Arpit Bansal

I am a PhD Candidate at the University of Maryland, College Park advised by Professor Tom Goldstein. My research primarily revolves around empirical analysis and optimization of Deep Neural Network architectures, focusing on enhancing their training and inference mechanisms. I have a pronounced interest in image synthesis, specifically in text-to-image generation models, involving methodologies to streamline image generation processes and an in-depth exploration of diffusion model mechanisms. Additionally, with the aim to build neural networks that can do logical extrapolation and learn problem-solving skills, I have built neural-networks that can learn algorithms and thus extrapolate from easy-to-hard tasks using recurrence. A complementary aspect of my research includes investigating the inherent behaviors and characteristics of neural networks in various scenarios. Prior to UMD, I worked as a software developer with Visa in Data Product Development divison and received my Bachelors and Masters (Dual Degree) from IIT Kharagpur in Electrical Engineering. My CV is available here.

Along with research, I enjoy playing keyboard, and hiking.