Brief Description

My research falls into two broad categories:

• Multilinear Algebra: Many data are naturally represented as multiway arrays or tensors, and as a result, tensor-based approaches have revolutionized feature extraction and compression. My research focuses on developing matrix-mimetic tensor frameworks that preserve desirable linear algebraic properties (think rank, orthogonality, and multiplication). The resulting framework looks and feels like matrix algebra and, as a result, we are able to naturally extend traditional algorithms to high-dimensions and obtain optimal representations of multiway data.

• Deep Learning: Deep neural networks (DNNs) have achieved undeniable success as high-dimensional function approximators in countless applications. However, this success comes at a significant hidden cost: the cost of training. Typically, the training problem is posed as a stochastic optimization problem with respect to the learnable DNN weights. With millions of weights, a non-convex objective function, and many hyperparameters to tune, solving the training problem well is no easy task. My research focuses on making training easier by exploiting commonly-used DNN architectures resulting in faster convergence, more accurate models, and automatic hyperparameter tuning.