Saved in:
| Main Authors: | , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2401.11806 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913203315802112 |
|---|---|
| author | César Jesús-Valls |
| author_facet | César Jesús-Valls |
| contents | Likelihood-based inference, central in modern particle physics data analysis requires the extensive evaluation of a likelihood function that depends on set of parameters defined by the statistical model under consideration. If an analytical expression for the likelihood can be defined from first principles the procedure is computationally straightforward. However, most experiments require approximating the likelihood numerically using large statistical samples of synthetic events generated using Monte Carlo methods. As a result, the likelihood consists of a comparison of the expected versus the observed event rates in a collection of histogram bins, defining binned likelihood functions. When this occurs, evaluating the likelihood function involves, on each occasion, recalculating the prediction in those bins, increasing the computational load of these analysis drastically. In this text, I highlight the importance of identifying which are the unique event configurations in the binned likelihood definition and I provide an exact formula to update the event rate predictions utilizing the minimum number of necessary calculations by means of factorization. The aim of the discussion is to decrease the computational load of widespread high-energy physics analyses, leading to substantial speed improvements and reduced carbon footprints. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2401_11806 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | On the importance of factorization for fast binned likelihood inference César Jesús-Valls High Energy Physics - Experiment Likelihood-based inference, central in modern particle physics data analysis requires the extensive evaluation of a likelihood function that depends on set of parameters defined by the statistical model under consideration. If an analytical expression for the likelihood can be defined from first principles the procedure is computationally straightforward. However, most experiments require approximating the likelihood numerically using large statistical samples of synthetic events generated using Monte Carlo methods. As a result, the likelihood consists of a comparison of the expected versus the observed event rates in a collection of histogram bins, defining binned likelihood functions. When this occurs, evaluating the likelihood function involves, on each occasion, recalculating the prediction in those bins, increasing the computational load of these analysis drastically. In this text, I highlight the importance of identifying which are the unique event configurations in the binned likelihood definition and I provide an exact formula to update the event rate predictions utilizing the minimum number of necessary calculations by means of factorization. The aim of the discussion is to decrease the computational load of widespread high-energy physics analyses, leading to substantial speed improvements and reduced carbon footprints. |
| title | On the importance of factorization for fast binned likelihood inference |
| topic | High Energy Physics - Experiment |
| url | https://arxiv.org/abs/2401.11806 |