Mukilan Karuppasamy

I'm a predoctoral researcher at IISc, Bangalore, working with Anirban Chakraborty on visual instruction tuning for multimodal large language models. Previously, I had the opportunity to work with CV Jawahar at IIIT Hyderabad on concept-based interpretability in video action recognition, and with Yogesh Singh Rawat at CRCV, UCF, on robustness analysis of vision-language models. I was also a MITACS Globalink Fellow in 2023, working on distributed message passing with Amine Mezhghani and Faouzi Bellili at the University of Manitoba. Earlier, I worked on virtual try-on systems using GANs with Brejesh Lall at IIT Delhi.

Email  /  CV  /  Scholar  /  Github  /  Linkedin

Research

I'm interested in computer vision, interpretability, and robustness analysis. My current interest lies in understanding the theoretical basis of interpretability in overparameterized deep neural networks. Some papers are highlighted.

Proposed a concept based inherently intrepretable model for Video models using concept bottleneck model and token merging. This was applied in Driver Intention Prediction task.

Derived a collaborative signal estimation from multiple agents with different measurement channels through distributed message passing algorithm.