Our GP-grid algorithm uses the multiplicative structure of most common kernels to reduce runtime complexity from O(N^3) to O(D*N^((D+1)/D), where D is the number of dimensions. Importantly, our GP-grid algorithm is exact, requiring no approximations or sparsification procedures to perform inference.
We generalize GP-grid to handle incomplete grids and heteroscedastic noise, which importantly enables GP-grid to naturally incorporate known statistical properties of the data.
- Changes to previous version:
Initial Announcement on mloss.org.
No one has posted any comments yet. Perhaps you'd like to be the first?
Leave a comment
You must be logged in to post comments.