Constraining the parameters of high-dimensional models with active learning

Abstract

Constraining the parameters of physical models with > 5−10 parameters is a widespread problem in fields like particle physics and astronomy. The generation of data to explore this parameter space often requires large amounts of computational resources. A reduction of the relevant physical parameters hampers the generality of the results. In this paper we show that this problem can be alleviated by the use of active learning. We illustrate this with examples from high energy physics, a field where computationally expensive simulations and large parameter spaces are common. We show that the active learning techniques query-by-committee and query-by-dropout-committee allow for the identification of model points in interesting regions of high-dimensional parameter spaces (e.g. around decision boundaries). This makes it possible to constrain model parameters more efficiently than is currently done with the most common sampling algorithms. Code implementing active learning can be found on GitHub .

Les Houches 2019 Physics at TeV Colliders: New Physics Working Group Report

“…Techniques like active learning can predominantly sample parameter regions that are difficult to learn for the ML algorithm and can help to mitigate these problems e.g. to learn exclusion boundaries of HEP models [357].…”

Active learning BSM parameter spaces

“…Since a neural network used as a predictor does not have a natural definition for this uncertainty, this is generally applied to approaches such as random forest classifiers, where the decision is made based on the results of a set of models, and then the uncertainty can be related to the differing predictions within the set. This has recently been applied in the HEP context [30,31].…”

Mapping out the thermodynamic stability of a QCD equation of state with a critical point using active learning

“…The problem of determining the acceptable parameter space range for a high-dimensional model using active learning has been shown to work effectively in Ref. [27].…”

“…Previous wellknown examples are an attempt to identify new signatures of the QCD phase transition [31,34] or classify jets origination from quarks vs. gluons [38,39]. We are also not aware of previous investigations in the context of heavy-ion collisions that have used active learning (though it has been employed in other contexts in nuclear theory [40] and highenergy physics [27,41,42]). For a recent review of artificial intelligence and machine learning applications in nuclear physics, see Ref.…”