Time and date: 23 November 2022 at 2:00 pm | Location: Abacws 1.04 | Speaker: Matt Hutchings
Kernel matrices appear in a variety of machine learning problems, such as kernel support vector machines and kernel principal component analysis. In large-scale problems, it is often too computationally expensive to diagonalise these matrices, so in practice, low-rank approximations are desirable. We describe the Nyström method for low-rank approximations of SPSD matrices, and discuss efficient sequential sampling strategies based on the notion of squared-kernel discrepancy.
