Miguel Crispim Romão’s research while affiliated with MIT Portugal and other places

The search for physics beyond the Standard Model (BSM) at collider experiments requires model-independent strategies to avoid missing possible discoveries of unexpected signals. Anomaly detection (AD) techniques offer a promising approach by identifying deviations from the Standard Model (SM) and have been extensively studied. The sensitivity of these methods to untunable hyperparameters has not been systematically compared, however. This study addresses it by investigating four semi-supervised AD methods -- Auto-Encoders, Deep Support Vector Data Description, Histogram-based Outlier Score, and Isolation Forest -- trained on simulated SM background events. In this paper, we study the sensitivity of these methods to BSM benchmark signals as a function of these untunable hyperparameters. Such a study is complemented by a proposal of a non-parametric permutation test using signal-agnostic statistics, which can provide a robust statistical assessment.

With the Large Hadron Collider's Run 3 in progress, the 125 GeV Higgs boson couplings are being examined in greater detail, while searching for additional scalars. Multi-Higgs frameworks allow Higgs couplings to significantly deviate from Standard Model values, enabling indirect probes of extra scalars. We consider the possibility of large pseudoscalar Yukawa couplings in the softly-broken Z2xZ2' three-Higgs doublet model with CP violating coefficients. To explore the parameter space of the model, we employ a Machine Learning algorithm that significantly enhances sampling efficiency. Using it, we find new regions of parameter space and observable consequences, not found with previous techniques. This method leverages an Evolutionary Strategy to quickly converge towards valid regions with an additional Novelty Reward mechanism. We use this model as a prototype to illustrate the potential of the new techniques, applicable to any Physics Beyond the Standard Model scenario.

We present a novel artificial intelligence approach to explore beyond Standard Model parameter spaces by leveraging a multi-objective optimisation algorithm. We apply this methodology to a non-minimal scotogenic model which is constrained by Higgs mass, anomalous magnetic moment of the muon, dark matter relic density, dark matter direct detection, neutrino masses and mixing, and lepton flavour violating processes. Our results successfully expand on the phenomenological realisations presented in previous work. We compare between multi- and single-objective algorithms and we observe more phenomenologically diverse solutions and an improved search capacity coming from the former. We use novelty detection to further explore sparsely populated regions of phenomenological interest. These results suggest a powerful search strategy that combines the global exploration of multi-objective optimisation with the exploitation of single-objective optimisation.

We introduce a novel approach to detecting microlensing events and other transients in light curves, utilising the isolation forest (iForest) algorithm for anomaly detection. Focusing on the Legacy Survey of Space and Time by the Vera C. Rubin Observatory, we show that an iForest trained on signal-less light curves can efficiently identify microlensing events by different types of dark objects and binaries, as well as variable stars. We further show that the iForest has real-time applicability through a drip-feed analysis, demonstrating its potential as a valuable tool for LSST alert brokers to efficiently prioritise and classify transient candidates for follow-up observations.

We propose symbolic regression as a powerful tool for the numerical studies of proposed models of physics beyond the Standard Model. In this paper we demonstrate the efficacy of the method on a benchmark model, namely the constrained minimal supersymmetric Standard Model, which has a four-dimensional parameter space. We provide a set of analytical expressions that reproduce three low-energy observables of interest in terms of the parameters of the theory: the Higgs mass, the contribution to the anomalous magnetic moment of the muon, and the cold dark matter relic density. To demonstrate the power of the approach, we employ the symbolic expressions in a global fits analysis to derive the posterior probability densities of the parameters, which are obtained two orders of magnitude more rapidly than is possible using conventional methods. Published by the American Physical Society 2025

We discuss modular family symmetry in effective theories based on generic properties of bottom-up local F-theory inspired grand unified theories (GUTs) broken by fluxes, which we refer to as fluxed GUTs. We argue that the Yukawa couplings will depend on the complex structure moduli of the matter curves in such a way that they can be modular forms associated with these symmetries. To illustrate the approach, we analyze in detail a concrete local fluxed S U ( 5 ) GUT with modular S 4 family symmetry. Published by the American Physical Society 2025

Finite modular family symmetry can emerge from top-down approaches based on heterotic string theory or Type IIB string theory. We show that, in addition to such approaches, it can also emerge from local F-Theory bottom-up constructions. As a first example of the new approach, we have analysed in detail a concrete F-Theory Fluxed SU(5) Grand Unified Theory (GUT) with modular $S_4$ family symmetry. The model fits the fermion mass and mixing data very well and serves as a demonstration of the bottom-up approach to modular family symmetry in F-Theory fluxed GUTs.

This paper presents a machine learning-based method for the detection of the unique gravitational microlensing signatures of extended dark objects, such as boson stars, axion miniclusters and subhalos. We adapt MicroLIA, a machine learning-based package tailored to handle the challenges posed by low-cadence data in microlensing surveys. Using realistic observational time stamps, our models are trained on simulated light curves to distinguish between microlensing by pointlike and extended lenses, as well as from other object classes which give a variable magnitude. We focus on boson stars and Navarro-Frenk-White (NFW) subhalos and show that the former, which are examples of objects with a relatively flat mass distribution, can be confidently identified for 0.8 ≲ r / r E ≲ 3 . Intriguingly, we also find that more sharply peaked structures, such as NFW subhalos, can be distinctly recognized from point lenses under regular observation cadence. Our findings significantly advance the potential of microlensing data in uncovering the elusive nature of extended dark objects. The code and dataset used are also provided. Published by the American Physical Society 2024

We propose symbolic regression as a powerful tool for studying Beyond the Standard Model physics. As a benchmark model, we consider the so-called Constrained Minimal Supersymmetric Standard Model, which has a four-dimensional parameter space defined at the GUT scale. We provide a set of analytical expressions that reproduce three low-energy observables of interest in terms of the parameters of the theory: the Higgs mass, the contribution to the anomalous magnetic moment of the muon, and the cold dark matter relic density. To demonstrate the power of the approach, we employ the symbolic expressions in a global fits analysis to derive the posterior probability densities of the parameters, which are obtained extremely rapidly in comparison with conventional methods.

We present a novel artificial intelligence approach for beyond the Standard Model parameter space scans by augmenting an evolutionary strategy with novelty detection. Our approach leverages the power of evolutionary strategies, previously shown to quickly converge to the valid regions of the parameter space, with a novelty reward to continue exploration once converged. Taking the Z 3 3HDM as our physics case, we show how our methodology allows us to quickly explore highly constrained multidimensional parameter spaces, providing up to eight orders of magnitude higher sampling efficiency when compared with pure random sampling and up to four orders of magnitude when compared to random sampling around the alignment limit. In turn, this enables us to explore regions of the parameter space that have been hitherto overlooked, leading to the possibility of novel phenomenological realizations of the Z 3 three Higgs doublet model that had not been considered before. Published by the American Physical Society 2024