Athresh Karanam: “Developing Explainable and Efficient Machine Learning Algorithms for Deep Tractable Probabilistic Models”

The advent and seemingly ubiquitous success of deep learning in the past decade, in wide-ranging applications and across multiple data modalities, has been contingent upon availability of an ever-growing amount of data. Although this data-intensive approach is particularly effective in relatively data-rich domains, it imposes prohibitively expensive requirements in terms of data collection, labeling, and computational resources. The problem is compounded by the need for explainable models to build trust amongst domain experts in critical domains such as healthcare. To this end, I am broadly interested in developing effective and explainable machine learning algorithms that are data- and compute-efficient. In particular, I am interested in developing these techniques in the context of deep tractable probabilistic models - a class of probabilistic models capable of answering a wide variety of probabilistic queries exactly and efficiently.