Speaker: Prof François-Xavier Briol, University College London
Abstract: To enable closed form conditioning, a common assumption in Gaussian process (GP) regression is independent and identically distributed Gaussian observation noise. This strong and simplistic assumption is often violated in practice, which leads to unreliable inferences and uncertainty quantification. Unfortunately, existing methods for robust GP regression break closed-form conditioning, which makes them less attractive to practitioners and significantly more computationally expensive. In this talk, I will discuss a recent line of work which has led to provably robust and conjugate Gaussian process (RCGP) regression at virtually no additional computational cost using generalised Bayesian inference. I will illustrate the desirable properties of RCGPs for problems ranging from Bayesian optimisation to sparse variational Gaussian process regression, spatio-temporal modelling, and multi-output regression. The latter will be illustrated in the context of cancer research, where we use RCGPs to perform robust modelling of the viability of cancer cells to varying doses of drugs.
-
Altamirano, M., Briol, F.-X., & Knoblauch, J. (2024). Robust and conjugate Gaussian process regression. International Conference on Machine Learning, PMLR 235, 1155–1185.
-
Laplante, W., Altamirano, M., Duncan, A., Knoblauch, J., & Briol, F.-X. (2025). Robust and conjugate spatio-temporal Gaussian processes. International Conference on Machine Learning (to appear).
-
Rooijakkers, J., Ronneberg, L., Briol, F.-X., Knoblauch, J. & Altamirano, M. (2025+). Multi-output robust and conjugate Gaussian process Regression. Under review.
This will be a free hybrid seminar. To register to attend remotely, please click here: https://cam-ac-uk.zoom.us/meeting/register/oURXKjAISECZrOvFPRw0Mw