Beecroft Institute for Particle Astrophysics and Cosmology

Abstract:

When choosing between competing symbolic models for a data set, a human will naturally prefer the “simpler” expression or the one which more closely resembles equations previously seen in a similar context. This suggests a non-uniform prior on functions, which is, however, rarely considered within a symbolic regression (SR) framework. In this paper we develop methods to incorporate detailed prior information on both functions and their parameters into SR. Our prior on the structure of a function is based on a ngram language model, which is sensitive to the arrangement of operators relative to one another in addition to the frequency of occurrence of each operator. We also develop a formalism based on the Fractional Bayes Factor to treat numerical parameter priors in such a way that models may be fairly compared though the Bayesian evidence, and explicitly compare Bayesian, Minimum Description Length and heuristic methods for model selection. We demonstrate the performance of our priors relative to literature standards on benchmarks and a real-world dataset from the field of cosmology.

Abstract:

<jats:p>The non-linear dynamics of scalar fields coupled to matter and gravity can lead to remarkable density-dependent screening effects. In this short review, we present the main classes of screening mechanisms, and discuss their tests in laboratory and astrophysical systems. We particularly focused on reviewing numerical and technical aspects involved in modeling the non-linear dynamics of screening and on tests using laboratory experiments and astrophysical systems, such as stars, galaxies, and dark matter halos.</jats:p>

Abstract:

The analysis of photometric large-scale structure data is often complicated by the need to account for many observational and astrophysical systematics. The elaborate models needed to describe them often introduce many "nuisance parameters’', which can be a major inhibitor of an efficient parameter inference. In this paper we introduce an approximate method to analytically marginalise over a large number of nuisance parameters based on the Laplace approximation. We discuss the mathematics of the method, its relation to concepts such as volume effects and profile likelihood, and show that it can be further simplified for calibratable systematics by linearising the dependence of the theory on the associated parameters. We quantify the accuracy of this approach by comparing it with traditional sampling methods in the context of existing data from the Dark Energy Survey, as well as futuristic Stage-IV photometric data. The linearised version of the method is able to obtain parameter constraints that are virtually equivalent to those found by exploring the full parameter space for a large number of calibratable nuisance parameters, while reducing the computation time by a factor 3-10. Furthermore, the non-linearised approach is able to analytically marginalise over a large number of parameters, returning constraints that are virtually indistinguishable from the brute-force method in most cases, accurately reproducing both the marginalised uncertainty on cosmological parameters, and the impact of volume effects associated with this marginalisation. We provide simple recipes to diagnose when the approximations made by the method fail and one should thus resort to traditional methods. The gains in sampling efficiency associated with this method enable the joint analysis of multiple surveys, typically hindered by the large number of nuisance parameters needed to describe them.