Professor Anandkumar's interests lie in the area of large-scale machine learning and high dimensional statistics. She has been spearheading development of spectral methods which involve decompositions of matrices and tensors. Tensors are higher order extensions of matrices that can encode rich information about higher order moments in the data. She has been applying spectral methods for efficient reinforcement learning (RL) in partially observable Markov decision processes (POMDP). She is investigating various aspects of reinforcement learning, including efficient Thompson sampling and hierarchical reinforcement learning.