The FERMIACC: Agents for Particle Theory

Abstract

We present the FERMIACC, a scaffolded reasoning model built on OpenAI agents designed to autonomously generate and quantitatively validate theory hypotheses for high energy physics data at scale.

Figures (9)

• Figure 1: Schematic architecture of the FERMIACC. Blue boxes correspond to agent modules. Green shaded boxes correspond to integrated software components.
• Figure 2: A zoom in on the schematic architecture of the FERMIACC model-building module. A proposal agent takes inputs from the experimental analysis and database of previous FERMIACC runs and designs a new BSM explanation for a given anomaly. It is then passed between a critic and patching agent until the proposal passes, at which point it is transferred to FeynRules for the first layer of verification.
• Figure 3: A zoom in on the schematic architecture of the FERMIACC analysis generator module.
• Figure 4: Background prediction and observed counts from the CMS diphoton search CMS:2015dxe in the EBEB (top) and EBEE (bottom) channels overlaid with the simulated best-fit signal counts from the FERMIACC's proposal 544c3421 for a pseudoscalar coupled to heavy vector-like quarks.
• Figure 5: Background prediction and observed counts from the ATLAS diphoton search TheATLAScollaboration:2015mdt overlaid with the simulated best-fit signal counts from the FERMIACC's proposal 5e41fd9e for a hypercharge axion.