Search

Distributed Machine Learning (DML) on resource-constrained edge devices holds immense potential for real-world applications. However, achieving fast convergence in DML in these heterogeneous environments remains a significant challenge. Traditional frameworks like Bulk Synchronous Parallel and Asynchronous Stochastic Parallel rely on frequent, small updates that incur substantial communication overhead and hinder convergence speed. Furthermore, these frameworks often employ static dataset sizes, neglecting the heterogeneity of edge devices and potentially leading to straggler nodes that slow down the entire training process. The straggler nodes, i.e., edge devices that take significantly longer to process their assigned data chunk, hinder the overall training speed. To address these limitations, this paper proposes Hermes, a novel probabilistic framework for efficient DML on edge devices. This framework leverages a dynamic threshold based on recent test loss behavior to identify statistically significant improvements in the model's generalization capability, hence transmitting updates only when major improvements are detected, thereby significantly reducing communication overhead. Additionally, Hermes employs dynamic dataset allocation to optimize resource utilization and prevents performance degradation caused by straggler nodes. Our evaluations on a real-world heterogeneous resource-constrained environment demonstrate that Hermes achieves faster convergence compared to state-of-the-art methods, resulting in a remarkable $13.22$x reduction in training time and a $62.1\%$ decrease in communication overhead., Comment: 11 pages, 19 figures, 3 tables; code: https://github.com/DaSH-Lab-CSIS/Hermes

It has recently been shown in Ref. [1] that the double-inflation scenario based on the Coleman-Weinberg potential can successfully generate primordial black holes (PBHs) with the inflationary predictions consistent with the Planck measurements. These PBHs can play the role of dark matter in our universe. In this paper, we propose the classically conformal minimal $B-L$ model as an ultra-violet (UV) completion of the scenario. In our model, the $B-L$ Higgs field is identified with the inflaton and the electroweak symmetry breaking is triggered by the radiative $B-L$ symmetry breaking with the Coleman-Weinberg potential. We show that this UV completion leads to a viable cosmological history after the double inflaton: the universe is reheated via inflaton decay into right-handed neutrinos whose mass is determined consistently by a relation between the number of e-folds and reheating temperature. Using the general parameterization for neutrino Dirac Yukawa couplings through the seesaw mechanism and the neutrino oscillation data, we also show that the observed baryon asymmetry of the universe is successfully reproduced by either resonant leptogenesis or non-thermal leptogenesis. Based on the scalar power spectrum shown in Ref. [1], we evaluate the scalar induced gravitational wave spectrum, which can be tested by various proposed gravitational wave observatories like BBO, DECIGO etc., Comment: 12 page with references, 2 Figures, new references added

The minimal $U(1)_{L_\mu-L_\tau}$ gauge symmetry extended Standard Model (SM) is a well motivated framework that resolves the discrepancy between the theoretical prediction and experimental observation of muon anomalous magnetic moment. We envisage the possibility of identifying the beyond Standard Model Higgs of $U(1)_{L_\mu-L_\tau}$ sector, non-minimally coupled to gravity, as the inflaton in the early universe, while being consistent with the $(g-2)_\mu$ data. Although the structure seems to be trivial, we observe that taking into consideration of a complete cosmological history starting from inflation through the reheating phase to late-time epoch along with existing constraints on $U(1)_{L_\mu-L_\tau}$ model parameters leave us a small window of allowed reheating temperature. This further results into restriction of $(n_s-r)$ plane which is far severe than the one in a generic non-minimal quartic inflationary set up., Comment: Improved version, accepted for publication in JCAP

We explore possible signatures of the interaction between dark matter (DM) and massive neutrinos during the post-reionization epoch. Using both Fisher matrix forecast analysis and Markov Chain Monte-Carlo (MCMC) simulation, we conduct a thorough investigation of the constraints and imprints of the scenario on the upcoming post-reionization and galaxy surveys. Our investigation focuses on two key parameters: the strength of the DM-massive neutrino interaction ($u$) and the total neutrino mass ($M_{\rm tot}$), on top of the usual 6 cosmological parameters. We utilize future 21-cm intensity mapping, galaxy clustering as well as cosmic shear observations in order to investigate the possible constraints of these parameters in the future observations: Square Kilometre Array (SKA1 and SKA2) and Euclid, taking both conservative and realistic approaches. All these missions show promise in constraining both the parameters $u$ and $M_{\rm tot}$ by few orders compared to the current constraints from Planck18 (SKA2 performing the best among them). Although we do not find much improvement in $H_0$ and $\sigma_8$ tensions from our forecast analysis, SKA2 constrains them better in conservative approach. We further perform a brief investigation of the prospects of some of the next generation Cosmic Microwave Background (CMB) missions in combinations with LSS experiments in improving the constraints. Our analysis reveals that both SKA2 and CMB-S4 + Euclid + SKA1 IM2 combination will put the strongest bounds on the model parameters., Comment: 13 pages, 12 figures (Accepted in MNRAS)

We study slow roll single field inflationary scenario and the production of non-thermal fermionic dark matter, together with standard model Higgs, during reheating. For the inflationary scenario, we have considered two models of polynomial potential - one is symmetric about the origin and another one is not. We fix the coefficients of the potential from the current Cosmic Microwave Background (CMB) data from Planck/Bicep. Next, we explore the allowed parameter space on the coupling $(y_\chi)$ with inflaton and mass $(m_\chi)$ of dark matter (DM) particles $(\chi)$ produced during reheating and satisfying CMB and several other cosmological constraints., Comment: Prepared for Proceedings of XXV Bled Workshop "What comes beyond the Standard models?"

The non-thermal production of dark matter (DM) usually requires very tiny couplings of the dark sector with the visible sector and therefore is notoriously challenging to hunt in laboratory experiments. Here we propose a novel pathway to test such a production in the context of a non-standard cosmological history, using both gravitational wave (GW) and laboratory searches. We investigate the formation of DM from the decay of a scalar field that we dub as the reheaton, as it also reheats the Universe when it decays. We consider the possibility that the Universe undergoes a phase %of kination with \textit{kination-like} stiff equation-of-state ($w_{\rm kin}>1/3$) before the reheaton dominates the energy density of the Universe and eventually decays into Standard Model and DM particles. We then study how first-order tensor perturbations generated during inflation, the amplitude of which may get amplified during the kination era and lead to detectable GW signals. Demanding that the reheaton produces the observed DM relic density, we show that the reheaton's lifetime and branching fractions are dictated by the cosmological scenario. In particular, we show that it is long-lived and can be searched by various experiments such as DUNE, FASER, FASER-II, MATHUSLA, SHiP, etc. We also identify the parameter space which leads to complementary observables for GW detectors such as LISA and u-DECIGO. In particular we find that a kination-like period with an equation-of-state parameter $w_{\rm kin}\approx 0.5$ and a reheaton mass $\mathcal O(0.5-5)$ GeV and a DM mass of $\mathcal O (10-100)$ keV may lead to sizeable imprints in both kinds of searches., Comment: 9 pages, 7 figures; Fig. 6 and Fig. 7 added for LISA detector; references added; conclusions unchanged

In this article, we have done a thorough investigation of the possible effects of interaction between dark matter (DM) and neutrinos on reionization history. We have constrained the interaction strength using 21 cm Cosmology and found out possible deviations from standard, non-interacting $\Lambda$CDM scenario. Comparing the results with the existing constraints from present cosmological observations reveals that 21 cm observations are more competent to constrain the interaction strength by a few orders of magnitude. We have also searched for prospects of detecting any such interaction in the upcoming 21 cm mission SKA1-Low by doing a forecast analysis and error estimation., Comment: Version considerably modified. To appear in MNRAS

We investigate for viable models of inflation that can successfully produce dark matter (DM) from inflaton decay process, satisfying all the constraints from Cosmic Microwave Background (CMB) and from some other observations. In particular, we analyze near-inflection-point small field inflationary scenario with non-thermal production of fermionic DM from the decaying inflaton field during the reheating era. To this end, we propose two different models of inflation with polynomial potential. The potential of Model I contains terms proportional to linear, quadratic, and quartic in inflaton; whereas in Model II, the potential contains only even power of inflaton and the highest term is sextic in inflaton. For both the models, we find out possible constraints on the model parameters which lead to proper inflationary parameters from CMB data with a very small tensor-to-scalar ratio, as expected from a small-field model. With the allowed parameter space from CMB, we then search for satisfactory relic abundance for DM, that can be produced from inflaton via reheating, to match with the present-day cold dark matter (CDM) relic density for the parameter spaces of the DM $\chi$ mass and Yukawa couplings in the range $10^{-9} \gtrsim y_{\chi} \gtrsim 10^{-15}$ and $10^3 \text{GeV} \lesssim m_{\chi} \lesssim 10^9 \text{GeV}$. The DM relic is associated with the inflection-points in each model via maximum temperature reached in the early universe during its production. Finally, we find out allowed parameter space coming out of combined constraints from stability analysis for both SM Higgs and DM decays from inflaton as well as from BBN and Lyman-$\alpha$ bounds., Comment: 39 pages, 18 figures, accepted for publication in JHEP

We study radiative plateau-like inflation \& Z$_{BL}$-portal freeze-in fermionic dark matter (DM) in a minimal B-L extended model. The U(1)$_{B-L}$ Higgs, responsible for heavy neutrino masses, also drives inflation in the early universe, thanks to radiative corrections from the heavy neutrinos \& the Z$_{BL}$ gauge boson. In our benchmark choice for the U(1)$_{B-L}$ gauge coupling $g_{B-L}\sim10^{-4}$, a light Z$_{BL}$ boson can be explored by current and future lifetime frontier experiments, such as FASER and FASER 2 at the LHC, SHiP, Belle II and LHCb. For the benchmark, the Hubble scale of inflation ($\mathcal{H}_{inf}$) is very low ($\mathcal{H}_{inf} = \mathcal{O}(100)$ eV) \& the inflaton turns out to be very light with mass of $\mathcal{O}(1)$ eV, and consequently the decay width of inflaton is extremely small. We investigate a 2-field system with the inflaton/B-L Higgs and the Standard Model (SM) Higgs, and find that the reheating with a suffuciently high temperature occurs when the water-fall direction to the SM Higgs direction opens up in the trajectory of the scalar field evolution., Comment: 19 pages, 5 figures, This version is accepted in PRD

We investigate a scale-invariant $B-L$ scenario where the Standard Model (SM) is supplemented with a dark scalar $\phi$ which has gauge \& Yukawa interactions, with the couplings $g_{BL}$ and $y$, respectively, leading to radiative plateau inflation at scale $\phi=M$ in the ultraviolet (UV), while dynamically generating the Electroweak and Seesaw scales \textit{\'a l\'a} Coleman-Weinberg in the infrared (IR). This is particularly achieved by implementing threshold corrections at an energy scale $\mu_T$ arising due to the presence of vector-like fermions. We show that implementing the inflationary observables makes the couplings solely dependent on the plateau scale $M$, leaving us with only two independent parameters $M$ and $\mu_T$. Within the theoretically consistent parameter space defined by $m_{Z_{BL}} > 850~\rm GeV$, from the assumption of independent evolution of the dark sector couplings from the SM couplings and $M < 5.67~M_P$ required for the realisation of inflationary \textit{plateau-like} behaviour of the potential around $\phi=M$, where $M_P=2.4\times10^{18}$ GeV is the reduced Planck mass, we identify the parameter space that is excluded by the current LHC results from the search for the heavy $Z_{BL}$ boson. For typical benchmark points in the viable parameter regions, we estimate the reheating temperature to be $\mathcal{O}(TeV)$ thus consistent with the standard Big Bang Nucleosynthesis (BBN) constraints. For typical benchmark points ($M=5.67,~1,~0.1~M_P$) we predict the scales of inflation to be $\mathcal{H}_{inf}=2.79\times10^{12}$ GeV, $1.53\times10^{10}$ GeV and $1.53\times10^7$ GeV, respectively., Comment: 27 pages, 4 figures, Published in PRD

In $\rm\Lambda$CDM cosmology, Dark Matter (DM) and neutrinos are assumed to be non-interacting. However, it is possible to have scenarios, where DM-neutrino interaction may be present, leading to scattering of DM with neutrinos and annihilation of DM into neutrinos. We investigate the viability of such scenarios in the light of cosmological data by making use of the Planck 2018 dataset (high-l TT+TE+EE, low-l TT, low-l EE) and constrain these processes in the light of the same. We also discuss a viable particle DM model where DM-neutrino interaction is present, and map the constraints obtained to the parameter space of the model., Comment: Accepted in JCAP

With a view to managing the increasing traffic in computer networks, round robin arbiter has been proposed to work with packet switching system to have increased speed in providing access and scheduling. Round robin arbiter is a doorway to a particular bus based on request along with equal priority and gives turns to devices connected to it in a cyclic order. Considering the rapid growth in computer networking and the emergence of computer automation which will need much more access to the existing limited resources, this paper emphasizes on designing a reconfigurable round robin arbiter over FPGA which takes parallel requests and processes them with high efficiency and less delay than existing designs. Proposed round robin arbiter encounters with 4 to 12 devices. Results show that with 200% increment in the number of connected devices, only 2.69% increment has been found in the delay. With less delay, proposed round robin arbiter exhibits high speed performance with higher traffic, which is a new feature in comparison with the existing designs., Comment: Published in 2015 International Conference on Electrical, Electronics, Signals, Communication and Optimization (EESCO)

We report non-zero weak lensing convergence correlation signal of Type-Ia supernovae from 5-year Supernovae Legacy Survey data. For our analysis we utilize 296 supernovae magnification data from 5-year SNLS in the weak lensing limit. The data we use consists of measurements from four different patches, each covering 1 square degree of the sky, merged together. We demonstrate that it is possible to have a very good estimate of the two point correlation function from this data using internal error estimate technique. In order to have a good estimate of the corresponding covariance matrix we apply bootstrap spatial re-sampling technique where we reshuffle the original data consisting of 296 data points 100-10000 times and compare the results with that obtained from original data points. We show that this technique helps us arrive at a reliable conclusion on weak lensing convergence even though the original dataset comprises of a small number of data points. This also allows us to compute the corresponding covariance matrix with great accuracy., Comment: A manuscript using the same data but with a preliminary level of analysis was submitted to arXiv on February, 2018 (arxiv:1802.09014, Now withdrawn). However, we have done a thorough modification to that version with a completely new method of analysis using updated technique leading to several new findings and request the arXiv admin to consider it as a fresh submission

The existence of Dark Matter (DM) has been well established from various cosmological and astrophysical evidences. However, the particle properties of DM are largely undetermined and attempts to probe its interactions with the Standard Model (SM) particles have, so far, not met with any success. The stringent constraints on the DM-SM interactions, while does not exclude the standard lore of producing weakly massive interacting particle DM candidates through thermal freeze-out mechanism in its entirety, have certainly cast shadow on the same. In this work, we consider non-thermal production of DM within a simple extension of the SM including an inflaton field and a scalar DM candidate. Assuming negligible interactions between the SM particles and the DM, we study the production of the latter at the end of inflation, during the (p)reheating epoch. In this context, we explore the role of DM self-interactions and its interaction with the inflaton field, and find that DM can be over produced in a significant region of the parameter space. We further demonstrate that large self-interaction of the DM can suppress its abundance during preheating and to a certain extent helps to achieve the observed relic abundance via cannibalization., Comment: Matched the published version on JCAP

A number of experimental anomalies involving neutrinos hint towards the existence of at least an extra (a very light) sterile neutrino. However, such a species, appreciably mixing with the active neutrinos, is disfavored by different cosmological observations like Big Bang Nucleosynthesis (BBN), Cosmic Microwave Background (CMB) and Large Scale Structure (LSS). Recently, it was shown that the presence of additional interactions in the sterile neutrino sector via light bosonic mediators can make the scenario cosmologically viable by suppressing the production of the sterile neutrinos from active neutrinos via matter-like effect caused by the mediator. This mechanism works assuming the initial population of this sterile sector to be negligible with respect to that of the Standard Model (SM) particles, before the production from active neutrinos. However, there is fair chance that such bosonic mediators may couple to the inflaton and can be copiously produced during (p)reheating epoch. Consequently, they may ruin this assumption of initial small density of the sterile sector. In this article we, starting from inflation, investigate the production of such a sterile sector during (p)reheating in a large field inflationary scenario and identify the parameter region that allows for a viable early Universe cosmology., Comment: 16 pages; minor changes in language, footnotes and plots, references added

Why does Deep Learning work? What representations does it capture? How do higher-order representations emerge? We study these questions from the perspective of group theory, thereby opening a new approach towards a theory of Deep learning. One factor behind the recent resurgence of the subject is a key algorithmic step called {\em pretraining}: first search for a good generative model for the input samples, and repeat the process one layer at a time. We show deeper implications of this simple principle, by establishing a connection with the interplay of orbits and stabilizers of group actions. Although the neural networks themselves may not form groups, we show the existence of {\em shadow} groups whose elements serve as close approximations. Over the shadow groups, the pre-training step, originally introduced as a mechanism to better initialize a network, becomes equivalent to a search for features with minimal orbits. Intuitively, these features are in a way the {\em simplest}. Which explains why a deep learning network learns simple features first. Next, we show how the same principle, when repeated in the deeper layers, can capture higher order representations, and why representation complexity increases as the layers get deeper., Comment: 2-page version of arXiv:1412.6621 prepared for presentation at ICLR 2015 workshop as required by ICLR PC). This version has some minor formatting changes as required by the conference

Why does Deep Learning work? What representations does it capture? How do higher-order representations emerge? We study these questions from the perspective of group theory, thereby opening a new approach towards a theory of Deep learning. One factor behind the recent resurgence of the subject is a key algorithmic step called pre-training: first search for a good generative model for the input samples, and repeat the process one layer at a time. We show deeper implications of this simple principle, by establishing a connection with the interplay of orbits and stabilizers of group actions. Although the neural networks themselves may not form groups, we show the existence of {\em shadow} groups whose elements serve as close approximations. Over the shadow groups, the pre-training step, originally introduced as a mechanism to better initialize a network, becomes equivalent to a search for features with minimal orbits. Intuitively, these features are in a way the {\em simplest}. Which explains why a deep learning network learns simple features first. Next, we show how the same principle, when repeated in the deeper layers, can capture higher order representations, and why representation complexity increases as the layers get deeper., Comment: 13 pages, 5 figures