Barnabas Poczos (CMU) Title: Distribution Regression and its Applications

Abstract: 
The most common machine learning algorithms operate on
finite-dimensional vectorial feature representations. In
many applications, however, the natural representation of the data
consists of more complex objects,
for example functions, distributions, and sets, rather than
finite-dimensional vectors. In this talk
we will discuss machine learning algorithms that can operate directly
on these complex
objects. For this purpose, we use nonparametric statistical methods
that can consistently estimate the inner product, distance, and
certain kernel functions between distributions, sets, and other
objects. We will discuss applications in various scientific areas
including cosmology (e.g. estimating the mass of galaxy clusters,
finding anomalous galaxy clusters, estimating the cosmological
parameters of our Universe, accelerating cosmological simulations),
fluid dynamics (finding anomalous events in turbulence data),
neuroimaging, and agriculture