- AutoML + Classical NN study. We have developed a novel way of applying AutoML techniques to the construction of linear regression models applied to experimental data. In a first stage we resorted to automated machine learning models to find the best preprocessing and model to fit our data. With this, we build a classical neural network in PyTorch implementing the configurations of the previous stage. In this way, using autoML techniques, model building and hyperparameter tuning is solved in a matter of seconds and by implementing the results of this part to classical PyTorch models we achieve more accurate results. (Paper of these results in progress).

​​​

- Different domain inputs? In nuclear physics experiments it is very common to carry out simulations in environments like Geant4 of our experiments. With these simulated data (Monte Carlo data) we train a ML model that we then apply to our experimental data. But, what if our simulation data is different from our experimental data?  In that case the model that has learned from the MC data cannot be applied to the experimental data, since what is known as a domain shift of the input data has occurred. The solution? The implementation of DANNs (Domain-Adversarial Neural Networks) that learn to switch from one input domain to another.  (Paper of these results in progress).