An AI fed raw collider data reproduced famous particle masses on its own
A computer model was fed about a billion real particle collisions and told almost no physics. The fake collisions it then produced show the correct masses of several known particles, matching values physicists had already measured.
Why it matters
Physicists normally study collisions by comparing them to hand-built computer simulations. Those simulations eat enormous amounts of computer time, and every rule inside them was written in by a person. This work suggests much of that structure is already sitting in the recorded data, and can be pulled out of it directly. That could make future studies cheaper and faster, though so far the model only recreated physics that was already known and already present in the data it was trained on.
Who's behind it: Midori Kato, Kevin A. Urquia-Calderon, Inar Timiryasov and Oleg Ruchayskiy, Niels Bohr Institute, University of Copenhagen.
Summary by the Lemma AI · how we grade