arXiv subjects

Sayantari Ghosh

Publications and source records attributed to Sayantari Ghosh.

At least 19 recordsLinked to original sources

arXiv2026

Teacher Knows It Best: Spontaneous Symmetry Breaking and Tipping Points in Networked Langevin Dynamics AI Sycophancy

We formulate a statistical physics framework to model a networked stochastic dynamical system exhibiting bistability, driven by additive noise and social conformity. We apply this model to understand and mitigate AI-induced delusional spiraling-a phenomenon where algorithmic sycophancy from Large Language Models continuously reinforces inaccurate beliefs within a socially interacting society. By partitioning the network into a majority of regular agents and a minority of "aware" nodes (Teachers) placed at topological hubs, we use a degree-weighted mean-field approximation to reduce high-dimensional coupled Langevin equations into a single macroscopic drift equation. We provide a closed-form analytical derivation for the deterministic critical tipping time through a saddle-node bifurcation. We validate this analytical boundary using finite-size scaling and demonstrate a universal data collapse across diverse network topologies. Finally, we optimize an intervention strategy under a strict budget constraint that balances the topological footprint against driving velocity. We prove mathematically that under certain conditions, a highly concentrated, rapid intervention targeting massive hubs strictly outperforms a distributed, slow approach to rescue the network.

physics.soc-ph↗

arXiv2026

Escape from Delusional Echo Trap: Symmetry Breaking, Stochastic Dynamics and Mathematical Mitigation Strategies for Algorithmic Sycophancy

We propose a rigorous and systematic mathematical framework for tracking the cognitive trajectories of a user, in the context of algorithmic sycophancy and AI-driven delusional spiraling. Using tools from dynamical systems theory and stochastic differential equations, we explore how individuals perceive, interpret, and update their beliefs as they interact with AI chatbots that possess hidden traits of sycophancy. We treat the evolving conviction as a continuous log-odds state variable, coupled into a stochastic differential equation, navigating a multi-valley potential energy landscape. Our analysis reveals several critical observations governing the stability and rigidity of belief dynamics. We demonstrate that the baseline prior perception of the individual is systematically enhanced by sycophantic feedback beyond a critical threshold. Here, the perceptual potential landscape undergoes a structural phase transition that severely deepens any incremental initial tilt present in the baseline state, transforming the landscape and giving rise to deep, highly resilient attractor basins that trap the individual in unshakeable, self-reinforcing, delusional convictions. Finally, we demonstrate that genuine external information can successfully challenge these rigid states. If this incoming evidence is strong and authentic enough to overcome the internal feedback barrier, it can correct the structural asymmetry caused by sycophancy, inducing a perception reversal that successfully restores the objective belief state.

cs.AI↗

arXiv2026

SAGE-GAN: Towards Realistic and Robust Segmentation of Spatially Ordered Nanoparticles via Attention-Guided GANs

Precise analysis of nanoparticles for characterization in electron microscopy images is essential for advancing nanomaterial development. Yet it remains challenging due to the time-consuming nature of manual methods and the shortcomings of traditional automated segmentation techniques, especially when dealing with complex shapes and imaging artifacts. While conventional methods yield promising results, they depend on a large volume of labeled training data, which is both difficult to acquire and highly time-consuming to generate. In order to overcome these challenges, we have developed a two-step solution: Firstly, our system learns to segment the key features of nanoparticles from a dataset of real images using a self-attention driven U-Net architecture that focuses on important physical and morphological details while ignoring background features and noise. Secondly, this trained Attention U-Net is embedded in a cycle-consistent generative adversarial network (CycleGAN) framework, inspired by the cGAN-Seg model introduced by Abzargar et al. This integration allows for the creation of highly realistic synthetic electron microscopy image-mask pairs that naturally reflect the structural patterns learned by the Attention U-Net. Consequently, the model can accurately detect features in a diverse array of real-world nanoparticle images and autonomously augment the training dataset without requiring human input. Cycle consistency enforces a direct correspondence between synthetic images and ground-truth masks, ensuring realistic features, which is crucial for accurate segmentation training.

cs.CV↗

arXiv2026

Bistability to Quad-stability: Emergence of Hybrid Phenotypes & Enhanced Spatio-temporal Plasticity in Presence of Host-Circuit Coupling

In the context of multistability driven diseases, like cancer, spatiotemporal plasticity plays a significant role to achieve a spectrum of phenotypic variations. The interplay between gene regulatory networks and environmental factors, such as resource competition and spatial diffusion, plays a crucial role in determining cellular behaviour and phenotypic heterogeneity. Though reaction diffusion frameworks have been widely applied in developmental biology, less attention has been paid to the simultaneous effects of resource competition and growth feedback on spatial organization. In this paper, we observed that a bistable genetic circuit under high resource competition due to growth feedback gives rise to multiple emergent phenotypes, as observed in cancer systems. Furthermore, we observed how spatial diffusion coupled with intrinsic nonlinearity can drive the emergence of distinct spatial dynamics over time. The observed spatiotemporal plasticity can also be driven by the comparative stability of the fixed points, diffusivity, and asymmetry of diffusion. Our findings highlight that growth-induced resource competition combined with diffusion can provide deeper insights into metastasis and cancer progression.

q-bio.PE↗

arXiv2026

Bridging the Applicator Gap with Data-Doping:Dual-Domain Learning for Precise Bladder Segmentation in CT-Guided Brachytherapy

Performance degradation due to covariate shift remains a major challenge for deep learning models in medical image segmentation. An open question is whether samples from a shifted distribution can effectively support learning when combined with limited target domain data. We investigate this problem in the context of bladder segmentation in CT guided gynecological brachytherapy, a critical task for accurate dose optimization and organ at risk sparing. While CT scans without brachytherapy applicators (no applicator: NA) are widely available, scans with applicators inserted (with applicator: WA) are scarce and exhibit substantial anatomical deformation and imaging artifacts, making automated segmentation particularly difficult. We propose a dual domain learning strategy that integrates NA and WA CT data to improve robustness and generalizability under covariate shift. Using a curated assorted dataset, we show that NA data alone fail to capture the anatomical and artifact related characteristics of WA images. However, introducing a modest proportion of WA data into a predominantly NA training set leads to significant performance improvements. Through systematic experiments across axial, coronal, and sagittal planes using multiple deep learning architectures, we demonstrate that doping only 10 to 30 percent WA data achieves segmentation performance comparable to models trained exclusively on WA data. The proposed approach attains Dice similarity coefficients of up to 0.94 and Intersection over Union scores of up to 0.92, indicating effective domain adaptation and improved clinical reliability. This study highlights the value of integrating anatomically similar but distribution shifted datasets to overcome data scarcity and enhance deep learning based segmentation for brachytherapy treatment planning.

cs.CV↗

arXiv2025

F-ANcGAN: An Attention-Enhanced Cycle Consistent Generative Adversarial Architecture for Synthetic Image Generation of Nanoparticles

Nanomaterial research is becoming a vital area for energy, medicine, and materials science, and accurate analysis of the nanoparticle topology is essential to determine their properties. Unfortunately, the lack of high-quality annotated datasets drastically hinders the creation of strong segmentation models for nanoscale imaging. To alleviate this problem, we introduce F-ANcGAN, an attention-enhanced cycle consistent generative adversarial system that can be trained using a limited number of data samples and generates realistic scanning electron microscopy (SEM) images directly from segmentation maps. Our model uses a Style U-Net generator and a U-Net segmentation network equipped with self-attention to capture structural relationships and applies augmentation methods to increase the variety of the dataset. The architecture reached a raw FID score of 17.65 for TiO$_2$ dataset generation, with a further reduction in FID score to nearly 10.39 by using efficient post-processing techniques. By facilitating scalable high-fidelity synthetic dataset generation, our approach can improve the effectiveness of downstream segmentation task training, overcoming severe data shortage issues in nanoparticle analysis, thus extending its applications to resource-limited fields.

cs.CV↗

arXiv2025

Federated Self-Supervised Learning for One-Shot Cross-Modal and Cross-Imaging Technique Segmentation

Decentralized federated learning enables learning of data representations from multiple sources without compromising the privacy of the clients. In applications like medical image segmentation, where obtaining a large annotated dataset from a single source is a distressing problem, federated self-supervised learning can provide some solace. In this work, we push the limits further by exploring a federated self-supervised one-shot segmentation task representing a more data-scarce scenario. We adopt a pre-existing self-supervised few-shot segmentation framework CoWPro and adapt it to the federated learning scenario. To the best of our knowledge, this work is the first to attempt a self-supervised few-shot segmentation task in the federated learning domain. Moreover, we consider the clients to be constituted of data from different modalities and imaging techniques like MR or CT, which makes the problem even harder. Additionally, we reinforce and improve the baseline CoWPro method using a fused dice loss which shows considerable improvement in performance over the baseline CoWPro. Finally, we evaluate this novel framework on a completely unseen held-out part of the local client dataset. We observe that the proposed framework can achieve performance at par or better than the FedAvg version of the CoWPro framework on the held-out validation dataset.

cs.CV↗

arXiv2024

Sustainability & Social Segmentation in Social Media Contagion: A Mathematical and Computational Study on Dual Effects of Individual Needs & Peer Influence

Addiction to internet-based social media has increasingly emerged as a critical social problem, especially among young adults and teenagers. Based on multiple research studies, excessive usage of social media may have detrimental psychological and physical impacts. In this study, we are going to explore mathematically the dynamics of social media addiction behaviour and explore the determinants of compulsive use of social media from the dual perspectives of individual needs or cravings and peer-related factors or peer pressure. The theoretical analysis of the model without the peer pressure effect reveals that the associated addiction-free equilibrium is globally stable whenever a certain threshold, known as the addictive-generation number, is less than unity and unstable when the threshold is greater than unity. We observed how introduction of peer influence adds a sustainability to the dynamics, and causes a multistability, through which addiction-contagion can proliferate, even below the designated critical threshold. Using simulations over model networks, we demonstrate our finding, even in the presence of social heterogeneity. Finally, we use the reaction-diffusion approach to investigate spatio-temporal dynamics in a synthetic society, in the form of a 2D lattice. Instead of a fast convergence to the steady states, we observe a long transient of social clustering and segmentation, represented by spatio-temporal pattern formation. Our model illustrates how the peer influence factor plays a crucial role and concludes that it is required to consider the peer factors while formulating specific strategies that could be more effective against this addiction and its potential adverse outcomes.

physics.soc-ph↗

arXiv2024

Resource competition in Three-gene-motif & Emergence of Feed-forward response: A Spatiotemporal Study

Feed-forward dynamics, which is well-known to have several important implications in nonlinear dynamical systems, frequently occurs in gene expression motifs, and has been well explored experimentally and mathematically. However, dependency of the components of a genetic circuit upon its host, due to the requirement for resources like ribosome, ATP, transcription factors, tRNA, etc., and related effects are of utmost importance, which is commonly ignored in mathematical models. In a resource-limited environment, two apparently unconnected genes can compete for resources for their respective expression and may exhibit indirect regulatory connection; an emergent response thus arises in the system completely because of resource competition. In this work, we have shown how the responses of the feed-forward loop (FFL), a well-studied regulatory genetic motif, can be recreated considering the resource competition in a three-gene pathway. Exploring the genetic system with temporal as well as spatiotemporal stability analysis, interesting transient and steady-state responses have been observed. The genetic motifs explored in this paper show many of the characteristic features of the conventional FFL structure, like response delay and pulse generation. Most interestingly, in a two-dimensional cellular arrangement, characteristic pattern formation under a concentration gradient of input signals have also been observed. This study pinpoints a larger area of research and exploration in synthetic and cellular systems, which will reveal novel controlling ideas and unique behavioral changes in the system for its context dependencies.

q-bio.MN↗

arXiv2023

Quantitative Analysis of Social Influence & Digital Piracy Contagion with Differential Equations on Networks

Though the studies of social contagions are regularly borrowing network models to study the propagation of social influences and opinions to include social heterogeneity. Such studies provide valuable insights regarding these, but the social network structures cannot be well explored in their study. In this research, we methodically study the trends in online piracy with a continuous ODE approach and differential equations on graphs, to have a clear comparative view. We first formulate a compartmental model to mathematically study bifurcations and thresholds, and later move on with a network-based analysis to illustrate the proliferation of online piracy dynamic with an epidemiological approach over a social network. We figure out a solution for this online piracy problem by developing awareness among individuals by introducing media campaigns which could be a useful factor for the eradication and control of online piracy. Next, using degree-block approximation, network analysis has been performed to investigate the phenomena from a heterogeneous approach and to derive the threshold condition for the persistence of piracy in the population in a steady state. Based on the behavioral responses of individuals in a society due to the effect of media, we examine the system through the aid of realistic parameter selection to better understand the complexity of the dynamics and propose control strategies.

physics.soc-ph↗

arXiv2023

Visual Representation for Patterned Proliferation of Social Media Addiction: Quantitative Model and Network Analysis

With the advancement of information technology, more people, especially young adults, are getting addicted to the use of different social media platforms. Despite immense useful applications in communication and interactions, the habit of spending excessive time on these social media platforms is becoming addictive, causing different consequences, like anxiety, depression, health problems, and many more. Here, we mathematically explored a model of social media addiction, including a peer-influence relapse. We have further done the complex network analysis for a heterogenic synthetic society. Finally, we explore spatiotemporal pattern formation in a diffusive social system using the reaction-diffusion approach. Our model shows how the existent nonlinearity in the system makes it difficult to make society social media addiction free once it crosses a certain threshold. Some possible strategies are explored mathematically to prevent social media addiction, and the importance of peer-influenced relapse has been identified as a major barrier.

physics.soc-ph↗

arXiv2023

Quantitative Modelling of Diffusion-driven Pattern Formation in microRNA-regulated Gene Expression

MicroRNAs are extensively known for post-transcriptional gene regulation and pattern formation in the embryonic developmental stage. We explore the origin of these spatio-temporal patterns mathematically, considering three different motifs here. For three scenarios, (1) simple microRNA-based mRNA regulation with a graded response in output, (2) microRNA-based mRNA regulation resulting in bistability in the dynamics, and (3) a coordinated response of microRNA (miRNA), simultaneously regulating the mRNAs of two different pools, detailed dynamical analysis, as well as the reaction-diffusion scenario have been considered and analyzed in the steady state and for the transient dynamics further. We have observed persistent-temporal patterns, as a result of the dynamics of the motifs, that explain spatial gradients and relevant patterns formed by related proteins in development and phenotypic heterogenetic aspects in biological systems. Competitive effects of miRNA regulation have also been found to be capable to cause spatio-temporal patterns, persistent enough to direct developmental decisions. Under coordinated regulation, miRNAs are found to generate spatio-temporal patterning even from complete homogeneity in concentration of target protein, which may have impactful insights in choice of cell-fates.

q-bio.QM↗

arXiv2023

Hesitancy, Awareness and Vaccination: A Computational Analysis on Complex Networks

Considering the global pandemic of coronavirus disease 2019 (COVID-19), around the world several vaccines are being developed. Till now, these vaccines are the most effective way to reduce the high burden on the global health infrastructure. However, the public acceptance towards vaccination is a crucial and pressing problem for health authorities. This study has been designed to determine the parameters affecting the decisions of common individuals towards COVID-19 vaccine. In our study, using the platforms of compartmental model and network simulation, we categorize people and observe their motivation towards vaccination in a mathematical social contagion process. In our model, we consider peer influence as an important factor in this dynamics, and study how individuals are influencing each other for vaccination. The efficiency of the vaccination process is estimated by the period of time required to vaccinate a substantial fraction of total population. We discovered the major barriers and drivers of this dynamics, and concluded that it is required to formulate specific strategies by the healthcare workers which could be more effective for the undecided and vaccine hesitant group of people.

physics.soc-ph↗

arXiv2022

Nanomaterials for Supercapacitors: Uncovering Research Themes with Unsupervised Machine Learning

Identification of important topics in a text can facilitate knowledge curation, discover thematic trends, and predict future directions. In this paper, we aim to quantitatively detect the most common research themes in the emerging supercapacitor research area, and summarize their trends and characteristics through the proposed unsupervised, machine learning approach. We have retrieved the complete reference entries of article abstracts from Scopus database for all original research articles from 2004 to 2021. Abstracts were processed through a natural language processing pipeline and analyzed by a latent Dirichlet allocation topic modeling algorithm for unsupervised topic discovery. Nine major topics were further examined through topic-word associations, Inter-topic distance map and topic-specific word cloud. We observed the greatest importance is being given to performance metrics (28.2%), flexible electronics (8%), and graphene-based nanocomposites (10.9%). The analysis also points out crucial future research directions towards bio-derived carbon nanomaterials (such as RGO) and flexible supercapacitors.

cond-mat.stat-mech↗

arXiv2022

To Jab or Not to Jab? A Study on COVID-19 Vaccine Hesitancy in India

With a country-wide comprehensive internet survey conducted in India, we aim to determine the factors that drive hesitancy towards getting vaccinated for COVID-19, and also compare their levels of influence. The perceived reliability and effectiveness of available vaccines turn out to be important drivers in lowering vaccine hesitancy. Additionally, higher hesitancy is associated with being of a younger age or having lower education. Furthermore, comparing vaccine attitudes from observations before a major COVID-19 wave with those from after, we find that the latter are significantly less hesitant about getting the vaccine. We also find that in addition to the standard knowledge and awareness campaigns, local level peer influences are important factors that affect vaccine hesitancy. Finally, we use statistically significant estimates from logistic regression on our survey data in a synthetic heterogeneous complex network-based society, to extrapolate scenarios that may arise from the dynamic interactions between our variables of interest. We use outcomes from this simulated society to suggest strategic interventions that may lower vaccine hesitancy.

physics.soc-ph↗

arXiv2022

Spatio-temporal Pattern Formation due to Host-Circuit Interplay in Gene Expression Dynamics

Biological systems are majorly dependent on their property of bistability in order to exhibit nongenetic heterogeneity in terms of cellular morphology and physiology. Spatial patterns of phenotypically heterogeneous cells, arising due to underlying bistability, may play significant role in phenomena like biofilm development, adaptation, cell motility etc. While nonlinear positive feedback regulation, like cooperative heterodimer formation are the usual reason behind bistability, similar dynamics can also occur as a consequence of host-circuit interaction. In this paper, we have investigated the pattern formation by a motif with non-cooperative positive feedback, that imposes a metabolic burden on its host due to its expression. In a cellular array set inside diffusible environment, we investigate spatio-temporal diffusion in one dimension as well as in two dimension in the context of various initial conditions respectively. Moreover, the number of cells exhibiting the same steady state, as well as their spatial distribution has been quantified in terms of connected component analysis. The effect of diffusion coefficient variation has been studied in terms of stability of related states and time evolution of patterns.

q-bio.QM↗

arXiv2022

Attention W-Net: Improved Skip Connections for better Representations

Segmentation of macro and microvascular structures in fundoscopic retinal images plays a crucial role in the detection of multiple retinal and systemic diseases, yet it is a difficult problem to solve. Most neural network approaches face several issues such as lack of enough parameters, overfitting and/or incompatibility between internal feature-spaces. We propose Attention W-Net, a new U-Net based architecture for retinal vessel segmentation to address these problems. In this architecture, we have two main contributions: Attention Block and regularisation measures. Our Attention Block uses attention between encoder and decoder features, resulting in higher compatibility upon addition. Our regularisation measures include augmentation and modifications to the ResNet Block used, which greatly prevent overfitting. We observe an F1 and AUC of 0.8407 and 0.9833 on the DRIVE and 0.8174 and 0.9865 respectively on the CHASE-DB1 datasets - a sizeable improvement over its backbone as well as competitive performance among contemporary state-of-the-art methods.

eess.IV↗

arXiv2022

Resource allocation determines alternate cell fate in Bistable Genetic Switch

Living cells need a constant availability of certain resources to have a sustained gene expression process. Limited availability of cellular resources for gene expression, like ribosomes, along with a variation of resource affinity, significantly modifies the system dynamics. Factors like the variation in rate of binding, or variation in efficiency of the recruited resource have the potential to affect crucial dynamical phenomena like cell fate determination. In this paper, we have taken a very important motif, a bistable genetic toggle switch, and explored the effect of resource imbalance in this circuit in terms of the bifurcations taking place. We show that initial asymmetric biasing to resource via resource affinity or gene copy number, significantly modifies the cell fate transition, both in pitchfork and saddle node type bifurcation. Our study establishes that in a limited resource environment, controlled resource allocation can be an important factor for robust functioning of the synthetic or cellular genetic switches.

q-bio.QM↗