The primary motivation of my research is to discover novel approaches to automatically recognize patterns in large datasets and develop tools to answer questions that affect people. Upon noting that, even with the availability of powerful tools, most of the collected data is still unanalyzed and under utilized, I am primarily motivated by how useful inferences can be made with little extra effort on the data that is already available. In this context, I am working on problems in medicine, systemic risk in financial systems, and sports.

My masters’ thesis focused on methods to improve the quality of multi-parameter physiological signal data in realtime. Automated systems in an intensive care unit critically depend on uninterrupted real-time monitoring of the physiological signals, but unfortunately, these signals are often corrupted by artifacts, and missing data. Inter-patient variations and the temporal evolution of the signals further exacerbate the problem. I investigated how to reconstruct the corrupted segments using the information available in correlated signals and signal history.

Going forward, I intend to work on learning predictors for global systemic risk using financial data such as consumer spending, security lending, etc. I am particularly interested in balancing the need for application specific solutions and the desire for data agnostic methods that will be useful across heterogeneous domains.