Darshana Saravanan

I am a PhD student at Saarland University, advised by Prof. Michael Hahn. Previously, I was a Research Intern at Microsoft Research India where I explored AI for Science, under the mentorship of Dr. Navin Goyal. I completed my MS by Research in Computer Science at IIIT Hyderabad, where I worked on understanding compositionality in Vision Language Models. I was advised by Dr. Vineet Gandhi and had the opportunity to collaborate with Dr. Makarand Tapaswi and Dr. Naresh Manwani.

Email  /  GitHub  /  Google Scholar  /  LinkedIn

Research

I am currently interested in understanding how neural networks function, with a focus on representations, training dynamics, and generalization in the context of LLMs. I have also worked on analysis and interpretability of Vision Language Models in the past.

Publications

Investigating Mechanisms for In-Context Vision Language Binding

VELOCITI: Benchmarking Video-Language Compositional Reasoning with Strict Entailment

Pseudo-labelling meets Label Smoothing for Noisy Partial Label Learning

Design and source code from Jon Barron's website