Hi! I am a 1st year PhD student at the department of Operations Research, school of Industrial Engineering at Purdue University. I am working at the Stochastic Systems Lab, under the guidance of Prof. Harsha Honnappa. My research is at the intersection of stochastic processes and machine learning. Specifically, I am working on observing probabilistic score-based diffusion as a control system, and trying to incorporate such models to work on stochastic data samples. My aim is to simplify a complex SDE based sampler to a simpler randomized ODE, which enables an easier understanding of stability and convergence of such ML models. Feel free to reach if interested in knowing more!
Hi! Previously, I was a research assistant at the Oden Institute of Computational Engineering and Sciences at the University of Texas at Austin. I was advised by Prof. Chandrajit Bajaj and have wonderful collaborators at the Computational Visualization Center (CVC). This is also part of my 5th year thesis of my undergraduate in B.E. Computer Science and Integrated MSc. Mathematics from BITS Pilani, Goa.
During my last 2 years at BITS, I was a part of APPCAIR, BITS Pilani, mentored by Prof. Snehanshu Saha. Here I researched on correcting the likelihood misspecification of models via generative adversarial networks, with the aid of approximate Bayesian inference. This made several well known regressors much more robust to noise. I have also spent some wonderful summers interning at Graphics Research Group, IIIT Delhi with Dr. Ojaswa Sharma and at Indian Statistical Institute, Kolkata with Prof. Shubhamoy Maitra.
Apart from spending my time doing fun research, I hold a keen interest in music, a passion I got to pursue as part of the Music Society at BITS, where I gave several performances as a vocalist. I am also an avid reader, and very fond of cooking and sketching portraits.
Replacing conventional gradient descent based optimizers such as stochastic gradient descent, mini-batch gradient descent and ADAM with RL and policy based parameter learning and optimization. Trying the idea with trust region policy optimization, proximal policy optimization and soft actor-critic frameworks. Applying a new framework called optimal control gradient flow for the same.
Learning an RL based soft classifier to learn acceptor-donor atoms for formation of hydrogen bonds in protein chains, while in motion and while interacting with other molecules.
We developed a generative modeling paradigm called skipGAN which combines GANs and Approximate Bayesian Computing (ABC-GAN), with skip connections. The architecture aims at correcting likelihood misspecification in prior models. Our research demonstrated performance improvements in well-known priors like TabNet, CatBoost, and Stats model for regression tasks across various noisy datasets. Preprint
Developed a factored-STVGP (spatio-temporal variational Gaussian Process) framework for directly mapping any readily available RGB image to hyperspectral. Our model surpassed the current state-of-the-art on hyperspectral imaging benchmarks, thus enabling better classification and object detection.
Implementing the genetic algorithm on various setups of the modular synthesizer, called "organisms". Each organism produces a particular kind of sound. The organisms evolve by undergoing the graph-based processes of mutation and crossover, and natural selection, resulting in changes in connections and setups, evolving from fragments of sound to a section of a musical piece. Process Document, Appendix, Team page
Advising a group of juniors on a deeper understanding of generative models, in correspondence with Dr. Snehanshu Saha, BITS Goa.
CS F213: Discrete Mathematics, BITS Goa
CS F111: Computer Programming, BITS Goa
MATH F243: Graphs and Networks, BITS Goa
Mentored and taught over 200 students over multiple colleges in India.
Tutorials and doubt-clearing sessions for first year students in MATH F113, BITS Goa.