Computational Challenges for Kernel Methods on Massive Data

Abstract: Kernel machines such as the Support Vector Machine are attractive because they can approximate functions or decision boundaries arbitrarily well with enough training data. Unfortunately, methods that operate on the kernel matrix (Gram matrix) of massive data sets in high dimensions often scale poorly with the size of the training dataset. For example, a dataset with half a million training examples might take days to train on modern workstations. On the other hand, specialized algorithms for linear Support Vector Machines and regularized regression run much more quickly when the dimensionality of the data is small because they operate on the covariance matrix rather than the kernel matrix of the training data. In this talk, I will discuss several open problems, both computational and theoretical related to this interesting research area.

Event Topic:

Computational Mathematics & Statistics