Dheeraj Varghese

I’m currently working on combining discrete diffusion and autoregression for multilingual multimodal models with Mohammad Mahdi Derakhshani and Cees Sneok at the VIS lab. Previously, for my thesis, I explored curriculum learning in Visual Language Models, supervised by Yuki Asano.

My experience with multimodality has made me curious to explore a few roads to improve efficiency in current models. I’d love to explore ‘Long context adaptation’ as a means to bring test and train closer; briding the gap and reducing the computational overhead of transformers. Also, I find hippocampal-cortical interactions very interesting 🧐