Enhancing Cosmological Model Selection with Interpretable Machine Learning.

Area of Science:

• Cosmology
• Astrophysics
• Machine Learning

Background:

• Cosmological models, such as the standard Lambda-CDM model and modified gravity theories like Hu-Sawicki f(R), describe the universe's evolution.
• Analyzing large-scale structure data, like galaxy clustering, is crucial for testing these models.
• Current methods may not fully exploit the information content in observational data.

Purpose of the Study:

• To develop and apply a novel neural network (NN) approach for differentiating between cosmological models.
• To utilize interpretability techniques (LIME) to identify key features driving the NN's decisions.
• To assess the potential of NNs in extracting cosmological information from galaxy survey data.

Main Methods:

• Implementation of neural networks for classification of cosmological models.
• Application of the LIME (Local Interpretable Model-agnostic Explanations) technique for model interpretability.
• Utilizing simulated galaxy clustering data based on current survey specifications.

Main Results:

• The developed NN successfully distinguished between the Lambda-CDM and f(R) cosmological models.
• The model achieved approximately 97% accuracy in predicting the correct cosmological model.
• LIME analysis identified crucial features in the large-scale structure data influencing the NN's classification.

Conclusions:

• Neural networks show significant potential for enhancing information extraction from cosmological large-scale structure data.
• NNs can effectively differentiate between competing cosmological models, aiding in testing fundamental physics.
• This approach can maximize the scientific return of current and future galaxy surveys for probing deviations from general relativity.