Research Interests
 |
What Excites Me?I am interested in computational aspects of unsupervised learning. I have proposed and analyzed stochastic optimization algorithms for solving linear and non-linear component analysis techniques including PCA, kernel PCA, PCA in noisy settings and its multiview siblings PLS and CCA. More recently, I have worked on deep learning theory. In particular, I am interested in approximation/optimization theoretic aspects of learning with ReLU activation functions, as well as understanding the implicit bias of algorithmic approaches such as dropout (here, here, here, and here) used in training deep networks. Elsewhere
|
orcid.org/0000-0002-9001-0688