Physics-Informed Models for Squared Amplitude Calculation

This project aims to advance the SYMBA framework for calculating squared scattering amplitudes in high-energy physics by incorporating physics-informed machine learning principles. While recent transformer-based approaches have shown strong performance by framing amplitude squaring as a sequence-to-sequence problem, they largely rely on syntactic patterns and do not explicitly capture the physical structure of Feynman diagrams or the semantic roles of symbolic entities. This work proposes to bridge that gap by integrating structural and physics-aware representations into the learning process. By embedding domain knowledge into model design, this project seeks to improve accuracy, generalization, and physical consistency, particularly for complex processes such as QCD where symbolic expressions grow rapidly in complexity. The outcome will contribute towards more scalable and interpretable machine learning approaches for symbolic physics computations, while providing reproducible tools and insights that can benefit the broader ML4Sci and high-energy physics communities.