Learning new physics efficiently with nonparametric methods.

Area of Science:

• High Energy Physics
• Machine Learning
• Computational Physics

Background:

• New physics searches require sophisticated analytical tools.
• Existing methods may be computationally intensive or rely on model assumptions.
• Kernel methods offer a powerful, non-parametric approach to function approximation.

Purpose of the Study:

• To develop a model-independent machine learning approach for new physics searches.
• To leverage large-scale kernel methods for hypothesis testing.
• To improve computational efficiency and performance compared to existing techniques.

Main Methods:

• Utilizing kernel methods, a class of non-parametric learning algorithms.
• Implementing a hypothesis testing procedure based on the likelihood ratio.
• Evaluating the compatibility between experimental data and a reference model without prior assumptions on new physics.

Main Results:

• The proposed machine learning approach demonstrates significant advantages in training times and computational resources.
• Performance is comparable to neural network implementations.
• The method is successfully applied to higher-dimensional datasets, advancing previous studies.

Conclusions:

• Kernel methods provide an efficient and effective tool for model-independent new physics searches.
• This approach offers a viable alternative to neural networks, particularly for complex, high-dimensional data.
• The study paves the way for more accessible and scalable new physics exploration.