I am a postdoctoral research fellow at Harvard University where I work with Prof. Hima Lakkaraju on the interpretability and robustness of deep learning models.
I completed my PhD at Idiap Research Institute & EPFL in Switzerland, where I was advised by Prof. François Fleuret. Before this, I completed my Masters (by Research) at the Indian Institute of Science, Bangalore advised by Prof. R. Venkatesh Babu.
During my PhD, I interned at Qualcomm AI Research in Amsterdam, where I worked with Tijmen Blankevoort.
I am broadly interested in mathematically and scientifically understanding deep learning models, and using this understanding to better apply them to real world applications. My research so far has focussed on the robustness, interpretability and efficiency of models at test time, but I am also interested in topics such as score-based generative modelling, and self-supervised representation learning. I am particularly interested in applications of machine learning in impactful domains such as healthcare and robotics. Some projects I have worked on in the past:
- understanding why gradients of deep classifiers are perceptually aligned with human perception (1, 2)
- providing a mathematical unification of feature attribution methods (3)
- proposing a method to train low-curvature neural nets that are “as linear as possible” (4)
My publications page contains more information.