Grosso Gaia, Letizia Marco
NSF AI Institute for Artificial Intelligence and Fundamental Interactions, Cambridge, MA USA.
MIT Laboratory for Nuclear Science, Cambridge, MA USA.
Eur Phys J C Part Fields. 2025;85(1):4. doi: 10.1140/epjc/s10052-024-13722-5. Epub 2025 Jan 4.
In this work, we address the question of how to enhance signal-agnostic searches by leveraging multiple testing strategies. Specifically, we consider hypothesis tests relying on machine learning, where model selection can introduce a bias towards specific families of new physics signals. Focusing on the New Physics Learning Machine, a methodology to perform a signal-agnostic likelihood-ratio test, we explore a number of approaches to multiple testing, such as combining -values and aggregating test statistics. Our findings show that it is beneficial to combine different tests, characterised by distinct choices of hyperparameters, and that performances comparable to the best available test are generally achieved, while also providing a more uniform response to various types of anomalies. This study proposes a methodology that is valid beyond machine learning approaches and could in principle be applied to a larger class model-agnostic analyses based on hypothesis testing.
在这项工作中，我们探讨了如何通过利用多重检验策略来增强与信号无关的搜索这一问题。具体而言，我们考虑依赖机器学习的假设检验，其中模型选择可能会对特定的新物理信号族产生偏差。聚焦于新物理学习机（一种执行与信号无关的似然比检验的方法），我们探索了多种多重检验方法，例如合并p值和汇总检验统计量。我们的研究结果表明，将以不同超参数选择为特征的不同检验进行组合是有益的，并且通常能够实现与最佳可用检验相当的性能，同时还能对各种类型的异常提供更一致的响应。本研究提出了一种方法，该方法不仅适用于机器学习方法，原则上还可应用于基于假设检验的更大类别的与模型无关的分析。
