‐FWI commonly relies on simple handcrafted priors, such as Gaussian or smoothing assumptions, which are often insufficient to represent complex geological structures. To address this limitation, we introduce a learned diffusion model (DM) as an implicit regularization mechanism within the ASVGD‐FWI framework. Instead of explicitly evaluating the diffusion prior density, the trained DM acts as a learned geological projection operator that guides updated particles toward the distribution of plausible velocity models. In this way, the proposed method combines data consistency from FWI, posterior diversity from ASVGD, and structural realism from the diffusion prior. Numerical experiments demonstrate that the diffusion‐regularized ASVGD‐FWI produces more geologically consistent mean models, suppresses nonphysical artifacts, and yields uncertainty estimates that are more concentrated around structurally complex and poorly constrained regions. These results suggest that learned diffusion priors can effectively enhance Bayesian FWI by improving both inversion stability and the interpretability of posterior uncertainty.

  • Open Access

    Physics‐Constrained Variational Autoencoder for Uncertainty Quantification of Full Waveform Inversion

    by Abdelrahman Elmeliegy, Mrinal K. Sen, Arnab Dhara, Jennifer Harding, Hongkyu Yoon

    We propose a variational autoencoder framework to directly assess uncertainties in subsurface models produced by single‐ and multiparameter full waveform inversion (FWI). The new method does not require pretraining on labeled data, thus it significantly reduces computational cost and storage requirements. The framework takes the seismic shot gathers as input and returns a set of possible velocity models that fit the input data. The network has three key components: (a) an encoder that maps the input seismic shot gathers to feature distributions in the latent space, (b) a latent vector sampled directly from feature distributions, and (c) a decoder that maps the sampled latent vector to the velocity model space. We incorporated Deep Image Prior principles, leveraging convolutional layers and LeakyReLU activations to regularize the inversion and improve reconstruction. At each epoch, the reconstructed velocity model is passed to a finite difference partial differential equation solver for forward modeling. We then calculate the data misfit between the input seismic data and the simulated data. Additionally, we compute the FWI gradient by cross‐correlating the adjoint and forward wavefield; the FWI gradient is then back‐propagated to guide the update of the neural networks weights and biases. The injection of physics‐based gradients further constrains the inversion and enhancing convergence. We tested the framework on synthetic acoustic and elastic seismic data sets, inverting models of varying complexity and quantifying their associated uncertainties. The results show accurate mean velocity models with meaningful uncertainty estimates, highlighting the potential of the proposed method for practical FWI applications.

  • Open Access

    Fixed-Weight Linear PINNs: Hopfield Networks as Globally Convex Finite-Difference PDE Solvers with Seismic Applications

    by Mrinal K Sen

    We establish a mathematically rigorous framework for casting finite-difference (FD) discretizations of linear partial differential equations (PDEs) as a Hopfield neural-network energy minimization problem. For any FD discretizations producing a linear system $\mathbf{A}\mathbf{u}=\mathbf{b}$, we show that the naturalHopfield energy $\frac{1}{2}\mathbf{u}^{T}\mathbf{A}\mathbf{u}$ fails whenever $\mathbf{A}$ is indefinite (as in wave equations, convection-dominated problems, and many coupled PDE systems), and we prove that the correct Hopfield energy is the \emph{residual energy} $E(\mathbf{u})=\|\mathbf{A}\mathbf{u}-\mathbf{b}\|^{2}$,which is globally convex with a unique minimum at the FD solution for any non-singular $\mathbf{A}$. The resulting Hopfield gradient-descent dynamics are equivalent to steepest descent on the normal equations (H-LSQR), while the optimal-step-size variant is the Conjugate Gradient on Normal Equations (CGNE), converging in$O(\kappa(\mathbf{A}))$ iterations.The framework is applied to the frequency-domain acoustic wave equation as a worked example. The Helmholtz FD system is complex, sparse, and indefinite, making it an ideal test of the residual-energy approach. Using the real-split operator $\mAt$ and complex coordinate-stretching PML boundaries, the Hopfield energy is$E(\vu)=\|\mAt\vu-\vb\|^{2}$. We demonstrate an application of this algorithm by computing frequency domain acoustic wavefields using one simple and one complex velocity model. Numerical experiments confirm that CGNE achieves relative errors of \sim\!10^{-7}$--$10^{-8}$ while the fixed-step Hopfield (H-LSQR)underestimates pressure amplitude by 20--30\% due to $\kappa(\mAt^{T}\mAt)=\kappa(\mAt)^{2}\sim10^{6}$. Finally, we identfy or classify our Hopfield FD solver as as \textbf{Fixed-Weight Linear PINN (FWL-PINN)} — the first PINN with a globally convex loss, guaranteed convergence, no spectral bias, and machine-precision accuracy, and compare our formulation with standard PINNs and Fourier Neural Operators (GNO). However, unlike the standard PINNs and FNO, FWI-PINN cannot be used as a surrogate.

  • Open Access

    3‐D Hybrid Full Waveform Inversion of Earthquake Data for the Crustal Velocity Estimation

    by Shuo Zhang, Zeyu Zhao, Mrinal K. Sen, Shuhua Hu

    As the state‐of‐the‐art method, full waveform inversion (FWI) is being broadly used to investigate the Earth's interior. However, the classical local‐optimization‐based FWI, relying on a good initial model, often encounters local minima issues. Benefiting from advancements in computational power, recent studies have integrated FWI with global optimization methods to mitigate the local minima issue by expanding search of the parameter space. Here, we introduce a hybrid optimization framework for FWI (HFWI) that can escape local minima by incorporating gradient and stochastic perturbations, with the modified stopping criterion to reduce the impact of hyper‐parameters. Synthetic tests are conducted to validate the performance of HFWI on the estimation of 3‐D model parameters. Although the poorly‐constrained initial models increase the likelihood of being trapped in local minima, the HFWI, because of the randomness in the searching direction, generally approaches the globally optimized model, with the residual data misfit less than a half period. The inverted model can be used as a good starting point for a local optimization based FWI to further refine its accuracies. Application of the elastic HFWI method to the earthquake recordings in Oklahoma, we validate the existence of sedimentary layers in the retrieved elastic model, and are able to constrain the spatial pattern of major geological units, such as the Anadarko Basin and the Cherokee Platform, with the illumination‐dependent uncertainties. Our results illustrate the ability of HFWI to estimate a 3‐D crustal model with a poor initial model, and the potential of HFWI on offering more detailed structural information.


  • Top Cited Papers

    Bayesian Inference for Subsurface Geophysical Inverse Problems

    by Mingliang Liu, Dario Grana, Klaus Mosegaard, Mrinal K. Sen, Minghui Xu, Tapan Mukerji

    In subsurface studies, geophysical inverse modeling aims to infer key Earth physical properties, such as deep geological structures, lithology, and fluid distribution from indirect observations, particularly geophysical data, while rigorously quantifying uncertainty. These inverse problems are typically high‐dimensional and computationally demanding, requiring efficient probabilistic inference methods. Bayesian inversion provides a coherent statistical framework that integrates prior geological knowledge with observed data, enabling systematic uncertainty quantification in subsurface characterization. Gradient‐free Bayesian sampling methods have long been used to characterize complex posterior distributions and remain foundational in geophysical inversion. Recently, in scenarios where gradient information can be efficiently obtained, gradient‐informed Bayesian inference methods have emerged as effective alternatives. By leveraging the local geometry of the posterior, these methods enable more efficient exploration of high‐dimensional parameter spaces. Concurrently, deep learning has further enhanced Bayesian inversion by facilitating implicit geological priors, surrogate forward modeling, and automatic differentiation for efficient gradient computation. This review provides a comprehensive synthesis of both gradient‐free and gradient‐informed Bayesian inference techniques, with an emphasis on the latter, and examines their applications in seismic, electromagnetic, gravity, and multiphysics inverse problems. Building on these developments, we introduce Differentiable Bayesian Inversion as a potential unifying conceptual framework that integrates deep‐learning‐based geological prior parameterization, physics‐based or surrogate forward modeling, and probabilistic reasoning within a modular, differentiable architecture. We conclude by outlining open challenges and future research directions toward developing robust, interpretable, and uncertainty‐aware inversion frameworks for increasingly complex geoscientific applications.


  • Open Access

    Bayesian full waveform inversion with learned prior using deep convolutional autoencoder

    by Shuhua Hu, Mrinal K Sen, Zeyu Zhao, Abdelrahman Elmeliegy, Shuo Zhang

    Full waveform inversion (FWI) can be expressed in a Bayesian framework, where the associated uncertainties are captured by the posterior probability distribution (PPD). In practice, solving Bayesian FWI with sampling-based methods such as Markov chain Monte Carlo (MCMC) is computationally demanding because of the extremely high dimensionality of the model space. To alleviate this difficulty, we develop a deep convolutional autoencoder (CAE) that serves as a learned prior for the inversion. The CAE compresses detailed subsurface velocity models into a low-dimensional latent representation, achieving more effective and geologically consistent model reduction than conventional dimension reduction approaches. The inversion procedure employs an adaptive gradient-based MCMC algorithm enhanced by automatic differentiation-based FWI to compute gradients efficiently in the latent space. In addition, we implement a transfer learning strategy through online fine-tuning during inversion, enabling the framework to adapt to velocity structures not represented in the original training set. Numerical experiments with synthetic data show that the method can reconstruct velocity models and assess uncertainty with improved efficiency compared to traditional MCMC methods.

  • Open Access

    Mapping Spatiotemporal Variations of Near‐Surface Seismic Velocity to Monitor Groundwater in Central Oklahoma

    by Shuo Zhang, Bingxu Luo, Yujie Zheng, David Lumley, Mrinal K. Sen, Hejun Zhu

    Global climate changes have intensified drought events, necessitating accurate groundwater monitoring. Traditional methods often lack the spatial and temporal resolution to track groundwater variations. This study calculates instantaneous travel‐time shifts, , from ambient noise cross‐correlation functions in central Oklahoma and projects them onto 2‐D near‐surface seismic velocity maps by using sensitivity kernels of coda‐wave interferometry. The lateral variations in these maps align with major bedrock aquifers in Oklahoma, revealing correlations between near‐surface seismic velocity, surface deformation, and gravity field, highlighting the role of water storage across time and space. The maps suggest that near‐surface seismic velocity responds are sensitive to groundwater dynamics, including discharge, recharge, and lateral flow. The confined and unconfined aquifers in Oklahoma, with diverse depth ranges and adaptive capacities, are characterized by distinct spatiotemporal patterns in our maps. This study demonstrates the potential of ambient seismic noise to improve spatial and temporal resolution in groundwater monitoring.


  • Bayesian full-waveform inversion using adaptive Markov chain Monte Carlo methods

    by Shuhua Hu, Zeyu Zhao, Mrinal K Sen

    Uncertainty quantification is crucial for seismic full-waveform inversion (FWI), which is a highly ill-posed inverse problem. In the framework of Bayesian inference, Markov chain Monte Carlo (MCMC) sampling algorithms can be used to quantify the uncertainties of FWI. However, the MCMC algorithm for FWI is faced with challenges in parameter tuning and low efficiency, which can be nontrivial for large-scale FWI problems. To solve this issue, we introduce adaptive MCMC algorithms for FWI, in which the step length is automatically tuned to optimize the sampling efficiency. Furthermore, we compare different preconditioning matrices for the proposal distribution, including an adaptive posterior covariance matrix estimated using previous samples and a local Hessian matrix. The method is implemented with an acoustic FWI with a frequency-domain finite-difference solver. The synthetic Marmousi and the 2004 BP velocity benchmark models are used to verify the effectiveness of our method. Numerical results suggest that a local Hessian matrix is preferred compared with the online estimated sample covariance matrix to be used as the preconditioning matrix for the proposal distribution. Numerical results using different starting models, data noise, and dimensionalities demonstrate the effectiveness of our method. By introducing the adaptive step length and the appropriate preconditioning matrix in the proposal distribution, the algorithm is able to draw samples efficiently from the posterior probability distribution of the FWI problem. The statistical features of the posterior samples are used to analyze uncertainties for the FWI problem.


  • Open Access

    Prediction Of CO2 Saturation and Elastic Properties Using Capillary Pressure-Based Rock Physics Model

    by Sohini Dasgupta, Denis Kiyashchenko, Asiya Kudarova, Suraj Pawar, Jonathan Hopkins, Pandu Devarakota, Kanglin Wang, Mrinal K Sen

    Carbon capture and Storage (CCS) plays an essential role in global decarbonisation. Accurate monitoring of injected CO2 volumes and migration is critical for the success of carbon sequestration projects. Geophysical measurements using time-lapse seismic surveys, well logs, core experiments, and reservoir simulations are widely used to monitor CO2 behaviour (potential leakage, plume movement, pressure, and saturation changes) over time for CCS sites. Supercritical CO2, exhibiting a hybrid of gas and liquid properties, is injected into the system. Seismic responses are susceptible to even a small concentration of gaseous CO2, thus making it difficult to pinpoint saturation numbers. The accuracy of a rock physics model heavily relies on the input parameters, such as rock properties (porosity, permeability) and fluid behaviour, which are often uncertain and can lead to inaccuracies in the predictions. Mostly, a homogeneous fluid distribution assumption is employed which may deviate from realistic CO2 saturation scenarios because of existing heterogeneities and in-situ conditions. This assumption might lead to a distorted saturation profile, uncertainties in the volume of CO2 injection, mismatch with actual field results, and long-term monitoring. In this paper, we focus on the choice of the fluid distribution model and how it impacts the characterisation of CO2 saturation mainly for the Quest CCS, Alberta fields, and the classic Sleipner, North Sea, project. We find that the two saturation states (homogeneous and patchy) at different durations of a CCS project can be captured using a capillary pressure-based rock physics model. We validate our results using core and well-log data, and we also observe the effects in reservoir simulation models. In the spirit of addressing the non-uniqueness of saturation analysis, this modeling strategy has the potential to become a key method for robust subsurface CO2 saturation monitoring.