arXiv subjects

Himanshu Tiwari

Publications and source records attributed to Himanshu Tiwari.

8 recordsLinked to original sources

arXiv2026

Digital Twin-assisted belief-state reinforcement learning for latency-robust ISAC in 6G networks

Integrated Sensing and Communication (ISAC) enables joint data transmission and environmental perception for sixth-generation (6G) networks, but centralized and virtualized RAN control loops introduce telemetry latency that yields stale observations and unstable control. This paper proposes a Digital Twin-assisted belief-state reinforcement learning framework for latency-robust ISAC. A Digital Twin (DT) reconstructs a synchronized belief state from delayed telemetry using an Extended Kalman Filter, and a Proximal Policy Optimization agent performs joint beamforming and power allocation for communication and sensing. Closed-loop simulations with telemetry delays up to 100 ms demonstrate consistent performance gains over latency-unaware deep reinforcement learning (DRL) and heuristic baselines. At 50 ms latency, the proposed method improves median throughput by 12% and reduces sensing error by 7% relative to a DT-only controller, while achieving an order-of-magnitude reduction in reliability violations. Even at 100 ms latency, the proposed approach retains approximately 88% of its zero-latency throughput. These results show that Digital Twin-assisted belief-state control enables stable and efficient ISAC operation under realistic telemetry delays in 6G networks.

cs.NI↗

arXiv2025

From ANN to BNN: Inferring Reionization Parameters using Uncertainty-aware Emulators of 21-cm Summaries

Inferring astrophysical parameters from radio interferometric observations of the redshifted 21-cm signal from the Epoch of Reionization (EoR) is a challenging yet crucial task. The 21-cm signal from EoR is expected to be highly non-Gaussian; therefore, we need to use higher-order statistics, e.g., bispectrum. Moreover, the forward modeling of the signal and its statistics for a varying set of model parameters requires rerunning the simulations many times, which is computationally very expensive. To overcome this challenge, many artificial neural network (ANN) based emulators have been introduced, which produce the 21-cm summaries in a fraction of the time. However, ANN emulators have a drawback: they can only produce point-value predictions; thus, they fail to capture the uncertainty associated with their predictions. Therefore, when such emulators are used in the Bayesian inference pipeline, they cannot naturally propagate their prediction uncertainties to the estimated model parameters. To address this problem, we have developed Bayesian neural network (BNN) emulators for the 21-cm signal statistics, which provide the posterior distribution of the predicted signal statistics, including their prediction uncertainty. We use these BNN emulators in our Bayesian inference pipeline to infer the EoR parameters through 21-cm summaries of the mock observation of 21-cm signal with telescopic noise for $1000$ hr of SKA-LOW observation. We show that BNN emulators can capture the prediction uncertainty for the 21-cm power spectrum and bispectrum, and using these emulators in the inference pipeline provides better and tighter constraints on them. We reduced the training dataset and showed that, for smaller training datasets, BNN outperforms the ANN emulators. We also show that using the bispectrum as a summary statistic gives better constraints on EoR parameters than the power spectrum.

astro-ph.CO↗

arXiv2025

Advancing Vulnerability Classification with BERT: A Multi-Objective Learning Model

The rapid increase in cybersecurity vulnerabilities necessitates automated tools for analyzing and classifying vulnerability reports. This paper presents a novel Vulnerability Report Classifier that leverages the BERT (Bidirectional Encoder Representations from Transformers) model to perform multi-label classification of Common Vulnerabilities and Exposures (CVE) reports from the National Vulnerability Database (NVD). The classifier predicts both the severity (Low, Medium, High, Critical) and vulnerability types (e.g., Buffer Overflow, XSS) from textual descriptions. We introduce a custom training pipeline using a combined loss function-Cross-Entropy for severity and Binary Cross-Entropy with Logits for types-integrated into a Hugging Face Trainer subclass. Experiments on recent NVD data demonstrate promising results, with decreasing evaluation loss across epochs. The system is deployed via a REST API and a Streamlit UI, enabling real-time vulnerability analysis. This work contributes a scalable, open-source solution for cybersecurity practitioners to automate vulnerability triage.

cs.CR↗

arXiv2024

21cm Epoch of Reionisation Power Spectrum with Closure Phase using the Murchison Widefield Array

The radio interferometric closure phases can be a valuable tool for studying cosmological {H\scriptsize{I}}~from the early Universe. Closure phases have the advantage of being immune to element-based gains and associated calibration errors. Thus, calibration and errors therein, which are often sources of systematics limiting standard visibility-based approaches, can be avoided altogether in closure phase analysis. In this work, we present the first results of the closure phase power spectrum of {H\scriptsize{I}}~21-cm fluctuations using the Murchison Widefield Array (MWA), with $\sim 12$ hours of MWA-phase II observations centered around redshift, $z\approx 6.79$, during the Epoch of Reionisation. On analysing three redundant classes of baselines -- 14~m, 24~m, and 28~m equilateral triads, our estimates of the $2σ$ ($95\%$ confidence interval) 21-cm power spectra are $\lesssim (184)^2 pseudo \rm ~mK^2$ at ${k}_{||} = 0.36 $ $pseudo~h {\rm Mpc^{-1}}$ in the EoR1 field for the 14~m baseline triads, and $\lesssim (188)^2 pseudo \rm ~mK^2$ at $k_{||} = 0.18 $ $pseudo~h {\rm Mpc^{-1}}$ in the EoR0 field for the 24~m baseline triads. The ``$pseudo$'' units denote that the length scale and brightness temperature should be interpreted as close approximations. Our best estimates are still 3-4 orders high compared to the fiducial 21-cm power spectrum; however, our approach provides promising estimates of the power spectra even with a small amount of data. These data-limited estimates can be further improved if more datasets are included into the analysis. The evidence for excess noise has a possible origin in baseline-dependent systematics in the MWA data that will require careful baseline-based strategies to mitigate, even in standard visibility-based approaches.

astro-ph.CO↗

arXiv2023

Measuring the global 21-cm signal with the MWA-II: improved characterisation of lunar-reflected radio frequency interference

Radio interferometers can potentially detect the sky-averaged signal from the Cosmic Dawn (CD) and the Epoch of Reionisation (EoR) by studying the Moon as a thermal block to the foreground sky. The first step is to mitigate the Earth-based RFI reflections (Earthshine) from the Moon, which significantly contaminate the FM band $\approx 88-110$ MHz, crucial to CD-EoR science. We analysed MWA phase-I data from $72-180$ MHz at $40$ kHz resolution to understand the nature of Earthshine over three observing nights. We took two approaches to correct the Earthshine component from the Moon. In the first method, we mitigated the Earthshine using the flux density of the two components from the data, while in the second method, we used simulated flux density based on an FM catalogue to mitigate the Earthshine. Using these methods, we were able to recover the expected Galactic foreground temperature of the patch of sky obscured by the Moon. We performed a joint analysis of the Galactic foregrounds and the Moon's intrinsic temperature $(T_{\rm Moon})$ while assuming that the Moon has a constant thermal temperature throughout three epochs. We found $T_{\rm Moon}$ to be at $184.4\pm{2.6}\rm ~K$ and $173.8\pm{2.5}\rm ~K$ using the first and the second methods, respectively, and the best-fit values of the Galactic spectral index $(α)$ to be within the $5\%$ uncertainty level when compared with the global sky models. Compared with our previous work, these results improved constraints on the Galactic spectral index and the Moon's intrinsic temperature. We also simulated the Earthshine at MWA between November and December 2023 to find suitable observing times less affected by the Earthshine. Such observing windows act as Earthshine avoidance and can be used to perform future global CD-EoR experiments using the Moon with the MWA.

astro-ph.IM↗

arXiv2022

Modelling the energy distribution in CHIME/FRB Catalog-1

We characterize the intrinsic properties of any FRB using its redshift $z$, spectral index $α$ and energy $E_{33}$ in units of $10^{33} \, {\rm J}$ emitted across $2128 - 2848\; {\rm MHz}$ in the FRB's rest frame. Provided that $z$ is inferred from the measured extra-galactic dispersion measure $DM_{\rm Ex}$, the fluence $F$ of the observed event defines a track in $(α, E_{33})$ space which we refer to as the "energy track". Here we consider the energy tracks for a sample of $254$ non-repeating low dispersion measure FRBs from the CHIME/FRB Catalog-1, and use these to determine $n(E_{33} \mid α)$ the conditional energy distribution i.e. the number of FRBs in the interval $ΔE_{33}$ given a value of $α$. Considering $-10 \le α\le 10$, we find that the entire energy scale shifts to higher energy values as $α$ is increased. For all values of $α$, we can identify two distinct energy ranges indicating that there are possibly two distinct FRB populations. At high energies, the distribution is well fitted by a modified Schechter function whose slope and characteristic energy both increase with $α$. At low energies, the number of FRBs are in excess of the predictions of the modified Schechter function indicating that we may have a distinctly different population of low-energy FRBs. We have checked that our main findings are reasonably robust to the assumptions regarding the Galactic Halo and Host galaxy contributions to the dispersion measure.

astro-ph.HE↗

arXiv2022

Improving constraints on the reionization parameters using 21-cm bispectrum

Radio interferometric experiments aim to constrain the reionization model parameters by measuring the 21-cm signal statistics, primarily the power spectrum. However the Epoch of Reionization (EoR) 21-cm signal is highly non-Gaussian, and this non-Gaussianity encodes important information about this era. The bispectrum is the lowest order statistic able to capture this inherent non-Gaussianity. Here we are the first to demonstrate that bispectra for large and intermediate length scales and for all unique $k$-triangle shapes provide tighter constraints on the EoR parameters compared to the power spectrum or the bispectra for a limited number of shapes of $k$-triangles. We use the Bayesian inference technique to constrain EoR parameters. We have also developed an Artificial Neural Network (ANN) based emulator for the EoR 21-cm power spectrum and bispectrum which we use to remarkably speed up our parameter inference pipeline. Here we have considered the sample variance and the system noise uncertainties corresponding to $1000$ hrs of SKA-Low observations for estimating errors in the signal statistics. We find that using all unique $k$-triangle bispectra improves the constraints on parameters by a factor of $2-4$ (depending on the stage of reionization) over the constraints that are obtained using power spectrum alone.

astro-ph.CO↗

arXiv2021

A maximum likelihood estimate of the parameters of the FRB population

We consider a sample of $82$ non-repeating FRBs detected at Parkes, ASKAP, CHIME and UTMOST each of which operates over a different frequency range and has a different detection criteria. Using simulations, we perform a maximum likelihood analysis to determine the FRB population model which best fits this data. Our analysis shows that models where the pulse scatter broadening increases moderately with redshift ($z$) are preferred over those where this increases very sharply or where scattering is absent. Further, models where the comoving event rate density is constant over $z$ are preferred over those where it follows the cosmological star formation rate. Two models for the host dispersion measure ($DM_{\rm host}$) distribution (a fixed and a random $DM_{\rm host}$) are found to predict comparable results. We obtain the best fit parameter values $α=-1.53^{+0.29}_{-0.19}$, $\overline{E}_{33}=1.55^{+0.26}_{-0.22}$ and $γ=0.77\pm 0.24$. Here $α$ is the spectral index, $γ$ is the exponent of the Schechter luminosity function and $\overline{E}_{33}$ is the mean FRB energy in units of $10^{33} \, {\rm J}$ across $2128 - 2848\; {\rm MHz}$ in the FRB rest frame.

astro-ph.HE↗