Anomaly Detection for Resonant New Physics with Machine Learning (1805.02664v3)

Abstract: Despite extensive theoretical motivation for physics beyond the Standard Model (BSM) of particle physics, searches at the Large Hadron Collider (LHC) have found no significant evidence for BSM physics. Therefore, it is essential to broaden the sensitivity of the search program to include unexpected scenarios. We present a new model-agnostic anomaly detection technique that naturally benefits from modern machine learning algorithms. The only requirement on the signal for this new procedure is that it is localized in at least one known direction in phase space. Any other directions of phase space that are uncorrelated with the localized one can be used to search for unexpected features. This new method is applied to the dijet resonance search to show that it can turn a modest 2 sigma excess into a 7 sigma excess for a model with an intermediate BSM particle that is not currently targeted by a dedicated search.

• The paper introduces a model-agnostic anomaly detection method leveraging machine learning to enhance the search for resonant new physics beyond the Standard Model at the LHC.
• It employs a Classification Without Labels (CWoLa) technique, training classifiers on data using auxiliary variables to identify signal-like characteristics in resonant distributions.
• Numerical results show the method significantly boosts detection sensitivity, turning a simulated 2dyf excess into a 7dyf discovery-level signal in an all-hadronic resonance search example.

Anomaly Detection for Resonant New Physics with Machine Learning

The paper "Anomaly Detection for Resonant New Physics with Machine Learning" presents an innovative approach to broadening the sensitivity of new physics searches beyond the established frameworks of the Standard Model (SM). Post the Higgs boson's discovery in 2012, the Large Hadron Collider (LHC) has rigorously pursued paths beyond the SM, yet the direct evidence for new physics remains elusive. This paper introduces a model-agnostic anomaly detection method benefiting from contemporary machine learning techniques, specifically designed to address potential gaps in the current search methodologies at the LHC.

Methodological Advancements

The anomaly detection approach delineated in the paper fundamentally extends and enhances traditional bump hunts by training machine learning classifiers directly on data. The pivotal requirement for this technique is that the new physics signals are localized in at least one known direction in the phase space, allowing the classifiers to utilize other uncorrelated dimensions to identify anomalous features indicative of new physics. Employing a Classification Without Labels (CWoLa) method, the algorithm seeks to classify events based on auxiliary variables that discern signal-like characteristics amidst known resonant variables such as invariant mass distributions.

Key Results and Numerical Evidence

A striking demonstration of this method's efficacy is provided in the context of a simulated all-hadronic resonance search at the LHC, targeting a scenario involving a new $W'$ particle decaying into jets. The paper reports a significant enhancement of resonance signals, where a $2\sigma$ excess turns into a notable $7\sigma$ discovery level following the application of the model-agnostic classifier. Fig.~\ref{fig:pvalues} in the paper visually underscores the marked increase in detection significance, facilitating the localization of new physical phenomena previously unexplored due to conventional methods’ limitations.

Implications

Practically, the introduction of such an anomaly detection technique stands to significantly impact the LHC's ongoing search for Beyond the Standard Model (BSM) physics. The method's model independence and ability to leverage machine learning directly translate into a robust mechanism adaptable to various potential new physics scenarios. Theoretically, this approach challenges existing borders within high energy physics by proposing a technique free from strict dependency on simulations or predefined signal characteristics. The adaptability and overfitting resistance makes it promising for unveiling novel structures in nature.

Future Developments

The implications of this research extend into shaping future developments in AI applications within the field of high energy physics. The paper's technique aligns with growing trends to employ unsupervised and semi-supervised learning models capable of detecting subtle anomalies without the need for extensive pre-labeling or simulation reliance. This are poised to play a crucial role in expanding anomaly search methodologies in forthcoming collider experiments and exploring deeper into the fabric of fundamental physics.

In conclusion, this paper introduces and validates a robust anomaly detection strategy advancing the sensitivity of new physics searches, leveraging machine learning's strengths in uncharted territories of particle physics. The innovative method holds promise in potentially unlocking yet-undiscovered facets of physics, paving the way for new explorations at the LHC and subsequent experiments.