Barnabás Póczos
Tuesday 9th June 2015
Time: 4.00pm
Basement Seminar Room
Alexandra House, 17 Queen Square, London, WC1N 3AR
Machine Learning on Functional Data
In this presentation we discuss two novel results in functional data
analysis. i) We present FuSSO, a functional analogue to the LASSO,
which can efficiently find a sparse set of functional input covariates
to regress a real-valued response against. ii) We will also discuss a
new method for function-to-function regression, when both input
covariates and output responses are functions from a nonparametric
function class. We analyze the statistical properties of these methods
and demonstrate their performance in several real-world problems from
cosmology, computer vision, and neuroimaging.