A computationally efficient symmetric diagonally dominant matrix projection-based Gaussian process approach

Although kernel approximation methods have been widely applied to mitigate the O(n3) cost of n×n matrix inverse in Gaussian process methods, they still face computational challenges. The ‘residual’ between covariance and approximating component is often discarded as it prevents reduction. In this paper, we propose a computationally efficient approach that achieves better efficiency, O(mn2), compared with standard when using m≪n data. proposed incorporates its symmetric diagonally dominant form which can be further approximated by Neumann series. We validated full approaches based variants, both on synthetic real air quality