I broadly work in the areas of machine learning, computational learning theory and statistics.
Over the past few decades, an increase in the availability of data has resulted in focus on algorithms that can make sense of large, complex datasets. Unfortunately, these algorithms often fail when the data doesn't satisfy the modeling assumptions the algorithm designer has in mind. Another problem with modern machine learning algorithms is the resources they require to be trained. The carbon footprint of AI has become a serious problem today, with the resources required to train them doubling every few months.
My work focuses on developing principled algorithms which are (1) efficient and (2) robust to input misspecification.
To do this, I exploit classical as well as modern observations about structure in the data as well as underlying signals (such as being sparse, or being representable by modern generative models).
Contact s.my-first-name@gmail.com