Abstract
Despite its importance, choosing the struc- tural form of the kernel in nonparametric regression remains a black art. We define a space of kernel structures which are built compositionally by adding and multiplying a small number of base kernels. We present a method for searching over this space of struc- tures which mirrors the scientific discovery process. The learned structures can often decompose functions into interpretable com- ponents and enable long-range extrapolation on time-series datasets. Our structure search method outperforms many widely used ker- nels and kernel combination methods on a variety of prediction tasks.
Author
David Duvenaud, James Robert Lloyd, Roger Grosse, Joshua B. Tenenbaum, Zoubin Ghahramani
Journal
Proceedings of the 30 th International Conference on Machine Learning,
Paper Publication Date
2013
Paper Type
Astroinformatics