A0727
Title: Mixture of normalizing flows for spherical density estimation
Authors: Tin Lok James Ng - Trinity College Dublin (Ireland) [presenting]
Andrew Zammit Mangion - University of Wollongong (Australia)
Abstract: The use of normalizing flows to model complex probability distributions has attracted much research interest in the machine-learning community in recent years. Normalizing flows offer great flexibility in modelling probability distributions, only requiring the specification of a simple reference distribution and a series of bijective transformations. More recently, research interests have shifted to developing normalizing flows for probability distributions on spaces with more complex geometries, such as spheres. However, using a global normalizing flow to model complex probability distributions proved challenging in some applications. The aim is to extend this framework to a mixture model by using normalizing flows as mixture components for density estimation on the sphere.