Multitask Jet Analysis with Vision-Language Models: A Physics-Informed Four-Panel Representation

This paper investigates whether Vision-Language Models can serve as a unified interface for structured jet analysis in high-energy physics colliders. Each JetClass jet is converted into a 224x224 RGB image with four panels encoding the transverse momentum flow of all constituents. The physics-informed image representation enables VLMs to perform multi-task jet tagging across different jet types.