Search

This paper investigates the distinctions between gradient methods applied to non-differentiable functions (NGDMs) and classical gradient descents (GDs) designed for differentiable functions. First, we demonstrate significant differences in the convergence properties of NGDMs compared to GDs, challenging the applicability of the extensive neural network convergence literature based on $L-smoothness$ to non-smooth neural networks. Next, we demonstrate the paradoxical nature of NGDM solutions for $L_{1}$-regularized problems, showing that increasing the regularization penalty leads to an increase in the $L_{1}$ norm of optimal solutions in NGDMs. Consequently, we show that widely adopted $L_{1}$ penalization-based techniques for network pruning do not yield expected results. Additionally, we dispel the common belief that optimization algorithms like Adam and RMSProp perform similarly in non-differentiable contexts. Finally, we explore the Edge of Stability phenomenon, indicating its inapplicability even to Lipschitz continuous convex differentiable functions, leaving its relevance to non-convex non-differentiable neural networks inconclusive. Our analysis exposes misguided interpretations of NGDMs in widely referenced papers and texts due to an overreliance on strong smoothness assumptions, emphasizing the necessity for a nuanced understanding of foundational assumptions in the analysis of these systems.

This work presents the magnetic field geometry in a complex of three cometary (with head-tail morphology) globules, namely LDN 323, LDN 328, and LDN 331, using R-band polarization measurements of background stars. These observations were combined with a Planck sky survey to study the large-scale morphology of the magnetic fields in the region. The distances of the target stars were adopted from the Gaia catalog. The variation of degree of polarization and polarization position angle with distances of stars is analyzed. The field geometry is mostly found to follow the cometary shape of the cloud, with some randomness at certain locations. For studying the correlation between cloud morphology and magnetic field orientations, a modified version of the Histogram of Relative Orientation analysis was employed.

The orbital, which represents the shape of the electron cloud, very often strongly influences the manifestation of various exotic phenomena, e.g., magnetism, metal-insulator transition, colossal magnetoresistance, unconventional superconductivity etc. in solid-state systems. The observation of the antiferromagnetism in $RE$TiO$_3$ ($RE$=rare earth) series has been puzzling since the celebrated Kugel-Khomskii model of spin-orbital super exchange predicts ferromagnetism in an orbitally degenerate $d^1$ systems. Further, the existence of the orbitally ordered vs. orbital liquid phase in both antiferromagnetic and paramagnetic phase have been unsettled issues thus far. To address these long-standing questions, we investigate single crystalline film of PrTiO$_3$. Our synchrotron X-ray diffraction measurements confirm the retention of bulk-like orthorhombic ($D_{2h}$) symmetry in the thin film geometry. We observe similar X-ray linear dichroism signal in both paramagnetic and antiferromagnetic phase, which can be accounted by ferro orbital ordering (FOO). While the presence of $D_{2h}$ crystal field does not guarantee lifting of orbital degeneracy always, we find it to be strong enough in these rare-earth titanates, leading to the FOO state. Thus, our work demonstrates the orthorhombic distortion is the driving force for the orbital ordering of antiferromagnetic $RE$TiO$_3$., Comment: 37 pages and 17 figures

The nature of energy dissipation in 2D superconductors under perpendicular magnetic field at small current excitations has been extensively studied over the past two decades. However, dissipation mechanisms at high current drives remain largely unexplored. Here we report on the distinct behavior of energy dissipation in the AlOx/KTaO$_3$ (111) system hosting 2D superconductivity in the intermediate disorder regime. The results show that below the Berezinskii Kosterlitz Thouless (BKT) phase transition temperature ($T_\mathrm{BKT}$), hot-spots and Larkin Ovchinnikov type flux-flow instability (FFI) are the major channels of dissipation, leading to pronounced voltage instability at large currents. Furthermore, such FFI leads to a rare observation of clockwise hysteresis in current-voltage characteristics within the temperature range $T_\mathrm{BKT} < T < T_\mathrm{C}$ ($T_\mathrm{C}$ is superconducting transition temperature). These findings deepen our understanding of how a BKT system ultimately transforms to a normal state under increasing current., Comment: 51 pages including supplemental

$RE$NiO$_3$ is a negative charge transfer energy system and exhibits a temperature-driven metal-insulator transition (MIT), which is also accompanied by a bond disproportionation (BD) transition. In order to explore how hole doping affects the BD transition, we have investigated the electronic structure of single-crystalline thin films of Nd$_{1-x}$Ca$_x$NiO$_3$ by synchrotron based experiments and {\it ab-initio} calculations. For a small value of $x$, we find that the doped holes are localized on one or more Ni sites around the dopant Ca$^{2+}$ ions, while the BD state for the rest of the lattice remains intact. The effective charge transfer energy ($\Delta$) increases with Ca concentration and the formation of BD phase is not favored above a critical $x$, suppressing the insulating phase. Our present study firmly demonstrates that the appearance of BD mode is essential for the MIT and settles a long-standing debate about the role of structural distortions for the MIT of the $RE$NiO$_3$ series., Comment: 4 Figures

Contradictory reports about the electronic and magnetic behavior of bulk Nd0:5Ba0:5MnO3 with uniform disorder exist in the literature. In this work, we investigate the thin films of Nd0:5Ba0:5MnO3, grown by pulsed laser deposition. The success of epitaxial growth in a layer-by-layer mode and high quality of the films have been confirmed by in-situ reflection high energy electron diffraction, atomic force microscopy, X-ray reflectivity, and, X-ray diffraction. These films are found to be insulting and exhibit spin-glass behavior at low temperatures. The study offers an opportunity to tune the competition between charge/orbital ordering and ferromagnetism in half-doped manganites through heterostructure engineering., Comment: Appl. Phys. Lett. 120, 061602 (2022) , 10 pages,5 figures

The choice of electrostatic gating over the conventional chemical doping for phase engineering of quantum materials is attributed to the fact that the former can reversibly tune the carrier density without affecting the system's level of disorder. However, this proposition seems to break down in field-effect transistors involving SrTiO$_3$ (STO) based two-dimensional electron gases. Such peculiar behavior is associated with the electron trapping under an external electric field. However, the microscopic nature of trapping centers remains an open question. In this paper, we investigate electric field-induced charge trapping/detrapping phenomena at the conducting interface between band insulators $\gamma$-Al$_2$O$_3$ and STO. Our transport measurements reveal that the charge trapping under +ve back gate voltage ($V_g$) above the tetragonal to cubic structural transition temperature ($T_c$) of STO is contributed by the electric field-assisted thermal escape of electrons from the quantum well, and the clustering of oxygen vacancies (OVs) as well. We observe an additional source of trapping below the $T_c$, which arises from the trapping of free carriers at the ferroelastic twin walls of STO. Application of -ve $V_g$ results in a charge detrapping, which vanishes above $T_c$ also. This feature demonstrates the crucial role of structural domain walls in the electrical transport properties of STO based heterostructures. The number of trapped (detrapped) charges at (from) the twin wall is controlled by the net polarity of the wall and is completely reversible with the sweep of $V_g$., Comment: 5 Figures

The quantum spin liquid (QSL) is an exotic phase of magnetic materials where the spins continue to fluctuate without any symmetry breaking down to zero temperature. Among the handful reports of QSL with spin $S\ge$1, examples with magnetic ions on a three-dimensional magnetic lattice are extremely rare since both larger spin and higher dimension tend to suppress quantum fluctuations. In this work, we offer a new strategy to achieve 3-D QSL with high spin by utilizing two types of transition metal ions, both are magnetically active but located at crystallographically inequivalent positions. We design a 3-D magnetic system Ba$_3$NiIr$_2$O$_9$ consisting of interconnected corner shared NiO$_6$ octahedra and face shared Ir$_2$O$_9$ dimer, both having triangular arrangements in \textit{a-b} plane. X-ray absorption spectroscopy measurements confirm the presence of Ni$^{2+}$ ($S$=1). Our detailed thermodynamic and magnetic measurements reveal that this compound is a realization of gapless QSL state down to at least 100 mK. Ab-initio calculations find a strong magnetic exchange between Ir and Ni sublattices and in-plane antiferromagnetic coupling between the dimers, resulting in dynamically fluctuating magnetic moments on both the Ir and Ni sublattice., Comment: 4 Figures

The rapid advancement of technology in online communication via social media platforms has led to a prolific rise in the spread of misinformation and fake news. Fake news is especially rampant in the current COVID-19 pandemic, leading to people believing in false and potentially harmful claims and stories. Detecting fake news quickly can alleviate the spread of panic, chaos and potential health hazards. We developed a two stage automated pipeline for COVID-19 fake news detection using state of the art machine learning models for natural language processing. The first model leverages a novel fact checking algorithm that retrieves the most relevant facts concerning user claims about particular COVID-19 claims. The second model verifies the level of truth in the claim by computing the textual entailment between the claim and the true facts retrieved from a manually curated COVID-19 dataset. The dataset is based on a publicly available knowledge source consisting of more than 5000 COVID-19 false claims and verified explanations, a subset of which was internally annotated and cross-validated to train and evaluate our models. We evaluate a series of models based on classical text-based features to more contextual Transformer based models and observe that a model pipeline based on BERT and ALBERT for the two stages respectively yields the best results.

The discovery of skyrmions has sparked tremendous interests about topologically nontrivial spin textures in recent times. The signature of noncoplanar nature of magnetic moments can be observed as topological Hall effect (THE) in electrical measurement. Realization of such nontrivial spin textures in new materials and through new routes is an ongoing endeavour due to their huge potential for future ultra-dense low-power memory applications. In this work, we report oxygen vacancy (OV) induced THE and anomalous Hall effect (AHE) in a 5$d^0$ system KTaO$_3$. The observation of weak antilocalization behavior and THE in the same temperature range strongly implies the crucial role of spin-orbit coupling (SOC) behind the origin of THE. Ab initio calculations reveal the formation of the magnetic moment on Ta atoms around the OV and Rashba-type spin texturing of conduction electrons. In the presence of Rashba SOC, the local moments around vacancy can form bound magnetic polarons (BMP) with noncollinear spin texture, resulting THE. Scaling analysis between transverse and longitudinal resistance establishes skew scattering driven AHE in present case. Our study opens a route to realize topological phenomena through defect engineering., Comment: Published in Advanced Quantum Technology

Thermal plumes are the energy containing eddy motions that carry heat and momentum in a convective boundary layer. The detailed understanding of their structure is of fundamental interest for a range of applications, from wall-bounded engineering flows to quantifying surface-atmosphere flux exchanges. We address the aspect of Reynolds stress anisotropy associated with the intermittent nature of heat transport in thermal plumes by performing an invariant analysis of the Reynolds stress tensor in an unstable atmospheric surface layer flow, using a field-experimental dataset. Given the intermittent and asymmetric nature of the turbulent heat flux, we formulate this problem in an event-based framework. In this approach, we provide structural descriptions of warm-updraft and cold-downdraft events and investigate the degree of isotropy of the Reynolds stress tensor within these events of different sizes. We discover that only a subset of these events are associated with the least anisotropic turbulence in highly-convective conditions. Additionally, intermittent large heat flux events are found to contribute substantially to turbulence anisotropy under unstable stratification. Moreover, we find that the sizes related to the maximum value of the degree of isotropy do not correspond to the peak positions of the heat flux distributions. This is because, the vertical velocity fluctuations pertaining to the sizes associated with the maximum heat flux, transport significant amount of streamwise momentum. A preliminary investigation shows that the sizes of the least anisotropic events probably scale with a mixed-length scale ($z^{0.5}\lambda^{0.5}$, where $z$ is the measurement height and $\lambda$ is the large-eddy length scale).

High entropy oxides (HEOs) are a class of materials, containing equimolar portions of five or more transition metal and/or rare-earth elements. We report here about the layer-by-layer growth of HEO [(La$_{0.2}$Pr$_{0.2}$Nd$_{0.2}$Sm$_{0.2}$Eu$_{0.2}$)NiO$_3$] thin films on NdGaO$_3$ substrates by pulsed laser deposition. The combined characterizations with in-situ reflection high energy electron diffraction, atomic force microscopy, and X-ray diffraction affirm the single crystalline nature of the film with smooth surface morphology. The desired +3 oxidation of Ni has been confirmed by an element sensitive X-ray absorption spectroscopy measurement. Temperature dependent electrical transport measurements revealed a first order metal-insulator transition with the transition temperature very similar to the undoped NdNiO$_3$. Since both of these systems have a comparable tolerance factor, this work demonstrates that the electronic behaviors of $A$-site disordered perovskite-HEOs are primarily controlled by the average tolerance factor., Comment: 4 figures

Depression has been the leading cause of mental-health illness worldwide. Major depressive disorder (MDD), is a common mental health disorder that affects both psychologically as well as physically which could lead to loss of lives. Due to the lack of diagnostic tests and subjectivity involved in detecting depression, there is a growing interest in using behavioural cues to automate depression diagnosis and stage prediction. The absence of labelled behavioural datasets for such problems and the huge amount of variations possible in behaviour makes the problem more challenging. This paper presents a novel multi-level attention based network for multi-modal depression prediction that fuses features from audio, video and text modalities while learning the intra and inter modality relevance. The multi-level attention reinforces overall learning by selecting the most influential features within each modality for the decision making. We perform exhaustive experimentation to create different regression models for audio, video and text modalities. Several fusions models with different configurations are constructed to understand the impact of each feature and modality. We outperform the current baseline by 17.52% in terms of root mean squared error., Comment: in Proceedings of the 9th International Workshop on Audio/Visual Emotion Challenge, AVEC 2019, ACM Multimedia Workshop, Nice, France

The metal-insulator transition (MIT) of bulk rare-earth nickelates is accompanied by a simultaneous charge ordering (CO) transition. We have investigated low-frequency resistance fluctuations (noise) across the MIT and magnetic transition of [EuNiO$_3$/LaNiO$_3$] superlattices, where selective suppression of charge ordering has been achieved by mismatching the superlattice periodicity with the periodicity of charge ordering. We have observed that irrespective of the presence/absence of long-range CO, the noise magnitude is enhanced by several orders with strong non-1/$f$ ($f$ = frequency) component when the system undergoes MIT and magnetic transition. The higher order statistics of resistance fluctuations reveal the presence of strong non-Gaussian components in both cases, further indicating inhomogeneous electrical transport arising from the electronic phase separation. Specifically, we find almost three orders of magnitude smaller noise in the insulating phase of the sample without long-range CO compared to the sample with CO. These findings suggest that digital synthesis can be a potential route to implement electronic transitions of complex oxides for device application., Comment: Accepted in Phys. Rev. B

It is well known that a determined adversary can fool a neural network by making imperceptible adversarial perturbations to an image. Recent studies have shown that these perturbations can be detected even without information about the neural network if the strategy taken by the adversary is known beforehand. Unfortunately, these studies suffer from the generalization limitation -- the detection method has to be recalibrated every time the adversary changes his strategy. In this study, we attempt to overcome the generalization limitation by deriving a metric which reliably identifies adversarial images even when the approach taken by the adversary is unknown. Our metric leverages key differences between the spectra of clean and adversarial images when an image is treated as a matrix. Our metric is able to detect adversarial images across different datasets and attack strategies without any additional re-calibration. In addition, our approach provides geometric insights into several unanswered questions about adversarial perturbations.

A proper initialization of the weights in a neural network is critical to its convergence. Current insights into weight initialization come primarily from linear activation functions. In this paper, I develop a theory for weight initializations with non-linear activations. First, I derive a general weight initialization strategy for any neural network using activation functions differentiable at 0. Next, I derive the weight initialization strategy for the Rectified Linear Unit (RELU), and provide theoretical insights into why the Xavier initialization is a poor choice with RELU activations. My analysis provides a clear demonstration of the role of non-linearities in determining the proper weight initializations., Comment: 9 pages, 4 figures