arXiv subjects

Mayank Sharma

Publications and source records attributed to Mayank Sharma.

At least 37 records · Page 2Linked to original sources

arXiv2024

Determining the absolute chemical abundance of nitrogen and sulfur in the quasar outflow of 3C298

Context. Quasar outflows are key players in the feedback processes that influence the evolution of galaxies and the intergalactic medium. The chemical abundance of these outflows provides crucial insights into their origin and impact. Aims. To determine the absolute abundances of nitrogen and sulfur and the physical conditions of the outflow seen in quasar 3C298. Methods. We analyze archival spectral data from the Hubble Space Telescope (HST) for 3C298. We measure Ionic column densities from the absorption troughs and compare the results to photoionization predictions made by the Cloudy code for three different spectral energy distributions (SED), including MF87, UVsoft, and HE0238 SEDs. We also calculate the ionic column densities of excited and ground states of N iii to estimate the electron number density and location of the outflow using the Chianti atomic database. Results. The MF87, UVsoft, and HE0238 SEDs yield nitrogen and sulfur abundances at super-solar, solar, and sub-solar values, respectively, with a spread of 0.4 to 3 times solar. Additionally, we determined an electron number density of log(ne) greater than 3.3 cm-3, with the outflow possibly extending up to a maximum distance of 2.8 kpc. Conclusions. Our results indicate solar metallicity within a 60 percent uncertainty range, driven by variations in the chosen SED and photoionization models. This study underscores the importance of SEDs impact on determining chemical abundances in quasars outflows. These findings highlight the necessity of considering a wider range of possible abundances, spanning from sub solar to super solar values.

astro-ph.GA↗

arXiv2024

Entropic pulling and diffusion diode in an Itô process

Biological environments at micrometer scales and below are often crowded, and experience incessant stochastic thermal fluctuations. The presence of membranes/pores, and multiple biological entities in a constricted space can make the damping/diffusion inhomogeneous. This effect of inhomogeneity is presented by the diffusion becoming coordinate-dependent. In this paper, we analyze the consequence of inhomogeneity-induced coordinate-dependent diffusion on Brownian systems in thermal equilibrium under the Itô's interpretation. We argue that the presence of coordinate-dependent diffusion under Itô's formulation gives rise to an effective diffusion potential (and, equivalently an entropy) that can have substantial contribution to system's transport. This emergent force when looked at as of entropic origin it provides a physical basis of the notion of the entropic pulling postulated in the context of working of some biological systems.

cond-mat.stat-mech↗

arXiv2024

AGN STORM 2: VIII. Investigating the Narrow Absorption Lines in Mrk 817 Using HST-COS Observations

We observed the Seyfert 1 galaxy Mrk817 during an intensive multi-wavelength reverberation mapping campaign for 16 months. Here, we examine the behavior of narrow UV absorption lines seen in HST/COS spectra, both during the campaign and in other epochs extending over 14 years. We conclude that while the narrow absorption outflow system (at -3750 km/s with FWHM=177 km/s) responds to the variations of the UV continuum as modified by the X-ray obscurer, its total column density (logNH =19.5 cm-2) did not change across all epochs. The adjusted ionization parameter (scaled with respect to the variations in the Hydrogen ionizing continuum flux) is log UH =-1.0. The outflow is located at a distance smaller than 38 parsecs from the central source, which implies a hydrogen density of nH > 3000 cm-3. The absorption outflow system only covers the continuum emission source and not the broad emission line region, which suggests that its transverse size is small (< 1e16 cm), with potential cloud geometries ranging from spherical to elongated along the line of sight.

astro-ph.GA↗

arXiv2024

OpenMedLM: Prompt engineering can out-perform fine-tuning in medical question-answering with open-source large language models

LLMs have become increasingly capable at accomplishing a range of specialized-tasks and can be utilized to expand equitable access to medical knowledge. Most medical LLMs have involved extensive fine-tuning, leveraging specialized medical data and significant, thus costly, amounts of computational power. Many of the top performing LLMs are proprietary and their access is limited to very few research groups. However, open-source (OS) models represent a key area of growth for medical LLMs due to significant improvements in performance and an inherent ability to provide the transparency and compliance required in healthcare. We present OpenMedLM, a prompting platform which delivers state-of-the-art (SOTA) performance for OS LLMs on medical benchmarks. We evaluated a range of OS foundation LLMs (7B-70B) on four medical benchmarks (MedQA, MedMCQA, PubMedQA, MMLU medical-subset). We employed a series of prompting strategies, including zero-shot, few-shot, chain-of-thought (random selection and kNN selection), and ensemble/self-consistency voting. We found that OpenMedLM delivers OS SOTA results on three common medical LLM benchmarks, surpassing the previous best performing OS models that leveraged computationally costly extensive fine-tuning. The model delivers a 72.6% accuracy on the MedQA benchmark, outperforming the previous SOTA by 2.4%, and achieves 81.7% accuracy on the MMLU medical-subset, establishing itself as the first OS LLM to surpass 80% accuracy on this benchmark. Our results highlight medical-specific emergent properties in OS LLMs which have not yet been documented to date elsewhere, and showcase the benefits of further leveraging prompt engineering to improve the performance of accessible LLMs for medical applications.

cs.CL↗

arXiv2024

Extreme FeLoBAL Outflow in the VLT/UVES Spectrum of Quasar SDSS J1321-0041

Context. Quasar outflows are often analyzed to determine their ability to contribute to active galactic nucleus (AGN) feedback. We identified a broad absorption line (BAL) outflow in the VLT/UVES spectrum of the quasar SDSS J1321-0041. The outflow shows troughs from Fe II, and is thus categorized as an FeLoBAL. This outflow is unusual among the population of FeLoBAL outflows, as it displays C II and Si II BALs. Aims. Outflow systems require a kinetic luminosity above $\sim0.5\%$ of the quasar's luminosity to contribute to AGN feedback. For this reason, we analyzed the spectrum of J1321-0041 to determine the outflow's kinetic luminosity, as well as the quasar's bolometric luminosity. Methods. We measured the ionic column densities from the absorption troughs in the spectrum and determined the Hydrogen column density and ionization parameter using those column densities as our constraints. We also determined the electron number density, $n_e$, based on the ratios between the excited-state and resonance-state column densities of Fe II and Si II. This allowed us to find the distance of the outflow from its central source, as well as its kinetic luminosity. Results. We determined the kinetic luminosity of the outflow to be $8.4^{+13.7}_{-5.4}\times 10^{45}\text{ erg s}^{-1}$, and the quasar's bolometric luminosity to be $1.72\pm0.13\times10^{47}\text{ erg s}^{-1}$, resulting in a ratio of $\dot{E}_k/L_{Bol}=4.8^{+8.0}_{-3.1}\%$. We conclude that this outflow has a sufficiently high kinetic luminosity to contribute to AGN feedback.

astro-ph.GA↗

arXiv2024

Quantum Inspired Chaotic Salp Swarm Optimization for Dynamic Optimization

Many real-world problems are dynamic optimization problems that are unknown beforehand. In practice, unpredictable events such as the arrival of new jobs, due date changes, and reservation cancellations, changes in parameters or constraints make the search environment dynamic. Many algorithms are designed to deal with stationary optimization problems, but these algorithms do not face dynamic optimization problems or manage them correctly. Although some optimization algorithms are proposed to deal with the changes in dynamic environments differently, there are still areas of improvement in existing algorithms due to limitations or drawbacks, especially in terms of locating and following the previously identified optima. With this in mind, we studied a variant of SSA known as QSSO, which integrates the principles of quantum computing. An attempt is made to improve the overall performance of standard SSA to deal with the dynamic environment effectively by locating and tracking the global optima for DOPs. This work is an extension of the proposed new algorithm QSSO, known as the Quantum-inspired Chaotic Salp Swarm Optimization (QCSSO) Algorithm, which details the various approaches considered while solving DOPs. A chaotic operator is employed with quantum computing to respond to change and guarantee to increase individual searchability by improving population diversity and the speed at which the algorithm converges. We experimented by evaluating QCSSO on a well-known generalized dynamic benchmark problem (GDBG) provided for CEC 2009, followed by a comparative numerical study with well-regarded algorithms. As promised, the introduced QCSSO is discovered as the rival algorithm for DOPs.

cs.NE↗

arXiv2023

Narrow absorption line Outflow in Seyfert 1 galaxy J1429+4518: Outflow's distance from the central source and its energetics

In the HST/COS spectrum of the Seyfert 1 galaxy 2MASX J14292507+4518318, we have identified a narrow absorption line (NAL) outflow system with a velocity of -151 km s$^{-1}$ This outflow exhibits absorption troughs from the resonance states of ions like CIV, NV, SiIV, and SiII, as well as excited states from CII*, and SiII*. Our investigation of the outflow involved measuring ionic column densities and conducting photoionization analysis. These yield the total column density of the outflow to be estimated as $\log N_{H}$=19.84 [cm$^{-2}]$, its ionization parameter to be $\log U_{H}$=$-$2.0 and its electron number density equal to $\log n_{e}$= 2.75[cm$^{-3}$]. These measurements enabled us to determine the mass-loss rate and the kinetic luminosity of the outflow system to be $Mdot$=0.22[$M_{Sun}$$yr^{-1}$] and $\log Edot_{K}$=39.3 [erg s$^{-1}$], respectively. We have also measured the location of the outflow system to be at $\sim$275 pc from the central source. This outflow does not contribute to the AGN feedback processes due to the low ratio of the outflow's kinetic luminosity to the AGN's Eddington luminosity ($Edot_{K}/L_{Edd}\approx 0.00025 \%$). This outflow is remarkably similar to the two bipolar lobe outflows observed in the Milky Way by XMM-Newton and Chandra.

astro-ph.GA↗

arXiv2023

Spontaneous collective transport in a heat--bath

We investigate emergence of spontaneous filtering of Brownian motion in higher dimensional space by many--body structures of symmetry broken dimer. Interacting dimer form structures which eventually restrict rotational degrees of freedom of individual constituents resulting in spontaneous emergence of collective transport. In this phenomenon, interaction and broken structural symmetry play cooperatively in emergence of collective transport out of thermal fluctuations. As a general phenomenon, spontaneous filtering of Brownian fluctuation might play an important role in the structural transition of complex molecules on top of what is known in terms of Kramers--rate process. This could also be a missing link to completely understand basics of bio--polymers' folding transitions (protein folding) where structure might encode motion in phase space.

cond-mat.stat-mech↗

arXiv2021

Learning Neural Network Classifiers with Low Model Complexity

Modern neural network architectures for large-scale learning tasks have substantially higher model complexities, which makes understanding, visualizing and training these architectures difficult. Recent contributions to deep learning techniques have focused on architectural modifications to improve parameter efficiency and performance. In this paper, we derive a continuous and differentiable error functional for a neural network that minimizes its empirical error as well as a measure of the model complexity. The latter measure is obtained by deriving a differentiable upper bound on the Vapnik-Chervonenkis (VC) dimension of the classifier layer of a class of deep networks. Using standard backpropagation, we realize a training rule that tries to minimize the error on training samples, while improving generalization by keeping the model complexity low. We demonstrate the effectiveness of our formulation (the Low Complexity Neural Network - LCNN) across several deep learning algorithms, and a variety of large benchmark datasets. We show that hidden layer neurons in the resultant networks learn features that are crisp, and in the case of image datasets, quantitatively sharper. Our proposed approach yields benefits across a wide range of architectures, in comparison to and in conjunction with methods such as Dropout and Batch Normalization, and our results strongly suggest that deep learning techniques can benefit from model complexity control methods such as the LCNN learning rule.

cs.LG↗

arXiv2020

Conversion of heat to work: An efficient inchworm

The paper discusses the natural emergence of directed motion in a dimer system due to a structural symmetry breaking. A generalised solution is obtained for the transport of such a system which is driven entirely by bath fluctuations. The result shows the existence of possibility of ratcheting driven by bath fluctuations. If this component of energy conversion driven by bath is taken into account the high efficiency of molecular motors as opposed to paradigmatic ratcheting models can probably be explained.

cond-mat.stat-mech↗

arXiv2019

Problems with automating translation of movie/TV show subtitles

We present 27 problems encountered in automating the translation of movie/TV show subtitles. We categorize each problem in one of the three categories viz. problems directly related to textual translation, problems related to subtitle creation guidelines, and problems due to adaptability of machine translation (MT) engines. We also present the findings of a translation quality evaluation experiment where we share the frequency of 16 key problems. We show that the systems working at the frontiers of Natural Language Processing do not perform well for subtitles and require some post-processing solutions for redressal of these problems

cs.CL↗

arXiv2019

Smaller Models, Better Generalization

Reducing network complexity has been a major research focus in recent years with the advent of mobile technology. Convolutional Neural Networks that perform various vision tasks without memory overhaul is the need of the hour. This paper focuses on qualitative and quantitative analysis of reducing the network complexity using an upper bound on the Vapnik-Chervonenkis dimension, pruning, and quantization. We observe a general trend in improvement of accuracies as we quantize the models. We propose a novel loss function that helps in achieving considerable sparsity at comparable accuracies to that of dense models. We compare various regularizations prevalent in the literature and show the superiority of our method in achieving sparser models that generalize well.

cs.LG↗

arXiv2019

Effect of Various Regularizers on Model Complexities of Neural Networks in Presence of Input Noise

Deep neural networks are over-parameterized, which implies that the number of parameters are much larger than the number of samples used to train the network. Even in such a regime deep architectures do not overfit. This phenomenon is an active area of research and many theories have been proposed trying to understand this peculiar observation. These include the Vapnik Chervonenkis (VC) dimension bounds and Rademacher complexity bounds which show that the capacity of the network is characterized by the norm of weights rather than the number of parameters. However, the effect of input noise on these measures for shallow and deep architectures has not been studied. In this paper, we analyze the effects of various regularization schemes on the complexity of a neural network which we characterize with the loss, $L_2$ norm of the weights, Rademacher complexities (Directly Approximately Regularizing Complexity-DARC1), VC dimension based Low Complexity Neural Network (LCNN) when subject to varying degrees of Gaussian input noise. We show that $L_2$ regularization leads to a simpler hypothesis class and better generalization followed by DARC1 regularizer, both for shallow as well as deeper architectures. Jacobian regularizer works well for shallow architectures with high level of input noises. Spectral normalization attains highest test set accuracies both for shallow and deeper architectures. We also show that Dropout alone does not perform well in presence of input noise. Finally, we show that deeper architectures are robust to input noise as opposed to their shallow counterparts.

cs.LG↗

arXiv2018

Radius-margin bounds for deep neural networks

Explaining the unreasonable effectiveness of deep learning has eluded researchers around the globe. Various authors have described multiple metrics to evaluate the capacity of deep architectures. In this paper, we allude to the radius margin bounds described for a support vector machine (SVM) with hinge loss, apply the same to the deep feed-forward architectures and derive the Vapnik-Chervonenkis (VC) bounds which are different from the earlier bounds proposed in terms of number of weights of the network. In doing so, we also relate the effectiveness of techniques like Dropout and Dropconnect in bringing down the capacity of the network. Finally, we describe the effect of maximizing the input as well as the output margin to achieve an input noise-robust deep architecture.

cs.LG↗

arXiv2018

Scalable Twin Neural Networks for Classification of Unbalanced Data

Twin Support Vector Machines (TWSVMs) have emerged an efficient alternative to Support Vector Machines (SVM) for learning from imbalanced datasets. The TWSVM learns two non-parallel classifying hyperplanes by solving a couple of smaller sized problems. However, it is unsuitable for large datasets, as it involves matrix operations. In this paper, we discuss a Twin Neural Network (Twin NN) architecture for learning from large unbalanced datasets. The Twin NN also learns an optimal feature map, allowing for better discrimination between classes. We also present an extension of this network architecture for multiclass datasets. Results presented in the paper demonstrate that the Twin NN generalizes well and scales well on large unbalanced datasets.

cs.LG↗

arXiv2018

Bounded-Velocity Stochastic Control for Dynamic Resource Allocation

We consider a general class of dynamic resource allocation problems within a stochastic optimal control framework. This class of problems arises in a wide variety of applications, each of which intrinsically involves resources of different types and demand with uncertainty and/or variability. The goal involves dynamically allocating capacity for every resource type in order to serve the uncertain/variable demand, modeled as Brownian motion, and maximize the discounted expected net-benefit over an infinite time horizon based on the rewards and costs associated with the different resource types, subject to flexibility constraints on the rate of change of each type of resource capacity. We derive the optimal control policy within a bounded-velocity stochastic control setting, which includes efficient and easily implementable algorithms for governing the dynamic adjustments to resource allocation capacities over time. Computational experiments investigate various issues of both theoretical and practical interest, quantifying the benefits of our approach over recent alternative optimization approaches.

math.OC↗

arXiv2017

The paradoxical zero reflection at zero energy

Usually, the reflection probability $R(E)$ of a particle of zero energy incident on a potential which converges to zero asymptotically is found to be 1: $R(0)=1$. But earlier, a paradoxical phenomenon of zero reflection at zero energy ($R(0)=0$) has been revealed as a threshold anomaly. Extending the concept of Half Bound State (HBS) of 3D, here we show that in 1D when a symmetric (asymmetric) attractive potential well possesses a zero-energy HBS, $R(0)=0$ $(R(0)<<1)$. This can happen only at some critical values $q_c$ of an effective parameter $q$ of the potential well in the limit $E \rightarrow 0^+$. We demonstrate this critical phenomenon in two simple analytically solvable models which are square and exponential wells. However, in numerical calculations even for these two models $R(0)=0$ is observed only as extrapolation to zero energy from low energies, close to a precise critical value $q_c$. By numerical investigation of a variety of potential wells, we conclude that for a given potential well (symmetric or asymmetric), we can adjust the effective parameter $q$ to have a low reflection at a low energy.

quant-ph↗

arXiv2016

Revisiting double Dirac delta potential

We study a general double Dirac delta potential to show that this is the simplest yet versatile solvable potential to introduce double wells, avoided crossings, resonances and perfect transmission ($T=1$). Perfect transmission energies turn out to be the critical property of symmetric and anti-symmetric cases wherein these discrete energies are found to correspond to the eigenvalues of Dirac delta potential placed symmetrically between two rigid walls. For well(s) or barrier(s), perfect transmission [or zero reflectivity, $R(E)$] at energy $E=0$ is non-intuitive. However, earlier this has been found and called "threshold anomaly". Here we show that it is a critical phenomena and we can have $ 0 \le R(0)<1$ when the parameters of the double delta potential satisfy an interesting condition. We also invoke zero-energy and zero curvature eigenstate ($ψ(x)=Ax+B$) of delta well between two symmetric rigid walls for $R(0)=0$. We resolve that the resonant energies and the perfect transmission energies are different and they arise differently.

quant-ph↗