Searching for Anomalies with Foundation Models
This paper investigates unexpected anomalies in CMS experiment data identified by the OmniLearned foundation model, finding that while background estimates perform well in validation regions, they fail to accurately model the signal region, prompting a call for further scrutiny.