Dheeraj Varghese

I’m a PhD candidate at the VIS Lab, supervised by Cees Snoek, with Efstratios Gavves as co-supervisor. I work on generalist multimodal foundation models as part of the Horizon Europe ELLIOT project. My research focuses on designing architectures that combine multiple modalities into a shared space, aiming for models that generalize well and assist meaningfully across a wide range of tasks.

Previously, I worked on multilingual image generation using a unified diffusion-autoregression backbone with Mohammad M. Derakhshani, and explored curriculum learning in vision-language models under the supervision of Yuki Asano.

At my core, I’m an applied engineer with an enthusiasm for recreating intelligence that serves as a tool, to make tasks easier for the human user. Sample efficiency in learning, blurring the context window, and unified representation spaces - all capture my attention at the moment.