For my research, I am using Gaussian Process Regression using 4-D inputs and 3 outputs. I have 3 separates GPs for the 3 separate outputs. I am using the Matern 3/2 covariance function. I am implementing everything in MATLAB. For real time purposes, I need the prediction to be carried out in around a millisecond (the performance requirement). I know there are a number of sparse, fast, parallel, online GP regression methods but none of them have been able to do such fast prediction. Does anybody know of any other fast prediction techniques that they have used and have had such performance ?
1
-
1$\begingroup$ How fast have you achieved and in what computing set up. What methods have you already tried? This might give people a reference point for making suggestions. $\endgroup$– Mark L. StoneCommented Aug 17, 2016 at 22:50
Add a comment
|