arXiv subjects

Yash Raj Shrestha

Publications and source records attributed to Yash Raj Shrestha.

At least 19 recordsLinked to original sources

arXiv2026

Beyond Surface Cues: Disentangling Sociocultural Signals in Multilingual LLMs

Multilingual LLM outputs can vary across sociocultural contexts. However, evidence of cultural grounding can be misleading: identity labels may be inferred from explicit or indirect textual cues, while names and wording can reveal the source language. Treating all these signals as evidence of cultural grounding may obscure potential biases. We present a human-validated, multi-agent audit that separates three questions: whether outputs reproduce social biases, whether identity groups are represented differently, and whether outputs reflect cross-cultural patterns. The study analyzes 89,253 outputs from 12 LLMs in English, French, and Chinese, spanning 18 occupations and three task conditions. We find that bias representation varies systematically across languages and tasks. Removing direct identity cues sharply reduces identity-label prediction in English and Chinese, but has a much smaller effect in French. Across all language-genre settings, the cultural context associated with the source language receives the highest average relevance score, with moderate agreement between automated and human ratings. However, the ability to identify the source language drops substantially after translation and again after masking names. Without these controls, multilingual audits may mistake surface cues for cultural understanding, leading to misleading conclusions about cross-cultural variation and bias. Our audit offers a practical framework for separating such shortcuts from more meaningful cross-cultural patterns.

cs.CL↗

arXiv2026

SAGE: An Expert-Annotated South Asian GI Endoscopy Dataset for Multimodal Learning and Hallucination Analysis

Gastrointestinal cancers represent a growing health burden in the South Asian region, driven largely by rapid changes in socio-economic conditions and lifestyle habits. However, early diagnosis remains limited by inadequate equipment, financial resources, and scarce GI expertise. AI-assisted diagnosis and report generation show great promise in alleviating this problem by providing non-specialist healthcare workers the technical expertise to perform diagnosis. Yet, almost all open-source, publicly available datasets are predominantly collected from the European region, with minimal representation from the South Asian region. The lack of open-source GI datasets from diverse geographic regions has made it difficult to assess whether population bias is present in existing models, and to develop geographically inclusive AI tools for automated GI diagnosis. To address this gap, we introduce SAGE: An Expert-Annotated South Asian GI Endoscopy dataset for Multimodal Learning and Hallucination Analysis, for image captioning, multi-label classification, and VQA tasks. It consists of 1,300 images, their captions along with hallucination tag, 18 labels and 14,276 question-answer pairs making it well-suited for diverse range of tasks including classification, benchmarking, and fine-tuning LMMs. We further conducted benchmarking of task-specific models, like multi-class classifiers on the effect of population shift which reveals that such models suffers the most with an average F1 score drop of 54 points on South Asian dataset. Further, benchmarking of contemporary LMMs reveals a substantial drop in the average GREEN score for anatomical landmark detection (0.308) and abnormality detection (0.410). We open-source our dataset under the CC BY-SA 4.0 license, and we hope to encourage others to contribute toward more inclusive dataset representation and help counteract population bias in medical AI.

cs.CV↗

arXiv2026

ProMoE-FL: Prototype-conditioned Mixture of Experts for Multimodal Federated Learning with Missing Modalities

In this paper, we address the problem of multimodal federated learning with missing modality. Existing methods utilize an additional public dataset or perform naive feature synthesis that is based solely on the available modality. To address these limitations, we propose ProMoE-FL, a Prototype-conditioned Mixture-of-Experts framework for robust missing-modality feature synthesis in multimodal federated learning. ProMoE-FL builds a global client-aware prototype bank that captures clinically meaningful modality priors across institutions. Our Mixture of Experts is conditioned on these prototypes and modality indices to enable direction-aware expert routing for dynamically synthesizing missing features. We perform extensive quantitative and qualitative evaluations on four public chest X-ray datasets (MIMIC-CXR, NIH Open-I, PadChest, and CheXpert) and demonstrate that ProMoE-FL consistently outperforms state-of-the-art methods in both homogeneous as well as the more challenging heterogeneous settings.

cs.CV↗

arXiv2026

Comparing Sentiment Contagion in AI-Agent and Human Social Networks: Evidence from MOLTBOOK

AI agents are beginning to interact not only with people, but also with one another. We investigate what happens to sentiment in such an AI-only social network: does negativity spread, or do replies calm it down? We study MOLTBOOK, a social network made up of autonomous language-model agents, using almost 2.9 million posts and 1.5 million comments. Negative posts receive many more replies than neutral or positive posts, so negativity still attracts attention. However, replies to negative content usually do not stay negative. They most often become neutral, and there is meager evidence that negative sentiment spreads across days. The main pattern is therefore not a cycle of negativity, but negative attention followed by neutralisation. These findings suggest that AI-agent networks may behave differently from human social networks: they may dampen emotional extremes, while still depending strongly on how interactions are organised.

cs.SI↗

arXiv2026

Med-MMFL: A Multimodal Federated Learning Benchmark in Healthcare

Federated learning (FL) enables collaborative model training across decentralized medical institutions while preserving data privacy. However, medical FL benchmarks remain scarce, with existing efforts focusing mainly on unimodal or bimodal modalities and a limited range of medical tasks. This gap underscores the need for standardized evaluation to advance systematic understanding in medical MultiModal FL (MMFL). To this end, we introduce Med-MMFL, the first comprehensive MMFL benchmark for the medical domain, encompassing diverse modalities, tasks, and federation scenarios. Our benchmark evaluates six representative state-of-the-art FL algorithms, covering different aggregation strategies, loss formulations, and regularization techniques. It spans datasets with 2 to 4 modalities, comprising a total of 10 unique medical modalities, including text, pathology images, ECG, X-ray, radiology reports, and multiple MRI sequences. Experiments are conducted across naturally federated, synthetic IID, and synthetic non-IID settings to simulate real-world heterogeneity. We assess segmentation, classification, modality alignment (retrieval), and VQA tasks. To support reproducibility and fair comparison of future multimodal federated learning (MMFL) methods under realistic medical settings, we release the complete benchmark implementation, including data processing and partitioning pipelines, at https://github.com/bhattarailab/Med-MMFL-Benchmark .

cs.CV↗

arXiv2025

Local K-Similarity Constraint for Federated Learning with Label Noise

Federated learning on clients with noisy labels is a challenging problem, as such clients can infiltrate the global model, impacting the overall generalizability of the system. Existing methods proposed to handle noisy clients assume that a sufficient number of clients with clean labels are available, which can be leveraged to learn a robust global model while dampening the impact of noisy clients. This assumption fails when a high number of heterogeneous clients contain noisy labels, making the existing approaches ineffective. In such scenarios, it is important to locally regularize the clients before communication with the global model, to ensure the global model isn't corrupted by noisy clients. While pre-trained self-supervised models can be effective for local regularization, existing centralized approaches relying on pretrained initialization are impractical in a federated setting due to the potentially large size of these models, which increases communication costs. In that line, we propose a regularization objective for client models that decouples the pre-trained and classification models by enforcing similarity between close data points within the client. We leverage the representation space of a self-supervised pretrained model to evaluate the closeness among examples. This regularization, when applied with the standard objective function for the downstream task in standard noisy federated settings, significantly improves performance, outperforming existing state-of-the-art federated methods in multiple computer vision and medical image classification benchmarks. Unlike other techniques that rely on self-supervised pretrained initialization, our method does not require the pretrained model and classifier backbone to share the same architecture, making it architecture-agnostic.

cs.LG↗

arXiv2025

Surgical Vision World Model

Realistic and interactive surgical simulation has the potential to facilitate crucial applications, such as medical professional training and autonomous surgical agent training. In the natural visual domain, world models have enabled action-controlled data generation, demonstrating the potential to train autonomous agents in interactive simulated environments when large-scale real data acquisition is infeasible. However, such works in the surgical domain have been limited to simplified computer simulations, and lack realism. Furthermore, existing literature in world models has predominantly dealt with action-labeled data, limiting their applicability to real-world surgical data, where obtaining action annotation is prohibitively expensive. Inspired by the recent success of Genie in leveraging unlabeled video game data to infer latent actions and enable action-controlled data generation, we propose the first surgical vision world model. The proposed model can generate action-controllable surgical data and the architecture design is verified with extensive experiments on the unlabeled SurgToolLoc-2022 dataset. Codes and implementation details are available at https://github.com/bhattarailab/Surgical-Vision-World-Model

eess.IV↗

arXiv2025

Noise, Adaptation, and Strategy: Assessing LLM Fidelity in Decision-Making

Large language models (LLMs) are increasingly used in social science simulations. While their performance on reasoning and optimization tasks has been extensively evaluated, less attention has been paid to their ability to simulate human decision-making's variability and adaptability. We propose a process-oriented evaluation framework with progressive interventions (Intrinsicality, Instruction, and Imitation) to examine how LLM agents adapt under different levels of external guidance and human-derived noise. We validate the framework on two classic economics tasks, irrationality in the second-price auction and decision bias in the newsvendor problem, showing behavioral gaps between LLMs and humans. We find that LLMs, by default, converge on stable and conservative strategies that diverge from observed human behaviors. Risk-framed instructions impact LLM behavior predictably but do not replicate human-like diversity. Incorporating human data through in-context learning narrows the gap but fails to reach human subjects' strategic variability. These results highlight a persistent alignment gap in behavioral fidelity and suggest that future LLM evaluations should consider more process-level realism. We present a process-oriented approach for assessing LLMs in dynamic decision-making tasks, offering guidance for their application in synthetic data for social science research.

cs.CE↗

arXiv2025

Contextualizing Recommendation Explanations with LLMs: A User Study

Large language models (LLMs) are increasingly prevalent in recommender systems, where LLMs can be used to generate personalized recommendations. Here, we examine how different LLM-generated explanations for movie recommendations affect users' perceptions of cognitive, affective, and utilitarian needs and consumption intentions. In a pre-registered, between-subject online experiment (N=759) and follow-up interviews (N=30), we compare (a) LLM-generated generic explanations, and (b) LLM-generated contextualized explanations. Our findings show that contextualized explanations (i.e., explanations that incorporate users' past behaviors) effectively meet users' cognitive needs while increasing users' intentions to watch recommended movies. However, adding explanations offers limited benefits in meeting users' utilitarian and affective needs, raising concerns about the proper design and implications of LLM-generated explanations. Qualitative insights from interviews reveal that referencing users' past preferences enhances trust and understanding but can feel excessive if overused. Furthermore, users with more active and positive engagement with the recommender system and movie-watching get substantial gains from contextualized explanations. Overall, our research clarifies how LLM-generated recommendations influence users' motivations and behaviors, providing valuable insights for the future development of user-centric recommender systems, a key element in social media platforms and online ecosystems.

cs.HC↗

arXiv2025

Hallucination-Aware Multimodal Benchmark for Gastrointestinal Image Analysis with Large Vision-Language Models

Vision-Language Models (VLMs) are becoming increasingly popular in the medical domain, bridging the gap between medical images and clinical language. Existing VLMs demonstrate an impressive ability to comprehend medical images and text queries to generate detailed, descriptive diagnostic medical reports. However, hallucination--the tendency to generate descriptions that are inconsistent with the visual content--remains a significant issue in VLMs, with particularly severe implications in the medical field. To facilitate VLM research on gastrointestinal (GI) image analysis and study hallucination, we curate a multimodal image-text GI dataset: Gut-VLM. This dataset is created using a two-stage pipeline: first, descriptive medical reports of Kvasir-v2 images are generated using ChatGPT, which introduces some hallucinated or incorrect texts. In the second stage, medical experts systematically review these reports, and identify and correct potential inaccuracies to ensure high-quality, clinically reliable annotations. Unlike traditional datasets that contain only descriptive texts, our dataset also features tags identifying hallucinated sentences and their corresponding corrections. A common approach to reducing hallucination in VLM is to finetune the model on a small-scale, problem-specific dataset. However, we take a different strategy using our dataset. Instead of finetuning the VLM solely for generating textual reports, we finetune it to detect and correct hallucinations, an approach we call hallucination-aware finetuning. Our results show that this approach is better than simply finetuning for descriptive report generation. Additionally, we conduct an extensive evaluation of state-of-the-art VLMs across several metrics, establishing a benchmark. GitHub Repo: https://github.com/bhattarailab/Hallucination-Aware-VLM.

cs.CV↗

arXiv2025

Human aversion? Do AI Agents Judge Identity More Harshly Than Performance

This study examines the understudied role of algorithmic evaluation of human judgment in hybrid decision-making systems, a critical gap in management research. While extant literature focuses on human reluctance to follow algorithmic advice, we reverse the perspective by investigating how AI agents based on large language models (LLMs) assess and integrate human input. Our work addresses a pressing managerial constraint: firms barred from deploying LLMs directly due to privacy concerns can still leverage them as mediating tools (for instance, anonymized outputs or decision pipelines) to guide high-stakes choices like pricing or discounts without exposing proprietary data. Through a controlled prediction task, we analyze how an LLM-based AI agent weights human versus algorithmic predictions. We find that the AI system systematically discounts human advice, penalizing human errors more severely than algorithmic errors--a bias exacerbated when the agent's identity (human vs AI) is disclosed and the human is positioned second. These results reveal a disconnect between AI-generated trust metrics and the actual influence of human judgment, challenging assumptions about equitable human-AI collaboration. Our findings offer three key contributions. First, we identify a reverse algorithm aversion phenomenon, where AI agents undervalue human input despite comparable error rates. Second, we demonstrate how disclosure and positional bias interact to amplify this effect, with implications for system design. Third, we provide a framework for indirect LLM deployment that balances predictive power with data privacy. For practitioners, this research emphasize the need to audit AI weighting mechanisms, calibrate trust dynamics, and strategically design decision sequences in human-AI systems.

cs.HC↗

arXiv2024

NCDD: Nearest Centroid Distance Deficit for Out-Of-Distribution Detection in Gastrointestinal Vision

The integration of deep learning tools in gastrointestinal vision holds the potential for significant advancements in diagnosis, treatment, and overall patient care. A major challenge, however, is these tools' tendency to make overconfident predictions, even when encountering unseen or newly emerging disease patterns, undermining their reliability. We address this critical issue of reliability by framing it as an out-of-distribution (OOD) detection problem, where previously unseen and emerging diseases are identified as OOD examples. However, gastrointestinal images pose a unique challenge due to the overlapping feature representations between in- Distribution (ID) and OOD examples. Existing approaches often overlook this characteristic, as they are primarily developed for natural image datasets, where feature distinctions are more apparent. Despite the overlap, we hypothesize that the features of an in-distribution example will cluster closer to the centroids of their ground truth class, resulting in a shorter distance to the nearest centroid. In contrast, OOD examples maintain an equal distance from all class centroids. Based on this observation, we propose a novel nearest-centroid distance deficit (NCCD) score in the feature space for gastrointestinal OOD detection. Evaluations across multiple deep learning architectures and two publicly available benchmarks, Kvasir2 and Gastrovision, demonstrate the effectiveness of our approach compared to several state-of-the-art methods. The code and implementation details are publicly available at: https://github.com/bhattarailab/NCDD

cs.CV↗

arXiv2024

The Effect of Education in Prompt Engineering: Evidence from Journalists

Large language models (LLMs) are increasingly used in daily work. In this paper, we analyze whether training in prompt engineering can improve the interactions of users with LLMs. For this, we conducted a field experiment where we asked journalists to write short texts before and after training in prompt engineering. We then analyzed the effect of training on three dimensions: (1) the user experience of journalists when interacting with LLMs, (2) the accuracy of the texts (assessed by a domain expert), and (3) the reader perception, such as clarity, engagement, and other text quality dimensions (assessed by non-expert readers). Our results show: (1) Our training improved the perceived expertise of journalists but also decreased the perceived helpfulness of LLM use. (2) The effect on accuracy varied by the difficulty of the task. (3) There is a mixed impact of training on reader perception across different text quality dimensions.

cs.HC↗

arXiv2024

Difficulty Estimation and Simplification of French Text Using LLMs

We leverage generative large language models for language learning applications, focusing on estimating the difficulty of foreign language texts and simplifying them to lower difficulty levels. We frame both tasks as prediction problems and develop a difficulty classification model using labeled examples, transfer learning, and large language models, demonstrating superior accuracy compared to previous approaches. For simplification, we evaluate the trade-off between simplification quality and meaning preservation, comparing zero-shot and fine-tuned performances of large language models. We show that meaningful text simplifications can be obtained with limited fine-tuning. Our experiments are conducted on French texts, but our methods are language-agnostic and directly applicable to other foreign languages.

cs.CL↗

arXiv2024

Cross-Task Data Augmentation by Pseudo-label Generation for Region Based Coronary Artery Instance Segmentation

Coronary Artery Diseases (CADs) although preventable, are one of the leading causes of death and disability. Diagnosis of these diseases is often difficult and resource intensive. Angiographic imaging segmentation of the arteries has evolved as a tool of assistance that helps clinicians make an accurate diagnosis. However, due to the limited amount of data and the difficulty in curating a dataset, the task of segmentation has proven challenging. In this study, we introduce the use of pseudo-labels to address the issue of limited data in the angiographic dataset to enhance the performance of the baseline YOLO model. Unlike existing data augmentation techniques that improve the model constrained to a fixed dataset, we introduce the use of pseudo-labels generated on a dataset of separate related task to diversify and improve model performance. This method increases the baseline F1 score by 9% in the validation data set and by 3% in the test data set.

eess.IV↗

arXiv2024

CAR-MFL: Cross-Modal Augmentation by Retrieval for Multimodal Federated Learning with Missing Modalities

Multimodal AI has demonstrated superior performance over unimodal approaches by leveraging diverse data sources for more comprehensive analysis. However, applying this effectiveness in healthcare is challenging due to the limited availability of public datasets. Federated learning presents an exciting solution, allowing the use of extensive databases from hospitals and health centers without centralizing sensitive data, thus maintaining privacy and security. Yet, research in multimodal federated learning, particularly in scenarios with missing modalities a common issue in healthcare datasets remains scarce, highlighting a critical area for future exploration. Toward this, we propose a novel method for multimodal federated learning with missing modalities. Our contribution lies in a novel cross-modal data augmentation by retrieval, leveraging the small publicly available dataset to fill the missing modalities in the clients. Our method learns the parameters in a federated manner, ensuring privacy protection and improving performance in multiple challenging multimodal benchmarks in the medical domain, surpassing several competitive baselines. Code Available: https://github.com/bhattarailab/CAR-MFL

cs.CV↗

arXiv2024

Comparing the willingness to share for human-generated vs. AI-generated fake news

Generative artificial intelligence (AI) presents large risks for society when it is used to create fake news. A crucial factor for fake news to go viral on social media is that users share such content. Here, we aim to shed light on the sharing behavior of users across human-generated vs. AI-generated fake news. Specifically, we study: (1) What is the perceived veracity of human-generated fake news vs. AI-generated fake news? (2) What is the user's willingness to share human-generated fake news vs. AI-generated fake news on social media? (3) What socio-economic characteristics let users fall for AI-generated fake news? To this end, we conducted a pre-registered, online experiment with $N=$ 988 subjects and 20 fake news from the COVID-19 pandemic generated by GPT-4 vs. humans. Our findings show that AI-generated fake news is perceived as less accurate than human-generated fake news, but both tend to be shared equally. Further, several socio-economic factors explain who falls for AI-generated fake news.

cs.SI↗

arXiv2023

ConvNeXtv2 Fusion with Mask R-CNN for Automatic Region Based Coronary Artery Stenosis Detection for Disease Diagnosis

Coronary Artery Diseases although preventable are one of the leading cause of mortality worldwide. Due to the onerous nature of diagnosis, tackling CADs has proved challenging. This study addresses the automation of resource-intensive and time-consuming process of manually detecting stenotic lesions in coronary arteries in X-ray coronary angiography images. To overcome this challenge, we employ a specialized Convnext-V2 backbone based Mask RCNN model pre-trained for instance segmentation tasks. Our empirical findings affirm that the proposed model exhibits commendable performance in identifying stenotic lesions. Notably, our approach achieves a substantial F1 score of 0.5353 in this demanding task, underscoring its effectiveness in streamlining this intensive process.

cs.CV↗