The research team formalized exploration of the SMEFT operator space as a reinforcement learning problem and used a lightweight actor-critic architecture with approximately 40K trainable parameters to identify diverse, high-quality theoretical hypotheses. The method was tested in complex environments spanning multiple areas of particle physics, including 132 observables and 24 measurements that deviate from Standard Model predictions by more than 1.5σ. Compared with independently selected random subsets and random sequence searches, the learned policies were able to construct multi-operator theories and improve overall fits.
Source:Hugging Face 社区论文精选 · Read the original
Leave a Reply