Multi-fidelity Gaussian Process
This is a probabilistic framework based on Gaussian process regression and nonlinear autoregressive schemes that is capable of learning complex nonlinear and space-dependent crosscorrelations between models of variable fidelity, and it can effectively safeguard against low-fidelity models that provide wrong trends.
The regression results comparisons between conventional Gaussian process method and multi-fidelity Gaussian process method are shown here.
Codes reimplemented here is based on the idea from the following paper:
- P. Perdikaris, M. Raissi, A. Damianou, N. Lawrence, and G. E. Karniadakis, “Nonlinear information fusion algorithms for data-efficient multifidelity modelling,” Proc. R. Soc. A, vol. 473, no. 2198, p. 20160751, 2017.
See run.sh.
python run.py test2.toml
Autograd: https://github.com/HIPS/autograd