Speaker
Description
Exploring parameter spaces in Beyond Standard Model (BSM) scenarios, especially in high-dimensional cases, is computationally prohibitive and inefficient with conventional sampling due to the curse of dimensionality. In this study, we implement a Machine Learning (ML)-assisted Nested Sampling (NS) approach to estimate the posterior distribution of the Type-II Seesaw Model. We use a generative framework, namely, Real-valued Non-Volume Preserving (Real NVP) normalizing flows as our ML framework. We use the predictions of such a simulator to guide the iterations of the NS, with much fewer likelihood evaluations, while another pre-trained classifier effectively selects valid points of the parameter space. The predicted points with both correct and incorrect predictions are saved with the actual observable/likelihood values and are used to periodically re-train the simulator, thereby refining sampling accuracy. This approach achieves convergence with a tolerance of ∼0.001 in a matter of days, significantly accelerating convergence relative to traditional sampling methods, which require several weeks.
| Field of contribution | Phenomenology |
|---|
