python - Would it be possible to predict experimental data with a predictive model trained on simulated data?

Question

I am on a project of building a Machine Learning Model in material science. The goal is to build a predictive model with experimental data. Due to limited experimental cost and time, however, we don't expect enough data from an experiment. So we are thinking of training a model with simulated data from Finite or Discrete element simulations for instance, and evaluate the model against experimental data. But I am doubt on this mixing. Even though the simulation is modelled based on experimental parameters, there is no guarantee that the distribution of target outputs would correspond to that of experiments.

How do you think?

Answer 1

I do not think i can give you an accurate answer. However, training on simulated data and deployment on a world-like environment is usual practice in the High Energy Physics environment. They use Monte-Carlo high statistics simulations to train the model and evaluate their prediction performances using real-data collected by experiments:

An example

One thing you can do to limit the experimental-simulation disagreement would be to implement a domain adaptation layer into your network:

This is a nice article explaining domain adaptation

All these considerations may depend on the type of architecture you are going to develop, for which specific task (binary classification or other?), and how reliable are your simulations.