Projects

Here are the links to the major projects I have undertaken. Click on the title for a detailed description.

LLM Unlearning via Gradient Ratio-Based Influence Estimation and Noise Injection

Proposed and developed a noise-injected selective LLM fine-tuning algorithm to enable targeted machine unlearning. This was work done at NVIDIA.




Mechanism Design for Heterogeneous Differentially Private Data Acquisition

Developed a payment mechanism to incentivize privacy-sensitive users to share their data for ML tasks while satisfying their heterogeneous differential privacy constraints




Compressive Sensing Matrix Design

Designed sensing matrices by minimizing weighted triple coherence (determined from training data). Also, proposed an approach to design weights to handle the rotation and scaling of images




Video Compression for Efficient Remote Support Video Storage and Retrieval

Video Compression algorithm specifically for Daikin video database. Uses a Hitomi Camera architecture-inspired compressed sensing algorithm



 
Supervised Detection of Tennis Ball from Camera Stream

Code implementation for the Ball detection task required in the autonomous operations of our rover. This was done as a part of the IITB Mars Rover Team.



 
Genetic Algorithms for Dimensionality Reduction of Hyperspectral Images Implementation of a Genetic search-based Algorithm for feature selection of Hyperspectral Images.



 
VAE-GANs for Compressive Medical Image Recovery

Code implementation of GAN+VAE network for image reconstruction from undersampled biomedical images.




Image Reconstruction for Parallel MRI

Implemented a modified version of GRAPPA algorithm on Xilinx Zynq-7000 FPGA Board for image reconstruction used in the indigenous MRI machine at IIT Bombay. This is set to be installed at AIIMS Delhi around October 2025 for trials.



 
Parallel Image Compression

Implemented image compression using PCA, DCT, and FFT and studied their computational speedup using OpenMP, MPI, and CUDA.