Learning new physics from an imperfect machine.

Area of Science:

• High-energy physics
• Particle physics
• Machine learning applications

Background:

• Standard Model predictions are subject to uncertainties.
• New physics searches require robust methods to handle these uncertainties.
• Artificial neural networks offer powerful tools for complex data analysis.

Purpose of the Study:

• To develop an agnostic new physics search strategy that incorporates uncertainties in Standard Model predictions.
• To leverage artificial neural networks for improved hypothesis testing in the presence of experimental uncertainties.
• To demonstrate the practical implementation and performance of the proposed method.

Main Methods:

• Utilizing the Maximum Likelihood ratio treatment of uncertainties as nuisance parameters.
• Implementing artificial neural networks for agnostic searches.
• Testing the method on a toy one-dimensional problem and a multivariate setup relevant to the LHC.

Main Results:

• The proposed method effectively handles uncertainties in Standard Model predictions within an agnostic search framework.
• Performance studies on a toy problem and LHC-relevant scenarios demonstrate the method's viability.
• The impact of experimental uncertainties in two-body final states at the LHC was analyzed.

Conclusions:

• Artificial neural networks provide a powerful framework for agnostic new physics searches that account for Standard Model uncertainties.
• The Maximum Likelihood ratio approach, when combined with neural networks, offers a robust strategy for hypothesis testing.
• This method has significant implications for future searches for new phenomena at particle colliders like the LHC.