Internship at COMPASS in Prague, Czech Republic I arrived in Prague in February 2018 for a 2 month internship in the Tokamak Group of Experimental Physics. My work was to apply artificial neural networks (NNs) to plasma tomography algorithms under the supervision of Prof. Jan Mlynar.
Training a Neural Network for COMPASS SXR tomography
When arriving I first needed to get acquainted with the existing Minimum Fisher Regularization algorithm developed at COMPASS as well as the database containing the tomographic signals from the SXR cameras. Afterwards, using this algorithm, we performed a big range of reconstructions in order to have a data-set capable of training the neural network. Different NN architectures were then tested in order to obtain the smallest reconstruction error. Lastly different training methods were tested to allow the NN to be more stable under the occurrence of missing/faulty detectors as well as changes in the geometry of the cameras.
Life outside of IPP
During my weekends I took my time to explore the city of Prague being it the beautiful architecture or the never ending world of Czech pubs. Thanks to Czech Republic’s central Europe position I also had the opportunity to visit friends in Munich and hitchhike to Karlsruhe for an Easter weekend among Portuguese friends.
I would like to thank all the COMPASS team that welcomed me, especially Martin Imrisek for all the time spent reviewing code and doing the necessary tomographic reconstructions. I would also like to thank EUROfusion and FuseNet for the support provided, as without it this internship would hardly have been possible.
- Diogo Carvalho