Vaibhav Choudhary

Vaibhav Choudhary

Ph.D. student

Boston University

About me

I’m a fourth-year Ph.D. student at Boston University, where I work under the mentorship of Dr. Vivek Goyal in the Signal Transformation and Information Representation (STIR) Group. My research focuses on developing model-based machine-learning solutions for inverse problems in computational imaging. Currently, I am focused on creating forward models and physics-based estimators for particle beam microscopy, particularly in the areas of material analysis and edge detection for segmentation, with applications in semiconductor device characterization and metrology.

Outside of my research, I’m passionate about running and hiking. I’m also an aspiring astronomer with a growing interest in astrophotography! The background image on this page is a photo of the aurora borealis that I captured in Gloucester, MA, during a solar storm in early 2024.

My full CV is available here

Interests
  • Computational Imaging
  • Inverse Problems
  • Machine Learning
Education
  • Ph.D. in Electrical Engineering

    Boston University

  • MS in Electrical Engineering

    North Carolina State University

  • B.E. in Electronics and Communications Engineering

    NSIT, Delhi University

Research

Quantitative Imaging for Particle Beam Microscopy
(Manuscript in preparation) Developed new quantitative and physics-informed techniques for secondary electron imaging in particle beam microscopy, enhancing material analysis and segmentation processes. These methods were applied to improve semiconductor device characterization and metrology.Read more…
Destructive Imaging in Particle Beam Microscopy
Develop new imaging techniques to minimize surface damage and sample modification when imaging under a high dose, resulting in low-noise, high-fidelity images.
Passive Stereo-Hyperspectral Imaging and Ranging
Designing a new imaging model that integrates hyperspectral and stereo imaging techniques, improving predictions of temperature, material composition, depth, and texture for enhanced 3D scene characterization.
Passive Stereo-Hyperspectral Imaging and Ranging
Active Interposer-based Multi-Chiplet Architectures
(Under Review) Harsh Sharma, Vaibhav Choudhary, Janna Doppa, Vivek Goyal, Umit Ogras, Partha Pratim Pande

Projects

Google Summer of Code
This project leveraged the Programmable-Real Time Unit (PRU) microprocessor on the BeagleBone Black to serve as an SPI and I2C master controller, by developing Device Drivers on the ARM side and Firmware on the PRU side. The aim was to introduce additional serial interfaces without the need for extra hardware controllers or the burden of CPU-intensive bit-banging. Completed as part of Google Summer of Code with BeagleBoard.org, all code is openly available under the GPL v2 license, with required reproduction of copyright notices and conditions included in the source files. For more details on how to use the code for your application, please refer to the Wiki and watch the explanatory video.
Collection Style Transfer
In this project, we utilized a modified GAN called CycleGAN to perform collection style transfer, where an image is re-drawn in the style of a particular artist based on a set of input images. Additionally, we employed a CNN to quantitatively assess the style of images generated by CycleGAN, comparing them to a baseline image. Our results demonstrate that the CNN can successfully distinguish between images created by a specific artist and random images, allowing it to objectively measure the style of images produced by CycleGAN. Furthermore, we show that our CycleGAN outperforms the pre-trained CycleGAN provided by its original creators on certain datasets.
ADMM Optimization based Lasso and Ridge Regression
The Alternating Direction Method of Multipliers (ADMM) is a powerful algorithm that utilizes parallelization to solve optimization problems efficiently. This project focused on deriving the ADMM algorithm for Lasso and Ridge regression and comparing its convergence performance with the standard implementation in Scikit-learn.
ADMM Optimization based Lasso and Ridge Regression