Searcharxiv⌕ Search

arXiv subjects

Maxim Laletin

Publications and source records attributed to Maxim Laletin.

15 recordsLinked to original sources
arXiv2026

Regularizing Attention Scores with Bootstrapping

Vision transformers (ViT) rely on attention mechanism to weigh input features, and therefore attention scores have naturally been considered as explanations for its decision-making process. However, attention scores are almost always non-zero, resulting in noisy and diffused attention maps and limiting interpretability. Can we quantify uncertainty measures of attention scores and obtain regularized attention scores? To this end, we consider attention scores of ViT in a statistical framework where independent noise would lead to insignificant yet non-zero scores. Leveraging statistical learning techniques, we introduce the bootstrapping for attention scores which generates a baseline distribution of attention scores by resampling input features. Such a bootstrap distribution is then used to estimate significances and posterior probabilities of attention scores. In natural and medical images, the proposed \emph{Attention Regularization} approach demonstrates a straightforward removal of spurious attention arising from noise, drastically improving shrinkage and sparsity. Quantitative evaluations are conducted using both simulation and real-world datasets. Our study highlights bootstrapping as a practical regularization tool when using attention scores as explanations for ViT. Code available: https://github.com/ncchung/AttentionRegularization

cs.CV↗
arXiv2025

Improved cosmological constraints on axion-lepton interactions

We present updated cosmological constraints on axion-lepton interactions based on state-of-the-art computations of the thermal axion abundance. By combining Planck Cosmic Microwave Background (CMB) data with baryon acoustic oscillation (BAO) measurements from DESI DR2, we derive improved limits on both lepton-flavor-conserving (LFC) and lepton-flavor-violating (LFV) axion couplings. Incorporating finite axion mass effects substantially strengthens the bounds for axion masses above 0.1 eV compared to those inferred from the $\Delta N_{\rm eff}$ constraint alone. The bounds on the LFC axion-tau coupling and LFV axion couplings to tau and muon or electron are improved by several orders of magnitude and the lower bound on the axion decay constant may exceed $10^6$ and $10^8$ GeV, respectively, for axion masses above 1 eV. Our cosmological constraints on LFC axion couplings to muons and taus and LFV axion couplings to tau and muon or electron are stronger than all other constraints for masses above 0.3 eV. In particular, they are stronger than recent collider constraints from Belle-II on $\tau \rightarrow la$ decays, where $l=e$ or $\mu$. The collider constraints on $\mu \rightarrow ea$ decays are weaker than the cosmological constraints for axion masses above 100 eV. Our results are relevant for both the QCD axion and axion-like particles (ALPs).

hep-ph↗
arXiv2025

Precise predictions for the QCD axion contribution to dark radiation with full phase-space evolution

We compute the QCD axion contribution to the energy density of dark radiation, parameterized by $ΔN_{\rm eff}$, by solving Boltzmann equations for the momentum distribution functions including the effects of quantum statistics for all particles involved in the axion production processes. This approach gives precise prediction for $ΔN_{\rm eff}$ independently on whether axions are produced via freeze-out or freeze-in. We focus on axions produced via flavor-conserving and flavor-violating interactions with leptons. Our precise predictions for $ΔN_{\rm eff}$ can differ from those assuming thermal shape for the momentum distribution functions, as commonly done in the literature, by more than the experimental precision of future Cosmic Microwave Background (CMB) observations. Current lower limits on the axion couplings from Planck constraints on $ΔN_{\rm eff}$ are also affected by our precise computation which, in particular, results in a strongly relaxed bound on flavor-violating axion couplings to tau lepton and muon or electron.

hep-ph↗
arXiv2024

Phase Transitions and Gravitational Waves in a Model of $\mathbb{Z}_{3}$ Scalar Dark Matter

Theories with more than one scalar field often exhibit phase transitions producing potentially detectable gravitational wave (GW) signal. In this work we study the semi-annihilating $\mathbb{Z}_3$ dark matter model, whose dark sector comprises an inert doublet and a complex singlet, and assess its prospects in future GW detectors. Without imposing limits from requirement of providing a viable dark matter candidate, i.e. taking into account only other experimental and theoretical constraints, we find that the first order phase transition in this model can be strong enough to lead to a detectable signal. However, direct detection and the dark matter thermal relic density constraint calculated with the state-of-the-art method including the impact of early kinetic decoupling, very strongly limit the parameter space of the model explaining all of dark matter and providing observable GW peak amplitude. Extending the analysis to underabundant dark matter thus reveals region with detectable GWs from a single-step or multi-step phase transition.

hep-ph↗
arXiv2022

Impact of dark matter self-scattering on its relic abundance

Elastic self-scatterings do not change the number of dark matter particles and as such have been neglected in the calculation of its relic abundance. In this work we highlight the scenarios where the presence of self-scatterings has a significant impact on the effectiveness of annihilation processes through the modification of dark matter momentum distribution. We study a few example freeze-out scenarios involving resonant and sub-threshold annihilations, as well as a model with an additional source of dark matter particles from the decays of a heavier mediator state. Interestingly, when the calculation is performed at the level of dark matter momentum distribution function, we find that the injection of additional energetic dark matter particles onto the thermal population can lead to a $\textit{decrease}$ of its final relic abundance.

hep-ph↗
arXiv2021

Dark matter freeze-in from semi-production

We study a novel dark matter production mechanism based on the freeze-in through semi-production, i.e. the inverse semi-annihilation processes. A peculiar feature of this scenario is that the production rate is suppressed by a small initial abundance of dark matter and consequently creating the observed abundance requires much larger coupling values than for the usual freeze-in. We provide a concrete example model exhibiting such production mechanism and study it in detail, extending the standard formalism to include the evolution of dark matter temperature alongside its number density and discuss the importance of this improved treatment. Finally, we confront the relic density constraint with the limits and prospects for the dark matter indirect detection searches. We show that, even if it was never in full thermal equilibrium in the early Universe, dark matter could, nevertheless, have strong enough present-day annihilation cross section to lead to observable signals.

hep-ph↗
arXiv2019

Strongly interacting dark matter and the DAMA signal

We show that models of strongly interacting (SIMP) dark matter built to reproduce the DAMA signal actually cannot account for its time dependence. We discuss the constraints on this type of models coming from direct detection experiments and study the propagation of thermalised dark matter particles in the ground for the allowed values of the parameters. We consider a simple 1D diffusion and a more detailed 3D diffusion. In both cases the predicted signal has either the wrong phase of the annual modulation or a much larger amplitude of the diurnal modulation.

hep-ph↗
arXiv2019

The DAMPE excess and gamma-ray constraints

The direct measurements of the cosmic electron-positron spectrum around 1 TeV made by DAMPE have induced many theoretical speculations about possible excesses in the data above the standard astrophysical predictions that might have the dark matter (DM) origin. These attempts mainly fall into two categories: i) DM annihilation (or decay) in the Galactic halo producing the broad spectrum excess; ii) DM annihilation in the nearby compact subhalo producing the sharp peak at 1.4 TeV. We investigate the gamma-ray emission accompanying $e^+e^-$ production in DM annihilation, as well as various theoretical means to suppress the prompt radiation, such as specific interaction vertices or multi-cascade modes, and conclude that these attempts are in tension with various gamma-ray observations. We show that the DM explanations of the broad spectrum excess tend to contradict the diffuse isotropic gamma-ray background (IGRB), measured by Fermi-LAT, while the nearby subhalo scenario is constrained by nonobservation in the surveys, performed by Fermi-LAT, MAGIC and HESS. We also briefly review other types of gamma-ray constraints, which seem to rule out the DM interpretations of the DAMPE broad spectrum excess as well.

astro-ph.HE↗
arXiv2019

Dark matter from CP symmetry of order 4: evolution in the asymmetric regime

Multi-Higgs models equipped with global symmetries produce scalar dark matter (DM) candidates stabilized by the unbroken symmetry. It is remarkable that a conserved CP symmetry can also stabilize DM candidates, provided it is a CP symmetry of order higher than two. CP4 3HDM, the three-Higgs-doublet model with CP symmetry of order 4, is the simplest example of this kind. It contains two mass-degenerate scalar DM candidates $φ$ and $\barφ$, each of them being a CP4 eigenstate and, therefore, its own antiparticle. A novel phenomenological feature of this model is the presence of $φφ\leftrightarrow \barφ\barφ$ conversion process, which conserves CP. It offers a rare example of DM models in which self-interaction in the dark sector can significantly affect cosmological and astrophysical observables. Here, we explore the thermal evolution of these DM species in the asymmetric regime. We assume that a mechanism external to CP4 3HDM produces an initial imbalance of the densities of $φ$ and $\barφ$. As the Universe cools down, we track the evolution of the asymmetry through different stages, and determine how the final asymmetry depends on the interplay between the conversion and annihilation $φ\barφ\to $ SM and on the initial conditions. We begin with the analytic treatment of Boltzmann equations, present a detailed qualitative description of the process, and then corroborate it with numerical results obtained using a dedicated computer code. Finally, we check if the model can produce an observable indirect detection signal.

hep-ph↗
arXiv2018

Multi-Higgs models with CP symmetries of increasingly high order

When building CP-symmetric models beyond the Standard Model, one can impose CP-symmetry of higher order. This means that one needs to apply the CP-transformation more than two times to get the identity transformation, but still the model is perfectly CP-conserving. A multi-Higgs-doublet model based on CP-symmetry of order 4, dubbed CP4, was recently proposed and its phenomenology is being explored. Here, we show that the construction does not stop at CP4. We build examples of renormalizable multi-Higgs-doublet potentials which are symmetric under CP8 or CP16, without leading to any accidental symmetry. If the vacuum conserves CP-symmetry of order 2k, then the neutral scalars become CP-eigenstates, which are characterized not by CP-parities but by CP-charges defined modulo 2k. One or more lightest states can be the dark matter candidates, which are protected against decay not by the internal symmetry but by the exotic CP. We briefly discuss their mass spectra and interaction patterns for CP8 and CP16.

hep-ph↗
arXiv2017

Fermi-LAT kills dark matter interpretations of AMS-02 data. Or not?

A number of papers attempt to explain the positron anomaly in cosmic rays, observed by PAMELA and AMS-02, in terms of dark matter (DM) decays or annihilations. However, the recent progress in cosmic gamma-ray studies challenges these attempts. Indeed, as we show, any rational DM model explaining the positron anomaly abundantly produces final state radiation and Inverse Compton gamma rays, which inevitably leads to a contradiction with Fermi-LAT isotropic diffuse gamma-ray background measurements. Furthermore, the Fermi-LAT observation of Milky Way dwarf satellites, supposed to be rich in DM, revealed no significant signal in gamma rays. We propose a generic approach in which the major contribution to cosmic rays comes from the dark matter disc and prove that the tension between the DM origin of the positron anomaly and the cosmic gamma-ray observations can be relieved. We consider both a simple model, in which DM decay/annihilate into charged leptons, and a model-independent minimal case of particle production, and we estimate the optimal thickness of DM disk. Possible mechanisms of formation and its properties are briefly discussed.

astro-ph.HE↗
arXiv2016

A no-go theorem for the dark matter interpretation of the positron anomaly

The overabundance of high-energy cosmic positrons, observed by PAMELA and AMS-02, can be considered as the consequence of dark matter decays or annihilations. We show that recent FERMI/LAT measurements of the isotropic diffuse gamma-ray background impose severe constraints on dark matter explanations and make them practically inconsistent.

astro-ph.HE↗
arXiv2014

Dark Atoms and their decaying constituents

The nonbaryonic dark matter of the Universe might consist of new stable charged species, bound by ordinary Coulomb interactions in various forms of heavy neutral "dark atoms". The existing models offer natural implementations for the dominant and subdominant forms of dark atom components. In the framework of Walking Technicolor the charge asymmetric excess of both stable negatively doubly charged technilepton $ζ^{--}$ and metastable but longliving positively doubly charged technibaryon $UU^{++}$ can be generated in the early Universe together with the observed baryon asymmetry. If the excess of $ζ$ exceeds by several orders of magnitude the excess of $UU$, dark matter might consist dominantly by $Heζ$ dark atoms of nuclear interacting O-helium ($OHe$) bound state of $ζ$ with primordial helium. Although sparse, this subdominant component can lead to observable effects, since leptonic decays of technibaryons $UU$ give rise to two positively charged leptons contrary to the pairs of opposite charge leptons created in decays of neutral particles. We show that decays of $UU^{++}\rightarrow e^+ e^+, μ^+ μ^+, τ^+ τ^+$ of the subdominant $UUζ$ component of dark matter, can explain the observed high energy positron excess in the cosmic rays if the fraction of $UUζ$ is $\sim 10^{-6}$ of the total dark matter density, the mass of $UU^{++}$ about 1 TeV and the lifetime about $10^{20} $s. Optimizing fit of recent AMS-02 data by model parameters, the predicted mass range of such long-living double charge particle is challenging for its search at the LHC. (abridged)

hep-ph↗
arXiv2014

Decaying Dark Atom constituents and cosmic positron excess

We present a scenario where dark matter is in the form of dark atoms that can accomodate the experimentally observed excess of positrons in PAMELA and AMS-02 while being compatible with the constraints imposed on the gamma-ray flux from Fermi/LAT. This scenario assumes that the dominant component of dark matter is in the form of a bound state between a helium nucleus and a $-2$ particle and a small component is in the form of a WIMP-like dark atom compatible with direct searches in underground detectors. One of the constituents of this WIMP-like state is a $+2$ metastable particle with a mass of 1 TeV or slightly below that by decaying to $e^+e^+$, $μ^+ μ^+$ and $τ^+ τ^+$ produces the observed positron excess. These decays can naturally take place via GUT interactions. If it exists, such a metastable particle can be found in the next run of LHC. The model predicts also the ratio of leptons over baryons in the Universe to be close to -3.

astro-ph.CO↗