🤖 AI Summary
The proliferation of complex, structurally diverse new physics models—coupled with experimental indistinguishability—hampers systematic model discrimination and interpretation.
Method: We propose a physics-informed unsupervised representation learning framework that integrates variational autoencoders with contrastive learning, incorporating LHC phenomenological distance constraints and model-classification consistency regularization to embed Standard Model and multiple new physics theory data into a unified latent space while preserving discriminative information.
Contribution/Results: This is the first method to yield cross-model discriminative latent representations, enabling geometric modeling of theoretical relationships, automated selection of representative benchmark scenarios, and identification of theoretical coverage gaps. Evaluated on three increasingly complex new physics scenarios, our approach achieves accurate latent-space clustering—distinguishing theoretically distinguishable models while merging experimentally indistinguishable ones—with significantly improved structural consistency over baseline methods.
📝 Abstract
We develop a machine learning method for mapping data originating from both Standard Model processes and various theories beyond the Standard Model into a unified representation (latent) space while conserving information about the relationship between the underlying theories. We apply our method to three examples of new physics at the LHC of increasing complexity, showing that models can be clustered according to their LHC phenomenology: different models are mapped to distinct regions in latent space, while indistinguishable models are mapped to the same region. This opens interesting new avenues on several fronts, such as model discrimination, selection of representative benchmark scenarios, and identifying gaps in the coverage of model space.
Published by the American Physical Society
2025