Being a quant, my research interests lie at the intersection of stats and computer science. Hence, Machine Learning was a natural choice for me, even before it became a hype!

Currently, I am interested in the broad area of integrating symbolic knowledge with neural networks. Specifically, looking at how to train neural models for solving symbolic tasks? What are the different ways of learning better neural models by augmenting them with domain knowledge? How can we efficiently integrate differentiable symbolic solvers with neural networks for end-to-end training of neuro-symbolic architectures?

During my 1st year of PhD, I started with an ambitious project of automatically generating radiology reports from a brain CT scan. Unfortunately I moved on to other interesting things and that project eventually died. I hope to come back to similar research problems some time in future.