Bayesian inference with multiple forward models, each one with different vector-valued outputs

UQ with UQLab Community Q&A and How To
1

Hi UQWorld!

Does UQLab support Bayesian inversion with multiple forward models, each one with a different vector-valued output? In that case, what is the right format of “Data(i).MOMap”? I could not find an example so I am working with multiple scalar-output forward models so far.

Thank you,
Giuseppe

2

Hi Giuseppe

Welcome to UQWorld :grinning:! Yes, that is possible. The MOMap maps tells UQLab which model model output to compare a certain supplied data point to. The proper way to specify the MOMap in the case of multiple forward models with vectorized outputs is by writing it as a matrix where the first row corresponds to the model index and the second row to the output index of the model defined in the first row.

Assume you have three forward models with two outputs each and a single data group. Assume further that every two measurement columns in Data.y correspond to a new forward model, you should then write the MOMap as

Data.MOMap = [1 1 2 2 3 3; ... % Model ID
              1 2 1 2 1 2];    % Output ID

I hope this helps! Let me know how it goes :slightly_smiling_face:!

3

Ciao Paul,
It worked perfectly fine :slight_smile:
Thanks a lot for your help!
Giuseppe