Speaker
Description
The Chiral Magnetic Effect (CME)—a QCD-anomaly–driven charge separation in the strong magnetic fields of non-central heavy-ion collisions—remains extremely difficult to isolate because flow-related backgrounds, especially from resonance decays, can mimic CME-like correlations. To address this weak-signal in a complicated background problem, we explore Transformer-based Artificial Intelligence models, originally developed for natural language processing, to study the CME in relativistic heavy-ion collisions. Unlike conventional neural networks that treat particle features more independently, Transformers use self-attention to model interactions among all particles in an event, aligning naturally with the system’s collective dynamics. We train on controlled synthetic datasets with known CME strength and use the learned event representation to regress a CME-sensitive proxy. We will show that the principle works: the model exhibits an approximately linear response to increasing CME strength in the presence of typical resonance-flow backgrounds, indicating complementary sensitivity beyond traditional observables.