Oct 2021 – Dec 2021 · 2 mos · Houston, Texas, United States

• Developed physics-based unsupervised deep learning models to obtain robust uncertainty estimates from the predictions. Uncertainty quantification from the employed model is not limited to the estimates of the distribution of the weights but also it also includes the estimates from the input data. The strategy utilized in the project will be further developed for evaluating subsurface properties using advanced physics.

Jun 2021 – Aug 2021 · 2 mos · Houston, Texas, United States

• Implemented multihead self-attention mechanism (vision transformers) to natural image and recognized the behavior and practicality of the employed deep learning model for other tasks. The is further developed and applied in the seismic domain for the purpose of salt segmentation from subsurface images. The newer approach shows encouraging results compared to the state-of-the-art salt segmentation approaches.

May 2020 – Aug 2020 · 3 mos · Houston, Texas, United States

• Developed parallel programs in Julia for stochastic full-waveform inversion on the HPC platform and tested the algorithm on three 2D synthetic models. Mitigated cycle-skipping in FWI, even if the initial model is a constant velocity background. Collaborated with Chevron’s scientists to understand the business needs and suggested research directions for realistic earth models.

May 2019 – Aug 2019 · 3 mos · Houston, Texas Area

• Developed a Fortran module using OpenMP for P- and S-wave mode separation algorithm in transversely isotropic media and integrated the module to BP’s production codes. Successfully applied this algorithm to further generate seismic images for all possible wave types. Compared the results using the newly developed technique with the existing methodologies and provided recommendations.

May 2017 – Aug 2017 · 3 mos · Gwanak-gu, Seoul, Korea

• Implemented the spectral-element-based finite-element method for the elastic wave modeling of the Earth’s subsurface. Processed and prepared the seismological data of the largest recorded earthquake in the Korean peninsula. Determined source signature (centroid-moment-tensor) and velocity model using subsurface waveform inversion.

Jan 2017 – Apr 2017 · 3 mos · Jeddah, Saudi Arabia

• Developed a high-performance elastic full-waveform inversion algorithm for anisotropic media on the GPU cluster. Included several utilities to the code: elastic wave modeling, free-surface, convoluted perfectly matched layer boundary conditions, anisotropy, gradient descent optimization strategy, and the objective function of the inverse problem. Investigated gradient preconditioning and data regularization schemes for performance optimization.

Jul 2016 – Jan 2017 · 6 mos · Hyderabad Area, India

• Developed CUDA C++-based algorithm for high-speed elastic full-waveform inversion in isotropic media for simultaneous estimation of P- and S-wave velocities. Tested the algorithm on the complex geologic model and observed around ten times computation speedup compare to the entirely CPU-based algorithm. Published the work in the 53 rd Annual Convention of Indian Geophysical Union at ISM Dhanbad, India

Jun 2016 – Jul 2016 · 1 mo · New Delhi Area, India

• Programmed OpenCV and machine-learning codes for automatic detection and recognition of traffic road signs available in India on panoramic-based images Developed a GUI for this program using PyQt.

May 2015 – Jul 2015 · 2 mos · Bengaluru Area, India

• Developed codes to understand the influence of various acquisition geometries on the reconstruction of seismic images. Three common different acquisition geometries (common offset, common midpoint, and asymmetric spread) are considered for the test. Provided recommendation that processing and inversion using common offset data produces high-quality images and requires less computation time.

Dec 2014 – Jan 2015 · 1 mo · Avcilar

• Designed a regression-based supervised back-propagation neural network (BP-ANN) to estimate the reliable reservoir parameters in the field data. Investigated and determined the correlated input parameters for the network. Collaborated with researchers in Computer Science and Geophysics departments and published two technical papers.

Jun 2014 – Jul 2014 · 1 mo · Dehradun