What was agreed
According to CERN, it and Mistral have signed a statement of intent to establish a long-term strategic partnership on developing foundation models for scientific research. The collaboration combines CERN's expertise in fundamental physics, scientific data and large-scale computing with Mistral's experience in training large models. Reuters reported on the partnership the same day and named the model Flare.
The planned model
At the core is a single pretrained Large Physics Model, which is meant to be the first foundation model for collision physics. It is to be trained on simulated proton-proton collisions at the Large Hadron Collider (LHC) and designed for several tasks, from reconstructing particle collisions and identifying particles to flagging unexpected signals and evaluating future detector designs.
Reconstruction today relies on a chain of specialised algorithms. Machine learning already plays a role in this, but a foundation model could cover the whole chain within one model.
Data basis
CERN states it has stored around one exabyte – one million terabytes – of experimental LHC data to date. With the HL-LHC, expected to begin operation around 2030, the number of collisions is set to increase tenfold.
Context
For German businesses the main signal is that a European model developer is working with a major European research institution on its own foundation models. According to CERN's announcement, spillover into applications beyond high-energy physics is possible but not guaranteed. The partnership is so far a statement of intent with no published term, funding or results.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
