We implemented a GP model for CDC mortality rates using a squared-exponential (7) covariance structure.
To analyze and compare the dierent choices available within the GP framework we have experimented
with:
1. Other covariance kernel families, in particular Matern-5/2;
2. MLE and Bayesian approaches to inference of hyperparameters ;
3. A variety of mean function specications;
4. Choice of inhomogeneous noise variance (x):
For easier reading, the Figures and Tables below show the results for the Males; in the Appendix we
report the corresponding Figures and Tables for Females. Most of the conclusions are identical for both
genders; where appropriate we make further remarks