Deep learning as a tool for heavy ion physics
This news is based an a press release by the Frankfurt Institute of Advanced Studies from 18 January 2018
A group of scientists from GSI, FIAS and the university Frankfurt used deep learning techniques to develop a tool for better understanding heavy ion collisions. The present study is a proof of principle study where Long-Gang Pang, Kai Zhou, Nan Su, Hannah Petersen, Horst Stöcker, former Scientific Director of GSI, and Xin-Nian Wang (University of California in Berkeley, USA) used more than 20,000 pictures from relativistic hydrodynamic simulations of heavy ion collisions, as they also occour in experiments with the GSI accelerators and the future FAIR accelerators, in a convolution neural network (CNN) to classify two regions in the phase diagram.
"We started the project when a human professional was defeated in the game of Go against AlphaGo designed by Google Deepmind. The news ignited our enthusiasm and we discussed a lot on whether artificial intelligence can assist scientists to tackle challenging unsolved scientific problems." explains Long-Gang Pang, a former FIAS postdoc from Hannah Petersens group, who is now at the University of California in Berkeley, USA.