Anomaly Detection for Resonant New Physics with Machine Learning.

Area of Science:

• High-energy particle physics
• Beyond Standard Model (BSM) physics
• Machine learning applications in physics

Background:

• Searches for new physics at the Large Hadron Collider (LHC) have yielded no significant evidence despite strong theoretical motivation.
• Existing search strategies may miss unexpected phenomena, necessitating broader approaches.
• The Standard Model of particle physics, while successful, is incomplete.

Purpose of the Study:

• To develop a novel, model-agnostic anomaly detection technique for particle physics searches.
• To leverage modern machine learning algorithms to enhance sensitivity to new physics signals.
• To broaden the search program's capability to detect unexpected BSM physics scenarios.

Main Methods:

• A new anomaly detection technique is introduced, requiring only that a signal is localized in at least one phase space direction.
• Machine learning algorithms are utilized to identify deviations from expected Standard Model processes.
• The method is applied to the dijet resonance search as a proof of concept.

Main Results:

• The new technique successfully identified an anomaly in the dijet resonance search.
• A modest 2σ excess was amplified to a significant 7σ excess using this method.
• This demonstrates the potential to uncover BSM particles not currently targeted by dedicated searches.

Conclusions:

• Model-agnostic anomaly detection using machine learning offers a powerful new tool for BSM physics searches.
• This approach significantly enhances the discovery potential for unexpected signals at the LHC.
• Broadening search strategies is crucial for uncovering new physics beyond the Standard Model.