arXiv subjects

Pawan Kumar

Publications and source records attributed to Pawan Kumar.

At least 91 records · Page 5Linked to original sources

arXiv2023

Adaptive Consensus Optimization Method for GANs

We propose a second order gradient based method with ADAM and RMSprop for the training of generative adversarial networks. The proposed method is fastest to obtain similar accuracy when compared to prominent second order methods. Unlike state-of-the-art recent methods, it does not require solving a linear system, or it does not require additional mixed second derivative terms. We derive the fixed point iteration corresponding to proposed method, and show that the proposed method is convergent. The proposed method produces better or comparable inception scores, and comparable quality of images compared to other recently proposed state-of-the-art second order methods. Compared to first order methods such as ADAM, it produces significantly better inception scores. The proposed method is compared and validated on popular datasets such as FFHQ, LSUN, CIFAR10, MNIST, and Fashion MNIST for image generation tasks\footnote{Accepted in IJCNN 2023}. Codes: \url{https://github.com/misterpawan/acom}

cs.LG↗

arXiv2023

Angle based dynamic learning rate for gradient descent

In our work, we propose a novel yet simple approach to obtain an adaptive learning rate for gradient-based descent methods on classification tasks. Instead of the traditional approach of selecting adaptive learning rates via the decayed expectation of gradient-based terms, we use the angle between the current gradient and the new gradient: this new gradient is computed from the direction orthogonal to the current gradient, which further helps us in determining a better adaptive learning rate based on angle history, thereby, leading to relatively better accuracy compared to the existing state-of-the-art optimizers. On a wide variety of benchmark datasets with prominent image classification architectures such as ResNet, DenseNet, EfficientNet, and VGG, we find that our method leads to the highest accuracy in most of the datasets. Moreover, we prove that our method is convergent.

cs.LG↗

arXiv2023

Light-weight Deep Extreme Multilabel Classification

Extreme multi-label (XML) classification refers to the task of supervised multi-label learning that involves a large number of labels. Hence, scalability of the classifier with increasing label dimension is an important consideration. In this paper, we develop a method called LightDXML which modifies the recently developed deep learning based XML framework by using label embeddings instead of feature embedding for negative sampling and iterating cyclically through three major phases: (1) proxy training of label embeddings (2) shortlisting of labels for negative sampling and (3) final classifier training using the negative samples. Consequently, LightDXML also removes the requirement of a re-ranker module, thereby, leading to further savings on time and memory requirements. The proposed method achieves the best of both worlds: while the training time, model size and prediction times are on par or better compared to the tree-based methods, it attains much better prediction accuracy that is on par with the deep learning based methods. Moreover, the proposed approach achieves the best tail-label prediction accuracy over most state-of-the-art XML methods on some of the large datasets\footnote{accepted in IJCNN 2023, partial funding from MAPG grant and IIIT Seed grant at IIIT, Hyderabad, India. Code: \url{https://github.com/misterpawan/LightDXML}

cs.LG↗

arXiv2023

Contact Angle Hysteresis on Rough Surfaces Part II: Energy Dissipation via Microscale Interface Dynamics

The wetting behaviour of surfaces is important for various applications like super-hydrophobic surfaces, enhanced oil recovery, mining of metal ores and anti-icing surfaces etc. For rough surfaces, which are the rule rather than the exception, designing textured surfaces that have wetting properties tailored to suit these applications generally involves either bio-mimicry or trial and error. Existing wetting theories such as the well-known Wenzel (1936) and Cassie & Baxter (1944) models do not predict wetting regimes and importantly, only give a single equilibrium angle rather than a range of stable contact angles (hysteresis) as observed in reality. In this paper, we use a roughness-scale mechanical energy balance (derived in part I) combined with simulations of micro-scale interface dynamics based on open-source software (Surface Evolver (Brakke 1992)) to calculate the energy dissipation during the motion of an interface over a chemically homogeneous rough surface. This dissipation is used to predict contact angle hysteresis (CAH) from knowledge of just the surface roughness topography and equilibrium contact angle. We simulate interface dynamics over a surface decorated with a periodic array of round-edge square pillars and show that the energy dissipated varies approximately as $ϕlnϕ$ with the area fraction ($ϕ$), and becomes zero as $ϕ \to 0$. The CAH predicted by our method is in good agreement with the experimental results of Forsberg et al. (2010), especially at low area fractions. We also compute CAH for an interface moving at 45$°$ to the surface periodicity direction, showing that the experimental measurements are bracketed by the 0$°$ and 45$°$ advance direction results.

physics.flu-dyn↗

arXiv2023

Wafer-scale growth of two-dimensional, phase-pure InSe

Two-dimensional (2D) indium monoselenide (InSe) has attracted significant attention as a III-VI two-dimensional semiconductor (2D) with a combination of favorable attributes from III-V semiconductors as well as van der Waals 2D transition metal dichalcogenides. Nevertheless, the large-area synthesis of phase-pure 2D InSe remains unattained due to the complexity of the binary In-Se system and the difficulties in promoting lateral growth. Here, we report the first polymorph-selective synthesis of epitaxial 2D InSe by metal-organic chemical deposition (MOCVD) over 2 inch diameter sapphire wafers. We achieve thickness-controlled, layer-by-layer epitaxial growth of InSe on c-plane sapphire via dynamic pulse control of Se/In flux ratio. The layer-by-layer growth allows thickness control over wafer scale with tunable optical properties comparable to bulk crystals. Finally, the gate-tunable electrical transport suggests that MOCVD-grown InSe could be a potential channel material for back-end-of-line integration in logic transistors with field-effect mobility comparable to single-crystalline flakes.

cond-mat.mtrl-sci↗

arXiv2023

Qualitative Data Augmentation for Performance Prediction in VLSI circuits

Various studies have shown the advantages of using Machine Learning (ML) techniques for analog and digital IC design automation and optimization. Data scarcity is still an issue for electronic designs, while training highly accurate ML models. This work proposes generating and evaluating artificial data using generative adversarial networks (GANs) for circuit data to aid and improve the accuracy of ML models trained with a small training data set. The training data is obtained by various simulations in the Cadence Virtuoso, HSPICE, and Microcap design environment with TSMC 180nm and 22nm CMOS technology nodes. The artificial data is generated and tested for an appropriate set of analog and digital circuits. The experimental results show that the proposed artificial data generation significantly improves ML models and reduces the percentage error by more than 50\% of the original percentage error, which were previously trained with insufficient data. Furthermore, this research aims to contribute to the extensive application of AI/ML in the field of VLSI design and technology by relieving the training data availability-related challenges.

cs.LG↗

arXiv2023

SynGraphy: Succinct Summarisation of Large Networks via Small Synthetic Representative Graphs

We describe SynGraphy, a method for visually summarising the structure of large network datasets that works by drawing smaller graphs generated to have similar structural properties to the input graphs. Visualising complex networks is crucial to understand and make sense of networked data and the relationships it represents. Due to the large size of many networks, visualisation is extremely difficult; the simple method of drawing large networks like those of Facebook or Twitter leads to graphics that convey little or no information. While modern graph layout algorithms can scale computationally to large networks, their output tends to a common "hairball" look, which makes it difficult to even distinguish different graphs from each other. Graph sampling and graph coarsening techniques partially address these limitations but they are only able to preserve a subset of the properties of the original graphs. In this paper we take the problem of visualising large graphs from a novel perspective: we leave the original graph's nodes and edges behind, and instead summarise its properties such as the clustering coefficient and bipartivity by generating a completely new graph whose structural properties match that of the original graph. To verify the utility of this approach as compared to other graph visualisation algorithms, we perform an experimental evaluation in which we repeatedly asked experimental subjects (professionals in graph mining and related areas) to determine which of two given graphs has a given structural property and then assess which visualisation algorithm helped in identifying the correct answer. Our summarisation approach SynGraphy compares favourably to other techniques on a variety of networks.

cs.SI↗

arXiv2023

Hybrid Pulsar-Magnetar Model for FRB 20191221A

We show that the 216.8$\pm$0.1 ms periodicity reported for the fast radio burst (FRB) 20191221A is very constraining for burst models. The high accuracy of burst periodicity (better than one part in 10$^3$), and the 2\% duty cycle (ratio of burst-duration and inter-burst interval), suggest a pulsar-like rotating beam model for the observed activity; the radio waves are produced along open field lines within $\sim 10^7$ cm of the neutron star surface, and the beam periodically sweeps across the observer as the star spins. According to this picture, FRB 20191221A is a factor $\sim 10^{12}$ scaled up version of galactic pulsars with one major difference: whereas pulsars convert rotational kinetic energy to EM waves, the outbursts of 20191221A require conversion of magnetic energy to radiation.

astro-ph.HE↗

arXiv2022

The plasma suppression effect can be ignored in realistic FRB models invoking bunched coherent radio emission

One widely discussed mechanism to produce highly coherent radio emission of fast radio bursts (FRBs) is coherent emission by bunches, either via curvature radiation or inverse Compton scattering (ICS). It has been suggested that the plasma oscillation effect can significantly suppress coherent emission power by bunches. We examine this criticism in this paper. The suppression factor formalism was derived within the context of radio pulsars in which radio waves are in the low-amplitude, linear regime and cannot directly be applied to the large-amplitude, non-linear regime relevant for FRBs. Even if one applies this linear treatment, plasma suppression is not important for two physical reasons. First, for an efficient radiation mechanism such as ICS, the required plasma density is not high so that a high-density plasma may not exist. Second, both bunched coherent mechanisms demand that a large global parallel electric field ($E_\parallel$) must exist in the emission region in order to continuously inject energy to the bunches to power an FRB. In order to produce typical FRB duration via coherent curvature or ICS radiation, a parallel electric field must be present to balance the acceleration and radiation back-reaction. The plasma suppression factor should be modified with the existence of $E_\parallel$. We show that the correction factor for curvature radiation, $f_{\rm cur}$, increases with $E_\parallel$ and becomes 1 when $E_\parallel$ reaches the radiation-reaction-limited regime. We conclude that the plasma suppression effect can be ignored for realistic FRB emission models invoking bunched coherent radio emission.

astro-ph.HE↗

arXiv2022

Transparency of Fast Radio Burst Waves in Magnetar Magnetospheres

At least some fast radio bursts (FRBs) are produced by magnetars. Even though mounting observational evidence points towards a magnetospheric origin of FRB emission, the question of the location for FRB generation continues to be debated. One argument suggested against the magnetospheric origin of bright FRBs is that the radio waves associated with an FRB may lose most of their energy before escaping the magnetosphere because the cross-section for $e^\pm$ to scatter large-amplitude EM waves in the presence of a strong magnetic field is much larger than the Thompson cross-section. We have investigated this suggestion and find that FRB radiation traveling through the open field line region of a magnetar's magnetosphere does not suffer much loss due to two previously ignored factors. First, the plasma in the outer magnetosphere ($r \gta 10^9 \ $cm), where the losses are potentially most severe, is likely to be flowing outward at a high Lorentz factor $γ_p \geq 10^3$. Second, the angle between the wave vector and the magnetic field vector, $θ_B$, in the outer magnetosphere is likely of the order of 0.1 radian or smaller due in part to the intense FRB pulse that tilts open magnetic field lines so that they get aligned with the pulse propagation direction. Both these effects reduce the interaction between the FRB pulse and the plasma substantially. We find that a bright FRB with an isotropic luminosity $L_{\rm frb} \gta 10^{42} \ {\rm erg \ s^{-1}}$ can escape the magnetosphere unscathed for a large section of the $γ_p-θ_B$ parameter space, and therefore conclude that the generation of FRBs in magnetar magnetosphere passes this test.

astro-ph.HE↗

arXiv2022

Ultrathin Broadband Metasurface Superabsorbers from a van der Waals Semimetal

Metamaterials and metasurfaces operating in the visible and near-infrared (NIR) offer a promising route towards next-generation photodetectors and devices for solar energy harvesting. While numerous metamaterials and metasurfaces using metals and semiconductors have been demonstrated, semimetals-based metasurfaces in the vis-NIR range are notably missing. Here, we experimentally demonstrate a broadband metasurface superabsorber based on large area, semimetallic, van der Waals PtSe2 thin films in agreement with electromagnetic simulations. Our results show that PtSe2 is an ultrathin and scalable semimetal that concurrently possesses high index and high extinction across the vis-NIR range. Consequently, the thin-film PtSe2 on a reflector separated by a dielectric spacer can absorb > 85 % for the unpatterned case and ~97 % for the optimized 2D metasurface in the 400-900 nm range making it one of the strongest and thinnest broadband perfect absorbers to date. Our results present a scalable approach to photodetection and solar energy harvesting, demonstrating the practical utility of high index, high extinction semimetals for nanoscale optics.

physics.optics↗

arXiv2022

Propagation of Alfvén waves in the charge starvation regime

We present numerical simulation results for the propagation of Alfvén waves in the charge starvation regime. This is the regime where the plasma density is below the critical value required to supply the current for the wave. We analyze a conservative scenario where Alfvén waves pick up charges from the region where the charge density exceeds the critical value and advect them along at a high Lorentz factor. The system consisting of the Alfvén wave and charges being carried with it, which we call charge-carrying Alfvén wave (CC-AW), moves through a medium with small, but non-zero, plasma density. We find that the interaction between CC-AW and the stationary medium has a 2-stream like instability which leads to the emergence of a strong electric field along the direction of the unperturbed magnetic field. The growth rate of this instability is of order the plasma frequency of the medium encountered by the CC-AW. Our numerical code follows the system for hundreds of wave periods. The numerical calculations suggest that the final strength of the electric field is of order a few percent of the Alfvén wave amplitude. Little radiation is produced by the sinusoidally oscillating currents associated with the instability during the linear growth phase. However, in the nonlinear phase, the fluctuating current density produces strong EM radiation near the plasma frequency and limits the growth of the instability.

astro-ph.HE↗

arXiv2022

Gravitational lensing in the presence of plasma scattering with application to Fast Radio Bursts

We describe how gravitational lensing of fast radio bursts (FRBs) is affected by a plasma screen in the vicinity of the lens or somewhere between the source and the observer. Wave passage through a turbulent medium affects gravitational image magnification, lensing probability (particularly for strong magnification events), and the time delay between images. The magnification is suppressed because of the broadening of the angular size of the source due to scattering by the plasma. The time delay between images is modified as the result of different dispersion measure (DM) along photon trajectories for different images. Each of the image lightcurve is also broadened due to wave scattering so that the images could have distinct temporal profiles. The first two effects are most severe for stellar and sub-stellar mass lens, and the last one (scatter broadening) for lenses and plasma screens at cosmological distances from the source/observer. This could limit the use of FRBs to measure their cosmic abundance. On the other hand, when the time delay between images is large, such that the lightcurve of a transient source has two or more well separated peaks, the different DMs along the wave paths of different images can probe density fluctuations in the IGM on scales $\lesssim 10^{-6}$ rad and explore the patchy reionization history of the universe using lensed FRBs at high redshifts. Different rotation measure (RM) along two image paths can convert linearly polarized radiation from a source to partial circular polarization.

astro-ph.HE↗

arXiv2022

Physical link of the polar field build-up with the Waldmeier effect broadens the scope of early solar cycle prediction: Cycle 25 is likely to be slightly stronger than Cycle 24

Prediction of the solar cycle is challenging but essential because it drives space weather. Several predictions with varying amplitudes of the ongoing Cycle~25 have been made. We show that an aspect of the Waldmeier effect (WE2), i.e., a strong positive correlation between the rise rate and the amplitude of the cycle, has a physical link with the build-up of the previous cycle's polar field after its reversal. We find that the rise rate of the polar field is highly correlated with the rise rate and the amplitude of the next solar cycle. Thus, the prediction of the amplitude of the solar cycle can be made just a few years after the reversal of the previous cycle's polar field, thereby extending the scope of the solar cycle prediction to much earlier than the usual time. Our prediction of Cycle 25 based on the rise rate of the previous polar field is $137\pm 23$, which is quite close to the prediction $138\pm 26$ based on the WE2 computed from the available 2 years sunspot data of the ongoing cycle.

astro-ph.SR↗

arXiv2022

High Density, Localized Quantum Emitters in Strained 2D Semiconductors

Two-dimensional chalcogenide semiconductors have recently emerged as a host material for quantum emitters of single photons. While several reports on defect and strain-induced single photon emission from 2D chalcogenides exist, a bottom-up, lithography-free approach to producing a high density of emitters remains elusive. Further, the physical properties of quantum emission in the case of strained 2D semiconductors are far from being understood. Here, we demonstrate a bottom-up, scalable, and lithography-free approach to creating large areas of localized emitters with high density (~150 emitters/um2) in a WSe2 monolayer. We induce strain inside the WSe2 monolayer with high spatial density by conformally placing the WSe2 monolayer over a uniform array of Pt nanoparticles with a size of 10 nm. Cryogenic, time-resolved, and gate-tunable luminescence measurements combined with near-field luminescence spectroscopy suggest the formation of localized states in strained regions that emit single photons with a high spatial density. Our approach of using a metal nanoparticle array to generate a high density of strained quantum emitters opens a new path towards scalable, tunable, and versatile quantum light sources.

cond-mat.mes-hall↗

arXiv2022

Nonlinear quantum spectroscopy with Parity-Time symmetric integrated circuits

We propose a novel quantum nonlinear interferometer design that incorporates a passive PT symmetric coupler sandwiched between two nonlinear sections where signal-idler photon pairs are generated. The PT-symmetry enables efficient coupling of the longer-wavelength idler photons and facilitates the sensing of losses in the second waveguide exposed to analyte under investigation, whose absorption can be inferred by measuring only the signal intensity at a shorter wavelength where efficient detectors are readily available. Remarkably, we identify a new phenomenon of sharp signal intensity fringe shift at critical idler loss values, which is distinct from the previously studied PT-symmetry breaking. We discuss how such unconventional properties arising from quantum interference can provide a route to enhancing the sensing of analytes and facilitate broadband spectroscopy applications in integrated photonic platforms.

physics.optics↗

arXiv2022

The large landscape of supernova, GRB and cocoon interactions

Long gamma ray bursts (LGRBs) are associated to the collapse of a massive star and the formation of a relativistic jet. As the jet propagates through the star, it forms an extended, hot cocoon. The dynamical evolution of the jet/cocoon system and its interaction with the environment has been studied extensively both analytically and numerically. On the other hand, the role played by the supernova (SN) explosion associated with LGRBs in determining the outcome of the system has been barely considered. In this paper, we discuss the large landscape of outcomes resulting from the interaction of the SN, jet and cocoon. We show that the outcome depends mainly on three timescales: the times for the cocoon and supernova shock wave to break through the surface of the progenitor star, and the time needed for the cocoon to engulf completely the progenitor star. The delay between the launch of the SN shock moving through the progenitor star and the jet can be related to these three timescales. Depending on the ordering of these time scales, the jet-cocoon might propagate inside the SN ejecta or the other way around, and the outcome for the properties of the explosion would be different. We discuss the imprint of the complex interaction between the jet-cocoon and the supernova shock on the emergent thermal and non-thermal radiation.

astro-ph.HE↗

arXiv2022

Group index matched frequency conversion in lithium niobate on insulator waveguides

Sources of spectrally engineered photonic states are a key resource in several quantum technologies. Of particular importance are the so-called factorizable biphoton states which possess no spectral entanglement and hence, are ideal for heralded generation of high-purity single photons. An essential prerequisite for generating these states through nonlinear frequency conversion is the control over the group indices of the photonic modes of the source. Here, we show that thin-film lithium niobate on insulator (LNOI) is an excellent platform for this purpose. We design and fabricate periodically poled ridge waveguides in LNOI to demonstrate group index engineering of its guided photonic modes and harness this control to experimentally realize on-chip group index matched type-II sum-frequency generation (SFG) and photon-pair creation through spontaneous parametric down-conversion (SPDC). Also, we numerically study the role of the top cladding layer in tuning the dispersion properties of the ridge waveguide structures and reveal a distinctive difference between the air and silica-clad designs which are currently among the two most common device cladding configurations in LNOI. We expect that these results will be relevant for various classical and quantum applications where dispersion control is crucial in tailoring the nonlinear response of the LNOI-based devices.

physics.optics↗