arXiv subjects

Nitin Kumar

Publications and source records attributed to Nitin Kumar.

At least 19 recordsLinked to original sources

arXiv2026

Asymmetry-Induced Chiral Dynamics in Coupled Self-Propelled Robots: Spinning and Circular Motion

Motivated by the chiral motility of microswimmers, we investigate how geometric asymmetry in a system of two self-propelled active Brownian robots coupled by a spring gives rise to rich collective dynamics. We demonstrate that asymmetry in the propulsion directions of the robots generates net torques that induce persistent rotational motion. Depending on the choice of propulsion angles $\alpha_1$ and $\alpha_2$, the system exhibits three distinct dynamical regimes -- run-and-tumble motion, circular trajectories, and spinning -- with the geometric configuration primarily determining the realized regime. We further show that spring stiffness and rotational noise act as additional tuning parameters governing the stability of these regimes. These results demonstrate how the interplay of mechanical coupling and activity produces diverse self-organized dynamics in simple robotic dimers, providing a bridge between artificial active systems and biological microswimmers such as bacteria, Chlamydomonas reinhardtii, and spermatozoa.

cond-mat.soft↗

arXiv2026

When Context Misleads: Surprisal, Energy and Attention Entropy as Metrics of Coherence Illusions in LLMs

Psycholinguistics studies show that human readers fall for coherence illusions: an incoherent discourse can seem coherent simply because a distractor matches what comes next. We investigate whether Dutch language models (6 monolingual and 4 multilingual) show the same behavior on texts that link back to earlier context with words such as 'again' and 'too'. First, we find that surprisal at the critical word tracks human acceptability judgments and eye-tracking data. Models are more surprised by incoherent continuations, but a matching distractor in the prior context reduces this surprisal. Second, attention entropy at the critical position identifies heads that behave differently under coherence vs. incoherence. We find that ablating these heads shows transfer effects across experiments, suggesting a shared mechanism. Third, we introduce energy from the associative-memory literature as a metric to quantify discourse coherence. Taken together, our results show that coherence illusions arise in Dutch LLMs, with entropy and energy exposing mechanisms that operate across settings.

cs.CL↗

arXiv2026

ARMA-C3: A Contrastive ARMA Convolutional Framework for Unsupervised and Semi-supervised Classification

In biomedical and neurodegenerative disorders, accurate and early disease identification remains challenging due to the scarcity of labeled data and the complexity of imaging patterns. To address these challenges, we introduce ARMA-C3, a unified unsupervised and semi-supervised graph learning framework for node classification based on contrastive learning and graph-cut regularization to learn structurally meaningful and discriminative representations. By modeling samples or images as graph nodes and exploiting inter-sample relationships, the proposed framework captures subject-level dependencies that conventional machine learning methods typically overlook. We conduct extensive binary classification experiments across five clinically relevant datasets: the Alzheimer's Disease Neuroimaging Initiative (ADNI), the Neuroimaging in Frontotemporal Dementia (NIFD) dataset, and three medical imaging benchmarks (BreastMNIST, PneumoniaMNIST, and a liver ultrasound dataset). Experimental results demonstrate that ARMA-C3 achieves competitive and frequently superior performance compared to classical clustering techniques, state-of-the-art machine learning models, and existing graph-based deep learning approaches across multiple evaluation settings, particularly under limited supervision and severe class imbalance. The proposed framework further demonstrates robust representation learning and strong cross-modal generalization across diverse biomedical imaging modalities.

cs.CV↗

arXiv2026

DMDSC: A Dynamic-Margin Deep Simplex Classifier for Open-Set Recognition on Medical Image Datasets

Medical imaging datasets are often characterized by extreme class imbalances, where rare pathologies are significantly underrepresented compared to common conditions. This imbalance poses a dual challenge for Open-Set Recognition (OSR): models must maintain high classification accuracy on known classes while reliably rejecting unknown samples unseen during training in the clinical settings. While recently proposed Deep Simplex Classifier (DSC)~\cite{cevikalp2024reaching} and UnCertainty-aware Deep Simplex Classifier (UCDSC)~\cite{Aditya_2026_WACV} successfully leverage Neural Collapse to ensure maximal inter-class separation, they rely on a uniform margin that does not account for the varying densities of medical classes. In this paper, we propose DMDSC an enhanced framework featuring a dynamic margin approach. Our approach automatically adapts class-specific margins based on label frequency, enforcing a higher penalty and tighter feature clustering for rare pathologies to counteract the effects of data imbalance. Extensive experiments conducted on diverse medical benchmarks on BloodMNIST\cite{medmnistv2}, OCTMNIST\cite{medmnistv2}, DermaMNIST\cite{medmnistv2}, and BreaKHis~\cite{spanhol2015dataset} datasets, demonstrate that our framework outperforms state-of-the-art methods.

cs.CV↗

arXiv2026

BiMol-Diff: A Unified Diffusion Framework for Molecular Generation and Captioning

Bridging molecular structures and natural language is essential for controllable design. Autoregressive models struggle with long-range dependencies, while standard diffusion processes apply uniform corruption across positions, which can distort structurally informative tokens. We present BiMol-Diff, a unified diffusion framework for the paired tasks of text-conditioned molecule generation and molecule captioning. Our key component is a token-aware noise schedule that assigns position-dependent corruption based on token recovery difficulty, preserving harder-to-recover substructures during the forward process. On ChEBI-20 and M3-20M, BiMol-Diff improves molecule reconstruction with a 15.4% relative gain in Exact Match and achieves strong captioning results, attaining best BLEU and BERTScore among compared baselines. These results indicate token-aware noising improves fidelity in molecular structure-language modelling.

cs.CL↗

arXiv2026

An active soft condensed matter approach to the Physics of living systems

This article aims to introduce the broad field of soft active matter physics and its relevance to the life sciences in simple, accessible language. Although this area of research is relatively new, it has already demonstrated significant potential in providing a physical understanding of many biological processes. While several review articles by leading researchers exist, they can be difficult to grasp for undergraduate students and even early-career researchers who wish to enter this field. In this article, I cover the basics, introduce the origins of soft active matter physics, and explain how it differs from traditional equilibrium condensed matter ideas at the fundamental level. For the most part, I will avoid mathematical equations and excessive technical precision in several statements. Instead, I will focus on communicating the core ideas and the overall spirit of the argument, using everyday examples to develop a physical intuition. The primary focus will be on the dynamical aspects of these systems. I will conclude by briefly discussing a published experimental study from our research group that examines universal features of the trajectories of homing and migrating organisms.

cond-mat.soft↗

arXiv2026

Bifurcation curve detection with deflation for multiparametric PDEs

This work presents a comprehensive framework for capturing bifurcating phenomena and detecting bifurcation curves in nonlinear multiparametric partial differential equations, where the system exhibits multiple coexisting solutions for given values of the parameters. Traditional continuation methods for one-dimensional parameterizations employ the previously computed solution as the initial guess for the next parameter value. These are usually very inefficient, since small step sizes increase computational cost, while larger steps could jeopardize the method convergence jumping to a different solution branch or missing the bifurcation point. To address these challenges, we propose a novel framework that combines: (i) arclength continuation, adaptively selecting new parameter values in higher dimension, and (ii) the deflation technique, discovering multiple branches to construct complete bifurcation diagrams without requiring a costly spectral analysis of the system. In particular, the arclength continuation method is designed to handle multiparametric scenarios, where the parameter vector $\lambda \in \mathbb{R}^p$ traces a curve $g(\lambda)$ within a $p$-dimensional parameter space. In addition, we introduce a zigzag path-following strategy to robustly track the bifurcation curves and surfaces, respectively, for two- and three-dimensional parametric spaces. Finally, we demonstrate its performance on three benchmark problems of increasing complexity: from the 1D/2D Bratu and Allen--Cahn equations to the 2D/3D Rayleigh--Benard convection problem.

math.NA↗

arXiv2026

ARMARecon: An ARMA Convolutional Filter based Graph Neural Network for Neurodegenerative Dementias Classification

Early detection of neurodegenerative diseases such as Alzheimer's Disease (AD) and Frontotemporal Dementia (FTD) is essential for reducing the risk of progression to severe disease stages. As AD and FTD propagate along white-matter regions in a global, graph-dependent manner, graph-based neural networks are well suited to capture these patterns. Hence, we introduce ARMARecon, a unified graph learning framework that integrates Autoregressive Moving Average (ARMA) graph filtering with a reconstruction-driven objective to enhance feature representation and improve classification accuracy. ARMARecon effectively models both local and global connectivity by leveraging 20-bin Fractional Anisotropy (FA) histogram features extracted from white-matter regions, while mitigating over-smoothing. Overall, ARMARecon achieves superior performance compared to state-of-the-art methods on the multi-site dMRI datasets ADNI and NIFD.

cs.CV↗

arXiv2025

UCDSC: Open Set UnCertainty aware Deep Simplex Classifier for Medical Image Datasets

Driven by advancements in deep learning, computer-aided diagnoses have made remarkable progress. However, outside controlled laboratory settings, algorithms may encounter several challenges. In the medical domain, these difficulties often stem from limited data availability due to ethical and legal restrictions, as well as the high cost and time required for expert annotations-especially in the face of emerging or rare diseases. In this context, open-set recognition plays a vital role by identifying whether a sample belongs to one of the known classes seen during training or should be rejected as an unknown. Recent studies have shown that features learned in the later stages of deep neural networks are observed to cluster around their class means, which themselves are arranged as individual vertices of a regular simplex [32]. The proposed method introduces a loss function designed to reject samples of unknown classes effectively by penalizing open space regions using auxiliary datasets. This approach achieves significant performance gain across four MedMNIST datasets-BloodMNIST, OCTMNIST, DermaMNIST, TissueMNIST and a publicly available skin dataset [29] outperforming state-of-the-art techniques.

cs.CV↗

arXiv2025

Low-Frequency Noise and Resistive Switching in $\beta$-Na$_{0.33}$V$_2$O$_5$

The interplay between charge ordering and its manifestation in macroscopic electrical transport in low-dimensional materials is crucial for understanding resistive switching mechanisms. In this study, we investigate the electronic transport and switching behavior of single-crystalline $\beta$-Na$_{0.33}$V$_2$O$_5$, focusing on low-frequency resistance noise dynamics of charge-order-driven resistive switching. Using electrical transport, low frequency noise spectroscopy, and X-ray diffraction, we probe electron dynamics across the Na-ion-ordering (IO) and charge-ordering (CO) transitions. Near room temperature, the weak temperature dependence of the noise spectral density points to a dominance of nearest-neighbor polaron hopping. Below IO transition temperature ($ T_{IO} \sim 240 \, \text{K} $), structural analysis reveals that Na-ions adopt a zig-zag occupancy pattern, breaking the two-fold rotational symmetry and influencing the electronic ground state. Subsequently, a sharp drop in resistance noise below the CO transition temperature ($ T_{CO} \sim 125 \, \text{K} $) indicates the emergence of correlated electron behavior. Furthermore, application of sufficient electric field leads to the destabilization of the CO state, and a transition to a high-conducting state. The material exhibits distinct resistive switching between 35~K and 110~K, with a resistance change spanning two orders of magnitude, primarily driven by electronic mechanisms rather than Joule heating. These findings provide new insights into charge-order-induced switching and electronic correlations in quasi-one-dimensional systems, with potential applications in cryogenic memory and neuromorphic computing devices owing to the low noise levels in their stable resistive states.

cond-mat.str-el↗

arXiv2025

Does the Skeleton-Recall Loss Really Work?

Image segmentation is an important and widely performed task in computer vision. Accomplishing effective image segmentation in diverse settings often requires custom model architectures and loss functions. A set of models that specialize in segmenting thin tubular structures are topology preservation-based loss functions. These models often utilize a pixel skeletonization process claimed to generate more precise segmentation masks of thin tubes and better capture the structures that other models often miss. One such model, Skeleton Recall Loss (SRL) proposed by Kirchhoff et al.~\cite {kirchhoff2024srl}, was stated to produce state-of-the-art results on benchmark tubular datasets. In this work, we performed a theoretical analysis of the gradients for the SRL loss. Upon comparing the performance of the proposed method on some of the tubular datasets (used in the original work, along with some additional datasets), we found that the performance of SRL-based segmentation models did not exceed traditional baseline models. By providing both a theoretical explanation and empirical evidence, this work critically evaluates the limitations of topology-based loss functions, offering valuable insights for researchers aiming to develop more effective segmentation models for complex tubular structures.

cs.CV↗

arXiv2025

Magnetic Triple-q State in Antiferromagnetic Monolayer Interfaced with Bismuthene

We have successfully fabricated the bismuthene covered Mn monolayer on Ag(111) by evaporating Mn atoms onto (p x root3)-Bi/Ag(111) at room temperature. By using spin-polarized scanning tunneling microscopy (SP-STM), we have resolved the magnetic triple-q (3Q) state. In combination with density-functional theory (DFT) calculations, the 3Q3-like spin texture is the magnetic ground state for the bismuthene covered Mn monolayer/Ag(111). Interestingly, the uniaxial magnetic anisotropy of 3Q3 state triggered by the bismuthene on top of Mn monolayer/Ag(111) has been revealed, which is consistent with the switching of 3Q3up and 3Q3down domains observed by SP-STM measurements with external magnetic fields.

cond-mat.mtrl-sci↗

arXiv2025

Nematic ordering via vertical stratification in drying clay nanotube suspensions

Evaporative self-assembly offers a simple, cost-effective method for producing functional nanostructured materials. However, achieving tunable and ordered assemblies remains challenging, especially when working with complex building blocks like nanoparticles that exhibit significant shape and size polydispersity. In this study, starting from an aqueous suspension of a polydisperse sample of rod-like Halloysite nanotubes, we present a physical protocol for producing a high degree of orientational ordering in the final dried deposit. By placing a sessile droplet on a substrate heated to 50{\deg}C, self-induced Marangoni flows suppress the coffee-ring effect, enabling more uniform deposition of colloidal rods. Subsequently, the vertical stratification during evaporation leads to the segregation of particles by aspect ratio, with longer rods (aspect ratio>=6.5) preferentially migrating to the top layers over the entire deposit. Since rods exceeding this threshold exhibit nematic ordering at high densities, the resulting top layer, spanning an area of the order of mm^2, displays a high degree of orientational order. Thus, our results highlight a robust strategy for engineering ordered structures from disordered colloidal suspensions despite the overall polydispersity of the system.

cond-mat.soft↗

arXiv2025

Spontaneous Emergence of Run-and-Tumble-Like Dynamics in a Robotic Analog of \textit{Chlamydomonas}: Experiment and Theory

Run-and-tumble (RT) motion is commonly observed in flagellated microswimmers, arising from synchronous and asynchronous flagellar beating. One such example is a biflagellated alga, called \textit{Chlamydomonas reinhardtii}. Its flagellar synchronization is not only affected by hydrodynamic interactions but also through contractile stress fibers that mechanically couple the flagella, enabling adaptable swimming behavior. To explore this, we design a macroscopic mechanical system that comprises dry, self-propelled robots linked by a rigid rod to model this organism. By varying the attachment points of the two ends of the rod on each robot, the model incorporates the effect of fiber contractility observed in the real organism. To mimic a low Reynolds number environment, we program each robot to undergo overdamped active Brownian (AB) motion. We find that such a system exhibits RT-like behavior, characterized by sharp, direction-reversing tumbles and exponentially distributed run times, consistent with the real organism. Moreover, we quantify tumbling frequency and demonstrate its tunability across experimental parameters. Additionally, we provide a theoretical model that reproduces our results, elucidating physical mechanisms governing RT dynamics. Thus, our robotic system not only replicates RT motion but also captures several subtle characteristics, offering valuable insights into the underlying physics of microswimmer motility.

cond-mat.soft↗

arXiv2025

Reconfigurable Filamentary Conduction in Thermally Stable Zeolitic Imidazolate Framework (ZIF-8) Resistive Switching Devices

The rapid growth of digital technology has driven the need for efficient storage solutions, positioning memristors as promising candidates for next-generation non-volatile memory (NVM) due to their superior electrical properties. Organic and inorganic materials each offer distinct advantages for resistive switching (RS) performance, while hybrid materials like metal-organic frameworks (MOFs) combine the strengths of both. In this study, we present a resistive random-access memory (ReRAM) device utilizing zeolitic imidazolate framework (ZIF-8), a MOF material, as the resistive switching layer. The ZIF-8 film was synthesized via a simple solution process method at room temperature and subsequently characterized. The Al/ZIF-8/ITO device demonstrates bipolar resistive switching behaviour with an on/off resistance ratio of 100, stable retention up to 10000 seconds, and consistent performance across 60 cycles while exhibiting robust thermal stability from -20 C to 100 C. Low-frequency noise and impedance spectroscopy measurements suggest a filamentary switching mechanism. Additionally, the memory state can be tuned by adjusting the reset voltage, pointing to potential as multi-level memory. Potentiation and depression experiments further highlight the devices promise for neuromorphic applications. With high stability, tunability, and strong performance, the ZIF-8 based ReRAM shows great promise for advanced NVM and neuromorphic computing applications.

cond-mat.mtrl-sci↗

arXiv2024

Noise Spectroscopy and Electrical Transport in NbO2 Memristors with Dual Resistive Switching

Negative differential resistance (NDR) behavior observed in several transition metal oxides is crucial for developing next-generation memory devices and neuromorphic computing systems. NbO2-based memristors exhibit two regions of NDR at room temperature, making them promising candidates for such applications. Despite this potential, the physical mechanisms behind the onset and the ability to engineer these NDR regions remain unclear, hindering further development of these devices for applications. This study employed electrical transport and ultra-low frequency noise spectroscopy measurements to investigate two distinct NDR phenomena in nanoscale thin films of NbO2. By analyzing the residual current fluctuations as a function of time, we find spatially inhomogeneous and non-linear conduction near NDR-1 and a two-state switching near NDR-2, leading to an insulator-to-metal (IMT) transition. The power spectral density of the residual fluctuations exhibits significantly elevated noise magnitudes around both NDR regions, providing insights into physical mechanisms and device size scaling for electronic applications. A simple theoretical model, based on the dimerization of correlated insulators, offers a comprehensive explanation of observed transport and noise behaviors near NDRs, affirming the presence of non-linear conduction followed by an IMT connecting macroscopic device response to transport signatures at atomic level.

cond-mat.mes-hall↗

arXiv2024

UnSegMedGAT: Unsupervised Medical Image Segmentation using Graph Attention Networks Clustering

The data-intensive nature of supervised classification drives the interest of the researchers towards unsupervised approaches, especially for problems such as medical image segmentation, where labeled data is scarce. Building on the recent advancements of Vision transformers (ViT) in computer vision, we propose an unsupervised segmentation framework using a pre-trained Dino-ViT. In the proposed method, we leverage the inherent graph structure within the image to realize a significant performance gain for segmentation in medical images. For this, we introduce a modularity-based loss function coupled with a Graph Attention Network (GAT) to effectively capture the inherent graph topology within the image. Our method achieves state-of-the-art performance, even significantly surpassing or matching that of existing (semi)supervised technique such as MedSAM which is a Segment Anything Model in medical images. We demonstrate this using two challenging medical image datasets ISIC-2018 and CVC-ColonDB. This work underscores the potential of unsupervised approaches in advancing medical image analysis in scenarios where labeled data is scarce. The github repository of the code is available on [https://github.com/mudit-adityaja/UnSegMedGAT].

cs.CV↗

arXiv2024

UnSeGArmaNet: Unsupervised Image Segmentation using Graph Neural Networks with Convolutional ARMA Filters

The data-hungry approach of supervised classification drives the interest of the researchers toward unsupervised approaches, especially for problems such as medical image segmentation, where labeled data are difficult to get. Motivated by the recent success of Vision transformers (ViT) in various computer vision tasks, we propose an unsupervised segmentation framework with a pre-trained ViT. Moreover, by harnessing the graph structure inherent within the image, the proposed method achieves a notable performance in segmentation, especially in medical images. We further introduce a modularity-based loss function coupled with an Auto-Regressive Moving Average (ARMA) filter to capture the inherent graph topology within the image. Finally, we observe that employing Scaled Exponential Linear Unit (SELU) and SILU (Swish) activation functions within the proposed Graph Neural Network (GNN) architecture enhances the performance of segmentation. The proposed method provides state-of-the-art performance (even comparable to supervised methods) on benchmark image segmentation datasets such as ECSSD, DUTS, and CUB, as well as challenging medical image segmentation datasets such as KVASIR, CVC-ClinicDB, ISIC-2018. The github repository of the code is available on \url{https://github.com/ksgr5566/UnSeGArmaNet}.

cs.CV↗