This page hosts my masters dissertation and talks related to it.
This dissertation explored model search, model scoring and model evaluation of an artificial intelligent (AI) system. This AI performed data analysis through compositional kernel search and was called Automatic Bayesian Covariance Discovery (ABCD).
We introduced novel methods to stabilise the predictive performance of ABCD. From our experimental results, we show that our methods perform better extrapolation and can also explore a much wider model space in hopes of searching a richer interpretable structure.
Thus, this dissertation was titled as Automatic Bayesian Covariance Discovery with Stable Extrapolation or ABCD-SE for short.