Shashwat Singh

I am an AI research student, currently pursuing an Undergrad + MS by Research program at IIIT Hyderabad under my advisor Dr. Ponnurangam Kumaraguru. I am interested in the science of Deep Learning, AI interpretability, and pretraining.

Some specific ideas I find interesting are:

• Post-facto interpretability and controllability of pretrained models.
• Disentanglement of ‘atomic concepts’ in dense representations and their composition.
• Pretraining tasks for efficient and generalizable learning.

In my free time, I like to read about history and watch movies. I love Pirates of the Caribbean :)

Publications

I have had the pleasure of working with Dr. Shauli Ravfogel, Dr. Ryan Cotterell, Dr. Ponnurangam Kumaraguru, Dr. Makarand Tapaswi, Saujas Vaduguru, and Shashwat Goel. Under their mentorship and guidance I have worked on project topics including interpretability of Language Model representations, training dynamics, diffusion, and vision-language models.