Barely Biased Learning for Gaussian Process Regression
This work addresses a specific bottleneck in Gaussian process regression for researchers, though it is incremental as it builds on prior scalable approximations without achieving computational competitiveness.
The paper tackles the bias-variance-computation trade-off in scalable Gaussian process regression by proposing a method that adaptively selects computation to ensure small bias in log marginal likelihood estimation, but the current implementation is not computationally competitive with existing approximations.
Recent work in scalable approximate Gaussian process regression has discussed a bias-variance-computation trade-off when estimating the log marginal likelihood. We suggest a method that adaptively selects the amount of computation to use when estimating the log marginal likelihood so that the bias of the objective function is guaranteed to be small. While simple in principle, our current implementation of the method is not competitive computationally with existing approximations.