Vansh Kapoor

Hey! I’m Vansh Kapoor, a graduate student at Carnegie Mellon University pursuing a Master’s in Machine Learning. I completed my undergraduate studies at the Indian Institute of Technology Bombay (IITB) in Electrical Engineering with Honors. My enthusiasm lies in exploring the realm of Machine Learning and applied probability & statistics. More specifically, I am interested in the fields of Deep Reinforcement Learning & Decision Making, Deep Learning, Online Algorithms, Stochastic Modeling, and Optimization.

During my undergraduate studies, I had the privilege of working with Prof. Jayakrishnan Nair and Prof. Nikhil Karamchandani. My research focused on a special case of discounted cost Partially Observable Markov Decision Processes (POMDPs), where the agent initially knows its state but remains uncertain about its exact state after taking an action unless a fixed cost is paid to reveal it. This work earned me the Undergraduate Research Award from IIT Bombay and was submitted as a first-author paper to AAAI-25.

For my Bachelor’s thesis, I collaborated with Google Research, where I am closely worked with Dr. Manish Jain and Prof. Nikhil Karamchandani to develop online learning algorithms for nullifying coordinated bot attacks that spread rumors on social networks like YouTube. My designed algorithm outperformed existing greedy methods and scaled effectively to large networks. Additionally, I served as a teaching assistant for a graduate-level course in Error Correcting Codes, assisting with course preparation, grading, and holding weekly office hours for over 40 students. In the summer of 2023, I gained industry experience through an internship at Google as a Silicon Intern. There, I improved the design verification process by 15% using Python-based automation for toggle coverage analysis and developed automated checkers for data retention flops in low-power mode applications.

I have also worked on various Deep Learning projects, including Text-to-Image Diffusion Models with Enhanced Semantic Understanding, Deep Recurrent Q-Networks (DRQN) for Partially Observable MDPs, CGAN-based Generative AI models & LSTM-based stock trading systems, and U-Net-based nuclei semantic segmentation.

I love solving brain teasers and puzzles and I somehow try fitting them into my daily routine. I enjoy in both playing table tennis and watching cricket (more of a cricket follower rather than a player )

If you have similar interests, feel free to hit me up with a mail here, I would love to chat.