arXiv subjects

Deepak Kumar

Publications and source records attributed to Deepak Kumar.

At least 73 records · Page 4Linked to original sources

arXiv2025

Krutrim LLM: Multilingual Foundational Model for over a Billion People

India is a diverse society with unique challenges in developing AI systems, including linguistic diversity, oral traditions, data accessibility, and scalability. Existing foundation models are primarily trained on English, limiting their effectiveness for India's population. Indic languages comprise only 1 percent of Common Crawl corpora despite India representing 18 percent of the global population, leading to linguistic biases. Thousands of regional languages, dialects, and code mixing create additional representation challenges due to sparse training data. We introduce Krutrim LLM, a 2 trillion token multilingual model designed for India's linguistic landscape. It incorporates the largest known Indic dataset, mitigating data scarcity and ensuring balanced performance across dialects. Krutrim outperforms or matches state-of-the-art models on Indic benchmarks while maintaining competitive English performance. Despite being significantly smaller in training flops, Krutrim LLM matches or exceeds models like LLAMA-2 on 10 out of 16 tasks, with an average score of 0.57 versus 0.55. This evidences Krutrim's flexible multilingual fluency across diverse linguistic contexts. Krutrim is integrated with real-time search to improve factual accuracy in conversational AI applications. This enhances accessibility for over 1 billion users worldwide. Through intentional design choices addressing data imbalances, Krutrim LLM signifies meaningful progress in building ethical, globally representative AI models.

cs.CL↗

arXiv2025

The footprint of nuclear saturation properties on the neutron star $f$ mode oscillation frequencies: a machine learning approach

We investigate the intricate relationships between the non-radial $f$ mode oscillation frequencies of neutron stars (NS)s and the corresponding nuclear matter equation of state (EOS) using a machine learning (ML) approach within the ambit of the relativistic mean field (RMF) framework for nuclear matter. With two distinct parameterizations of the Walecka model, namely, (1) with non-linear self interactions of the scalar field (NL) and, (2) a density dependent Bayesian model (DDB), we perform a thorough examination of the $f$ mode frequency in relation to various nuclear saturation properties. The correlations between the $f$ mode frequencies and nuclear saturation properties reveal, through various analytical and ML methods, the complex nature of NSs and their potential as the cosmic laboratory for studying extreme states of matter. A principal component analysis (PCA) has been performed using mixed datasets from DDB and NL models to discriminate the relative importance of the different components of the EOS on the $f$ mode frequencies. Additionally, a {\it Random forest feature importance} analysis also elucidates the distinct roles of these properties in determining the $f$ mode frequency across a spectrum of NS masses. Our findings are further supported by symbolic regression searches, yielding high-accuracy relations with strong Pearson coefficients and minimal errors. These relations suggest new methodologies for probing NS core characteristics, such as energy density, pressure, and speed of sound from observations of non-radial $f$ mode oscillations of NSs.

nucl-th↗

arXiv2025

Uplink Rate Splitting Multiple Access with Imperfect Channel State Information and Interference Cancellation

This article investigates the performance of uplink rate splitting multiple access (RSMA) in a two-user scenario, addressing an under-explored domain compared to its downlink counterpart. With the increasing demand for uplink communication in applications like the Internet-of-Things, it is essential to account for practical imperfections, such as inaccuracies in channel state information at the receiver (CSIR) and limitations in successive interference cancellation (SIC), to provide realistic assessments of system performance. Specifically, we derive closed-form expressions for the outage probability, throughput, and asymptotic outage behavior of uplink users, considering imperfect CSIR and SIC. We validate the accuracy of these derived expressions using Monte Carlo simulations. Our findings reveal that at low transmit power levels, imperfect CSIR significantly affects system performance more severely than SIC imperfections. However, as the transmit power increases, the impact of imperfect CSIR diminishes, while the influence of SIC imperfections becomes more pronounced. Moreover, we highlight the impact of the rate allocation factor on user performance. Finally, our comparison with non-orthogonal multiple access (NOMA) highlights the outage performance trade-offs between RSMA and NOMA. RSMA proves to be more effective in managing imperfect CSIR and enhances performance through strategic message splitting, resulting in more robust communication.

eess.SP↗

arXiv2025

Characterizing the MrDeepFakes Sexual Deepfake Marketplace

The prevalence of sexual deepfake material has exploded over the past several years. Attackers create and utilize deepfakes for many reasons: to seek sexual gratification, to harass and humiliate targets, or to exert power over an intimate partner. In part enabling this growth, several markets have emerged to support the buying and selling of sexual deepfake material. In this paper, we systematically characterize the most prominent and mainstream marketplace, MrDeepFakes. We analyze the marketplace economics, the targets of created media, and user discussions of how to create deepfakes, which we use to understand the current state-of-the-art in deepfake creation. Our work uncovers little enforcement of posted rules (e.g., limiting targeting to well-established celebrities), previously undocumented attacker motivations, and unexplored attacker tactics for acquiring resources to create sexual deepfakes.

cs.CY↗

arXiv2025

Multimodal AI on Wound Images and Clinical Notes for Home Patient Referral

Chronic wounds affect 8.5 million Americans, particularly the elderly and patients with diabetes. These wounds can take up to nine months to heal, making regular care essential to ensure healing and prevent severe outcomes like limb amputations. Many patients receive care at home from visiting nurses with varying levels of wound expertise, leading to inconsistent care. Problematic, non-healing wounds should be referred to wound specialists, but referral decisions in non-clinical settings are often erroneous, delayed, or unnecessary. This paper introduces the Deep Multimodal Wound Assessment Tool (DM-WAT), a machine learning framework designed to assist visiting nurses in deciding whether to refer chronic wound patients. DM-WAT analyzes smartphone-captured wound images and clinical notes from Electronic Health Records (EHRs). It uses DeiT-Base-Distilled, a Vision Transformer (ViT), to extract visual features from images and DeBERTa-base to extract text features from clinical notes. DM-WAT combines visual and text features using an intermediate fusion approach. To address challenges posed by a small and imbalanced dataset, it integrates image and text augmentation with transfer learning to achieve high performance. In evaluations, DM-WAT achieved 77% with std 3% accuracy and a 70% with std 2% F1 score, outperforming prior approaches. Score-CAM and Captum interpretation algorithms provide insights into specific parts of image and text inputs that influence recommendations, enhancing interpretability and trust.

cs.LG↗

arXiv2025

Effect of Magnetic Field on Aqueous Humor Flows Inside Anterior Chamber of Human Eye

Aqueous humor (AH) dynamics is responsible for maintaining intraocular pressure, ocular health and targeted drug delivery within the eye. This study investigates the flow of AH within the anterior chamber (AC) under the combined influence of a uniform magnetic field and natural convection. Different orientations of the magnetic field and temperaature gradient are considered. A lubrication approximation is employed and the resulting equations are solved using regular perturbation method. The analytical solutions are validated using numerical simulations performed in COMSOL Multiphysics 6.2. In the standing position, AH flow field is characterised by a single vortex, while in the supine position, it forms two counter-rotating vortices. The velocity is found to be higher in standing position. The effect of a uniform magnetic field on the velocity is more significant in the supine position. The magnetic field does not change the flow field qualitatively as buoyancy is the primary driving force. In the standing position a magnetic field oriented perpendicular to the eye resulted in a greatest reduction of AH velocity, as compared to a magnetic field along the eye. This study is a step towards holistic approach for targeted drug delivery using magnetic fields in eye.

physics.flu-dyn↗

arXiv2025

Realistic Total-Body J-PET Geometry Optimization -- Monte Carlo Study

Total-Body PET is one of the most promising medical diagnostics modalities. The high sensitivity provided by Total-Body technology can be advantageous for novel tomography methods like positronium imaging. Several efforts are ongoing to lower the price of the TB-PET systems. Among the alternatives, the Jagiellonian PET (J-PET) technology, based on plastic scintillator strips, offers a low-cost alternative. The work aimed to compare five Total-Body J-PET geometries as a possible next generation J-PET scanner design. We present comparative studies of performance characteristics of the cost-effective Total-Body PET scanners using J-PET technology. We investigated in silico five Total-Body scanner geometries. Monte Carlo simulations of the XCAT phantom, the 2-meter sensitivity line source and positronium sensitivity phantoms were performed. We compared the sensitivity profiles for 2-gamma and 3-gamma tomography, relative cost of the setups and performed quantitative analysis of the reconstructed images. The analysis of the reconstructed XCAT images reveals the superiority of the seven-ring scanners over the three-ring setups. However, the three-ring scanners would be approximately 2-3 times cheaper. The peak sensitivity values for two-gamma vary from 20 to 34 cps/kBq. The sensitivity curves for the positronium tomography have a similar shape to the two-gamma sensitivity profiles. The peak values are lower compared to the two-gamma cases, from about 20-28 times, with a maximum of 1.66 cps/kBq. The results show the feasibility of multi-organ imaging of all the systems to be considered for the next generation of TB J-PET designs. The relative cost for all the scanners is about 10-4 times lower compared to the cost of the uExplorer. These properties coupled together with J-PET cost-effectiveness, make the J-PET technology an attractive solution for broad application in clinics.

physics.med-ph↗

arXiv2024

Statistical mechanics of an active wheel rolling in circles

Vibrated granular matter constitutes a useful system for studying the physics of active matter. Usually, self-propulsion is induced in grains through suitable asymmetry in the particle design. In this paper, we show that a symmetrical mini wheel placed on a vibrating plate self-propels along circular trajectories, showing chiral active dynamics. The chiral activity emerges through a sequence of spontaneous symmetry breaking in the particle's kinetics. The fact that isotropy, fore-aft, and chiral symmetries are broken spontaneously leads to distinct statistics, which include a temporal evolution involving stochastic resetting, a non-Gaussian velocity distribution with multiple peaks, broad power-law curvature distribution, and a bounded chirality probability, along with a phase transition from passive achiral to active chiral state as a function of vibration amplitude. Our study establishes the vibrated wheel as a three-state chiral active system that can serve as a model experimental system to study the non-equilibrium statistical mechanics and stochastic thermodynamics of chiral active systems and can inspire novel locomotion strategies in robotics.

cond-mat.soft↗

arXiv2024

Global Calderón-Zygmund theory for fractional Laplacian type equations

We establish several fine boundary regularity results of weak solutions to non-homogeneous $s$-fractional Laplacian type equations. In particular, we prove sharp Calderón-Zygmund type estimates of $u/d^s$ depending on the regularity assumptions on the associated kernel coefficient including VMO, Dini continuity or the Hölder continuity, where $u$ is a weak solution to such a nonlocal problem and $d$ is the distance to the boundary function of a given domain. Our analysis is based on point-wise behaviors of maximal functions of $u/d^s$.

math.AP↗

arXiv2024

Complexities in the growth and stabilization of polar phase in the Hf$_{0.5}$Zr$_{0.5}$O$_2$ thin films grown by Pulsed Laser Deposition

After the discovery of ferroelectricity in HfO$_2$ based thin films a decade ago, ferroelectric Hf$_{0.5}$Zr$_{0.5}$O$_2$ (HZO) thin films are frequently being utilized in the CMOS (Complementary Metal- Oxide Semiconductor) and logic devices, thanks to their large remnant polarization, high retention and endurance. A great deal of effort has been made towards understanding the origin of ferroelectricity in epitaxial HZO thin films and controlling the microstructure at the atomic level which governs the ferroelectric phase. Nevertheless, the HZO films still suffer from fundamental questions, such as (1) the vagueness of interfacial mechanisms between HZO, buffer layer and the substrate which controls the polar phase; (2) the nature of the metastable polar phase responsible for the ferroelectricity, be it orthorhombic or rhombohedral; which are poorly understood. Here, we have addressed these issues by employing the in-situ reflection high energy electron diffraction -- assisted pulsed laser deposition and mapping the asymmetrical polar maps on high quality HZO films grown on functional perovskite oxide substrates. The interface between La$_{0.7}$Sr$_{0.3}$MnO$_3$ (LSMO) and the substrate is shown to be quite important, and a slightly rougher interface of the former destabilizes the ferroelectric phase of HZO irrespective of well-controlled growth of the ferroelectric layers. A rhombohedral-like symmetry of HZO unit cell is extracted through the x-ray diffraction asymmetrical polar maps. The ferroelectric measurements on a nearly 7 nm HZO film on STO(001) substrate display a remnant polarization close to 8 uC/cm$^2$. These results highlight the complexities involved at the atomic scale interface in the binary oxides thin films and can be of importance to the HfO$_2$-based ferroelectric community which is still at its infancy.

cond-mat.mtrl-sci↗

arXiv2024

Motion-Driven Neural Optimizer for Prophylactic Braces Made by Distributed Microstructures

Joint injuries, and their long-term consequences, present a substantial global health burden. Wearable prophylactic braces are an attractive potential solution to reduce the incidence of joint injuries by limiting joint movements that are related to injury risk. Given human motion and ground reaction forces, we present a computational framework that enables the design of personalized braces by optimizing the distribution of microstructures and elasticity. As varied brace designs yield different reaction forces that influence kinematics and kinetics analysis outcomes, the optimization process is formulated as a differentiable end-to-end pipeline in which the design domain of microstructure distribution is parameterized onto a neural network. The optimized distribution of microstructures is obtained via a self-learning process to determine the network coefficients according to a carefully designed set of losses and the integrated biomechanical and physical analyses. Since knees and ankles are the most commonly injured joints, we demonstrate the effectiveness of our pipeline by designing, fabricating, and testing prophylactic braces for the knee and ankle to prevent potentially harmful joint movements.

physics.med-ph↗

arXiv2024

PressProtect: Helping Journalists Navigate Social Media in the Face of Online Harassment

Social media has become a critical tool for journalists to disseminate their work, engage with their audience, and connect with sources. Unfortunately, journalists also regularly endure significant online harassment on social media platforms, ranging from personal attacks to doxxing to threats of physical harm. In this paper, we seek to understand how to make social media usable for journalists who face constant digital harassment. To begin, we conduct a set of need-finding interviews with Asian American and Pacific Islander journalists to understand where existing platform tools and newsroom resources fall short in adequately protecting journalists, especially those of marginalized identities. We map journalists' unmet needs to concrete design goals, which we use to build PressProtect, an interface that provides journalists greater agency when engaging with readers on Twitter/X. Through user testing with eight journalists, we evaluate PressProtect and find that participants felt it effectively protected them against harassment and could also generalize to serve other visible and vulnerable groups. We conclude with a discussion of our findings and recommendations for social platforms hoping to build defensive defaults for journalists facing online harassment.

cs.CY↗

arXiv2024

Comparing Femtosecond Optical Tweezers with Conventional CW Optical Tweezers

In this work, we present a comparative study between continuous-wave (CW) and pulsed optical tweezers for 250 nm, 500 nm and 1-micron radius polystyrene beads at 5 different laser powers. We have used a Ti:Sapphire (MIRA 900F) laser that can be easily switched from CW to pulsed mode of operation, so there is no change in the experimental conditions in the two cases. We have measured the difference in the trap strength in both cases by fitting the power spectrum curve with Lorentzian. As it turns out, trapping with pulsed tweezers seems to be more effective for the smaller particles and as the particle size is increased both CW and pulsed tweezers appear to be equally effective at lower average laser powers but as the power is increased pulsed tweezers do a better job at stable trapping.

physics.optics↗

arXiv2024

A framework to identify supercritical and subcritical Turing bifurcations: Case study of a system sustaining cubic and quadratic autocatalysis

In this work, we focus on an autocatalytic reaction-diffusion model and carry out multiple scale weakly nonlinear analysis. A cubic and a quadratic autocatalytic reaction system is analysed. We develop a framework to identify the critical surfaces in parameter space across which the nature of the Turing bifurcation changes from supercritical to subcritical. These are verified by direct numerical simulations of the system. Using weakly nonlinear analysis, we derive equations up to the fifth order that governs the amplitude of the spatial patterns. The limit point of the bifurcating solution is captured accurately by extending the analysis to the fifth order for the case of subcritical bifurcation. The numerical solutions are in good agreement with the predictions of the weakly nonlinear analysis for supercritical bifurcations. We show that when multiple steady states arise Turing patterns can coexist with another spatially uniform steady states. Furthermore, we show that our framework can be extended to get different patterns like squares and hexagons in a two-dimensional domain. We show that the shape of Turing patterns is influenced by the domain size. This shows that the geometry can influence the kind of patterns formed in natural systems. This study will aid the experimentalist identify operating conditions where Turing patterns can be obtained.

nlin.AO↗

arXiv2024

Calderón-Zygmund theory of nonlocal parabolic equations with discontinuous coefficients

We prove Calderón-Zygmund type estimates of weak solutions to non-homogeneous nonlocal parabolic equations under a minimal regularity requirement on kernel coefficients. In particular, the right-hand side is presented by a sum of fractional Laplacian type data and a non-divergence type data. Interestingly, even though the kernel coefficients are discontinuous, we obtain a significant increment of fractional differentiability for the solutions, which is not observed in the corresponding local parabolic equations.

math.AP↗

arXiv2024

End-to-End Waveform and Beamforming Optimization for RF Wireless Power Transfer

Radio frequency (RF) wireless power transfer (WPT) is a key technology for future low-power wireless systems. However, the inherently low end-to-end power transfer efficiency (PTE) is challenging for practical applications. The main factors contributing to it are the channel losses, transceivers' power consumption, and losses related, e.g., to the digital-to-analog converter (DAC), high-power amplifier, and rectenna. Optimizing PTE requires careful consideration of these factors, motivating the current work. Herein, we consider an analog multi-antenna power transmitter that aims to charge a single energy harvester. We first provide a mathematical framework to calculate the harvested power from multi-tone signal transmissions and the system power consumption. Then, we formulate the joint waveform and analog beamforming design problem to minimize power consumption and meet the charging requirements. Finally, we propose an optimization approach relying on swarm intelligence to solve the specified problem. Simulation results quantify the power consumption reduction as the DAC, phase shifters resolution, and antenna length are increased, while it is seen that increasing system frequency results in higher power consumption.

eess.SP↗

arXiv2024

Guided Conditional Diffusion Classifier (ConDiff) for Enhanced Prediction of Infection in Diabetic Foot Ulcers

To detect infected wounds in Diabetic Foot Ulcers (DFUs) from photographs, preventing severe complications and amputations. Methods: This paper proposes the Guided Conditional Diffusion Classifier (ConDiff), a novel deep-learning infection detection model that combines guided image synthesis with a denoising diffusion model and distance-based classification. The process involves (1) generating guided conditional synthetic images by injecting Gaussian noise to a guide image, followed by denoising the noise-perturbed image through a reverse diffusion process, conditioned on infection status and (2) classifying infections based on the minimum Euclidean distance between synthesized images and the original guide image in embedding space. Results: ConDiff demonstrated superior performance with an accuracy of 83% and an F1-score of 0.858, outperforming state-of-the-art models by at least 3%. The use of a triplet loss function reduces overfitting in the distance-based classifier. Conclusions: ConDiff not only enhances diagnostic accuracy for DFU infections but also pioneers the use of generative discriminative models for detailed medical image analysis, offering a promising approach for improving patient outcomes.

cs.CV↗

arXiv2024

Geometry-induced friction at a soft interface

Soft and biological matter come in a variety of shapes and geometries. When soft surfaces that do not fit into each other due to a mismatch in Gaussian curvatures form an interface, beautiful geometry-induced patterns emerge. In this paper, we study the effect of geometry on the dynamical response of soft surfaces moving relative to each other. Using a novel experimental scheme, we measure friction between a highly bendable thin polymer sheet and a hydrogel substrate. At this soft and low-friction interface, we find a strong dependence of friction on the relative geometry of the two surfaces - a flat sheet experiences significantly larger friction on a spherical substrate than on planar or cylindrical substrate. We show that the stress developed in the sheet due to its geometrically incompatible confinement is responsible for the enhanced friction. This mechanism also leads to a transition in the nature of friction as the sheet radius is increased beyond a critical value. Our finding reveals a hitherto unnoticed non-specific mechanism of purely geometrical origin that may influence friction significantly in soft, biological, and nano-scale systems. In particular, it provokes us to re-examine our understanding of phenomena such as the curvature dependence of biological cell mobility.

cond-mat.soft↗