The proposed schedule contains talks of 45 minutes length with fifteen minutes of discussion.
|Thursday 9th June|
Coffee & Welcome
20:30 Dinner in Monsal Head Hotel
|Friday 10th June|
|16:30||Joaquin Quinonero Candela||Abstract|
If you haven't already decided what you are going to say, Carl and Joaquin have sent a list of issues that they would like to hear about. These are certainly things we should also discuss. If you are not giving a full talk you might like to bring some of your thoughts on some of these issues.
Applications for which Gaussian processes are particularly suited, and seem to perform better than other alternative modelling approaches.
Though in regression it seems that optimizing the marginal likelihood leads to good generalization performance, the same cannot be said for classification where at times maximizing the marginal likelihood makes the test error become worse.
the question is whether there exists enough (if any) well designed empirical comparisons that allow making assessments on the performance of Gaussian Processes compared to competing methods. If not, one may want to motivate the design of good empirical comparisons.
Stein's book "Interpolation of Spatial Data" claims that "the lengthscales" are not important, only the shape of the covariance function is. The ubiquitous squared exponential covariance function suffers from limitations (ie. it is too smooth). How much effort is yet to be devoted to investigating new covariance functions?
In classification, as well as in regression with general noise models, analytic inference is impossible, and use is made of approximations. The number and variety of these is high, and no clear consensus seems to exist on which are better than others.