arXiv subjects

Alexander Segner

Publications and source records attributed to Alexander Segner.

6 recordsLinked to original sources

arXiv2025

Machine-learning techniques as noise reduction strategies in lattice calculations of the muon $g-2$

Lattice calculations of the hadronic contributions to the muon anomalous magnetic moment are numerically highly demanding due to the necessity of reaching total errors at the sub-percent level. Noise-reduction techniques such as low-mode averaging have been applied successfully to determine the vector-vector correlator with high statistical precision in the long-distance regime, but display an unfavourable scaling in terms of numerical cost. This is particularly true for the mixed contribution in which one of the two quark propagators is described in terms of low modes. Here we report on an ongoing project aimed at investigating the potential of machine learning as a cost-effective tool to produce approximate estimates of the mixed contribution, which are then bias-corrected to produce an exact result. A second example concerns the determination of electromagnetic isospin-breaking corrections by combining the predictions from a trained model with a bias correction.

hep-lat↗

arXiv2022

Fair Interpretable Representation Learning with Correction Vectors

Neural network architectures have been extensively employed in the fair representation learning setting, where the objective is to learn a new representation for a given vector which is independent of sensitive information. Various representation debiasing techniques have been proposed in the literature. However, as neural networks are inherently opaque, these methods are hard to comprehend, which limits their usefulness. We propose a new framework for fair representation learning that is centered around the learning of "correction vectors", which have the same dimensionality as the given data vectors. Correction vectors may be computed either explicitly via architectural constraints or implicitly by training an invertible model based on Normalizing Flows. We show experimentally that several fair representation learning models constrained in such a way do not exhibit losses in ranking or classification performance. Furthermore, we demonstrate that state-of-the-art results can be achieved by the invertible model. Finally, we discuss the law standing of our methodology in light of recent legislation in the European Union.

cs.LG↗

arXiv2022

Fair Group-Shared Representations with Normalizing Flows

The issue of fairness in machine learning stems from the fact that historical data often displays biases against specific groups of people which have been underprivileged in the recent past, or still are. In this context, one of the possible approaches is to employ fair representation learning algorithms which are able to remove biases from data, making groups statistically indistinguishable. In this paper, we instead develop a fair representation learning algorithm which is able to map individuals belonging to different groups in a single group. This is made possible by training a pair of Normalizing Flow models and constraining them to not remove information about the ground truth by training a ranking or classification model on top of them. The overall, ``chained'' model is invertible and has a tractable Jacobian, which allows to relate together the probability densities for different groups and ``translate'' individuals from one group to another. We show experimentally that our methodology is competitive with other fair representation learning algorithms. Furthermore, our algorithm achieves stronger invariance w.r.t. the sensitive attribute.

cs.LG↗

arXiv2022

Fair Interpretable Learning via Correction Vectors

Neural network architectures have been extensively employed in the fair representation learning setting, where the objective is to learn a new representation for a given vector which is independent of sensitive information. Various "representation debiasing" techniques have been proposed in the literature. However, as neural networks are inherently opaque, these methods are hard to comprehend, which limits their usefulness. We propose a new framework for fair representation learning which is centered around the learning of "correction vectors", which have the same dimensionality as the given data vectors. The corrections are then simply summed up to the original features, and can therefore be analyzed as an explicit penalty or bonus to each feature. We show experimentally that a fair representation learning problem constrained in such a way does not impact performance.

cs.LG↗

arXiv2019

Shining Light on the Scotogenic Model: Interplay of Colliders and Cosmology

In the framework of the scotogenic model, which features radiative generation of neutrino masses, we explore light dark matter scenario. Throughout the paper we chiefly focus on keV-scale dark matter which can be produced either via freeze-in through the decays of the new scalars, or from the decays of next-to-lightest fermionic particle in the spectrum, which is produced through freeze-out. The latter mechanism is required to be suppressed as it typically produces a hot dark matter component. Constraints from BBN are also considered and in combination with the former production mechanism they impose the dark matter to be light. For this scenario we consider signatures at High Luminosity LHC and proposed future hadron and lepton colliders, namely FCC-hh and CLIC, focusing on searches with two leptons and missing energy as a final state. While a potential discovery at High Luminosity LHC is in tension with limits from cosmology, the situation greatly improves for future colliders.

hep-ph↗

arXiv2019

Pairwise Learning to Rank by Neural Networks Revisited: Reconstruction, Theoretical Analysis and Practical Performance

We present a pairwise learning to rank approach based on a neural net, called DirectRanker, that generalizes the RankNet architecture. We show mathematically that our model is reflexive, antisymmetric, and transitive allowing for simplified training and improved performance. Experimental results on the LETOR MSLR-WEB10K, MQ2007 and MQ2008 datasets show that our model outperforms numerous state-of-the-art methods, while being inherently simpler in structure and using a pairwise approach only.

cs.IR↗