Saurav K. Shastri

Experience

• 

  Crucible of Research and Innovation

  Jun 2016 - Jul 2016

  Research Intern
• 

  IEEE PESU Student Chapter

  Jun 2016 - Sept 2017

  • Treasurer

    Nov 2016 - Sept 2017

  • Event Manager

    Jun 2016 - Nov 2016
• 

  Crucible of Research and Innovation

  Jan 2017 - May 2017

  Undergraduate Researcher
• 

  Simon Fraser University

  May 2017 - Aug 2017

  Research Assistant

  MITACS Globalink Research InternDepartment of MathematicsAdvisor: Prof. Ben Adcock (SFU)Worked on self-calibration of accelerated multi-coil MRI (e.g., 4x acceleration) to recover unknown sensitivity maps and image

• 

  Indian Institute of Science (IISc)

  Jan 2018 - Jul 2019

  Spectrum Lab, Department of Electrical EngineeringAdvisor: Prof. Chandra Sekhar Seelamantula (IISc)Collaborator: Dr. Sunil Rudresh (IISc), Prof. Arun Thittai (IITM)Reconstructed 1-D and 2-D real ultrasound signals sampled at sub-Nyquist rates (10x times lower than standard sampling rate)Managed collaboration with IIT Madras, experimentally verifying the algorithm's capability to resolve thicknesses 3x below the ultrasound system's axial resolution limit.Worked on structured nonuniform sampling (SNS) and reconstruction for sub-Nyquist radar. The technique had a superior hit-rate and lower errors in delay and Doppler estimation compared to the uniform sampling (e.g., ~100% hit-rate at -10 dB SNR)Developed generalized design of compactly supported sampling kernels for 2-D Finite-rate-of-Innovation signals Show less

  • Project Assistant

    Jun 2018 - Jul 2019

  • Research Intern

    Jan 2018 - May 2018
• 

  The Ohio State University

  Aug 2019 - May 2025

  Advisor: Prof. Philip Schniter (OSU)Collaborators: Prof. Rizwan Ahmad (OSU), Prof. Christopher A. Metzler (UMD)Devised an expectation-consistent (EC) + diffusion-based posterior sampling method for inverse problems.Innovated deepEC, a novel EC approximation framework for solving non-linear inverse problems. In phase retrieval experiments, deepEC demonstrated 1.5 dB higher PSNR and required 5x fewer denoiser calls compared to competitors when recovering colored images from corrupted phaseless Fourier measurements.Developed a state-of-the-art autotuning plug-and-play (PnP) algorithm called D-GEC for Fourier-based forward operators that offer predictable error statistics at each iteration, unlike other PnP algorithms. On average, D-GEC’s PSNR performance exceeds its closest competitor by over 1.5 dB for 8x accelerated multi-coil MR image recovery. Crafted “corr+corr,” a deep neural network for denoising MR images affected by colored noise, surpassing its closest competitor by over 5 dB PSNR in specific noise scenarios.Engineered a robust auto-tuning PnP primal-dual splitting algorithm for 4x accelerated multi-coil MR image recovery which achieved 0.83 dB better PSNR than end-to-end deep neural network (U-Net).Authored and published multiple research papers in top-tier peer-reviewed signal processing journals and conferences, effectively communicating findings and advancing the field. Show less

  • Graduate Teaching Associate

    Aug 2024 - May 2025

  • Graduate Research Associate

    Aug 2019 - Dec 2024
• 

  Mitsubishi Electric Research Laboratories

  May 2023 - Aug 2023

  Intern

  Hosts: Dr. Hassan Mansour, Dr. Yanting MaPioneered a machine-learning-based approach for accurate underground imaging in a realistic ground-penetrating radar setup, leveraging a Fourier Neural Operator-based forward model and a neural network-based calibration. The method outperforms its competitor by over 4.5 dB in PSNR and is currently in the process of being patented.Initiated the use of diffusion model-based generative prior, implementing and validating its effectiveness.Collaborated with researchers to parallelize and optimize training data generation, achieving a 20x speed increase. Show less