arXiv subjects

Amulya Yadav

Publications and source records attributed to Amulya Yadav.

At least 19 recordsLinked to original sources

arXiv2026

CHAI for LLMs: Improving Code-Mixed Translation in Large Language Models through Reinforcement Learning with AI Feedback

Large language models (LLMs) show strong performance across many tasks but remain weak at understanding code-mixed (CM) language. Despite this limitation, improving LLMs for CM tasks has received little attention. To address this gap, we propose CHAI, a general-purpose framework for enhancing LLM performance on CM tasks, focusing on CM translation. CHAI leverages four key ideas. First, we investigate the use of LLMs as annotators to address the scarcity of high-quality CM datasets. Second, we leverage these LLM annotations to generate large-scale preference data and apply reinforcement learning from AI feedback (RLAIF) to improve CM translation. Third, we incorporate LLM-generated domain knowledge as a constitution, enabling iterative response refinement. Fourth, we perform extensive evaluations on real-world datasets and settings. Results show that CHAI-powered LLMs outperform state-of-the-art open-source models by 68.45% on average in human-adjudicated win rate on CM translation tasks. This work is a step toward more inclusive open-source code-mixed LLMs.

cs.CL↗

arXiv2026

The Mirage of LLM Guardrails: A Case Study in AI-Assisted Medical Note Manipulation

The rapid deployment of large language models (LLMs) in healthcare settings makes the reliability of their built-in guardrails against malicious queries a question of urgent practical consequence. Yet the robustness of these mechanisms against deliberate misuse (in the healthcare context) remains poorly understood. In this paper, we investigate this question empirically, using AI-assisted medical note manipulation as a concrete case study. We make four novel contributions. First, we develop a reproducible manipulation pipeline that takes publicly available seed medical note templates and use commercial LLMs to produce customized manipulated notes by substituting patient names, provider identities, dates, and medical conditions across multiple model families, input formats, and prompt phrasings. Second, we conduct a systematic empirical evaluation of LLM guardrail robustness for medical note manipulation. Our experimental results reveal substantial weaknesses and inconsistencies in contemporary commercial LLM guardrails, including low refusal rates for several model families. Third, we utilize a combination of automated metrics and human annotation-based metrics to assess the correctness of requested manipulations. Fourth, we conduct a user-study to assess the believability of manipulated medical notes, finding that the best manipulations are visually indistinguishable from original documents to human raters. Finally, we discuss implications for responsible guardrail design in LLMs, AI safety policies, and the broader ethics of deploying LLMs in healthcare settings.

cs.CR↗

arXiv2026

The Hardness of Achieving Impact in AI for Social Impact Research: A Ground-Level View of Challenges & Opportunities

AI for Social Impact (AI4SI) is an emergent field harnessing interdisciplinarities between the fields of artificial intelligence (AI), machine learning (ML), and the social sciences to address societal issues aligned with the United Nations Sustainable Development Goals (UN SDGs), such as universal healthcare, climate action, etc. Despite AI4SI's rising popularity, achieving tangible, on-the-ground impact remains a significant challenge. In particular, identifying collaborators open to co-designing and deploying AI4SI-based solutions in real-world settings is often difficult. Thus, many projects stall at the proof-of-concept stage, unable to scale to production-level deployment. Drawing on twenty-six AI4SI researchers' interviews, primarily from academic institutions though also including some industry researchers and practitioners, and the authors' own lived experiences, this paper employs thematic analysis to highlight structural, organizational, communication, collaboration, and operational challenges hindering socially impactful AI4SI deployments. While there are no easy fixes, the authors synthesize best practices and actionable strategies from interviews and personal experiences, positioning this paper as a practical guide for AI4SI researchers and organizations pursuing socially impactful collaborations$^1$. $^1$We note that our findings are most directly applicable to academic research groups in the global north, as governmental, startup, and global south researchers' perspectives are underrepresented in our sample.

cs.HC↗

arXiv2026

Designing User-Centric Metrics for Evaluation of Counterfactual Explanations

Counterfactual Explanations (CFEs) have grown in popularity as a means of offering actionable guidance by identifying the minimum changes in feature values required to flip an ML model's prediction to something more desirable. Unfortunately, most prior research on CFEs relies on artificial evaluation metrics, such as proximity, which may overlook end-user preferences and constraints, e.g., the user's perception of effort needed to make certain feature changes may differ from that of the model designer. To address this research gap, this paper makes three novel contributions. First, we conduct a pilot study with 20 crowd-workers on Amazon MTurk to experimentally validate the alignment of existing CF evaluation metrics with real-world user preferences. Results show that user-preferred CFEs matched those based on proximity in only 63.81% of cases, highlighting the limited applicability of these metrics in real-world settings. Second, inspired by the need to design a user-informed evaluation metric for CFEs, we conduct a more detailed two-day user study with 41 participants facing realistic credit application scenarios to find experimental support for or against three intuitive hypotheses that may explain how end users evaluate CFEs. Third, based on the findings of this second study, we propose the AWP model, a novel user-centric, two-stage model that describes one possible mechanism by which users evaluate and select CFEs. Our results show that AWP predicts user-preferred CFEs with 84.37% accuracy. Our study provides the first human-centered validation for personalized cost models in CFE generation and highlights the need for adaptive, user-centered evaluation metrics.

cs.LG↗

arXiv2025

Evaluating Large Language Models on Rare Disease Diagnosis: A Case Study using House M.D

Large language models (LLMs) have demonstrated capabilities across diverse domains, yet their performance on rare disease diagnosis from narrative medical cases remains underexplored. We introduce a novel dataset of 176 symptom-diagnosis pairs extracted from House M.D., a medical television series validated for teaching rare disease recognition in medical education. We evaluate four state-of-the-art LLMs such as GPT 4o mini, GPT 5 mini, Gemini 2.5 Flash, and Gemini 2.5 Pro on narrative-based diagnostic reasoning tasks. Results show significant variation in performance, ranging from 16.48% to 38.64% accuracy, with newer model generations demonstrating a 2.3 times improvement. While all models face substantial challenges with rare disease diagnosis, the observed improvement across architectures suggests promising directions for future development. Our educationally validated benchmark establishes baseline performance metrics for narrative medical reasoning and provides a publicly accessible evaluation framework for advancing AI-assisted diagnosis research.

cs.CL↗

arXiv2025

Dr. GPT Will See You Now, but Should It? Exploring the Benefits and Harms of Large Language Models in Medical Diagnosis using Crowdsourced Clinical Cases

The proliferation of Large Language Models (LLMs) in high-stakes applications such as medical (self-)diagnosis and preliminary triage raises significant ethical and practical concerns about the effectiveness, appropriateness, and possible harmfulness of the use of these technologies for health-related concerns and queries. Some prior work has considered the effectiveness of LLMs in answering expert-written health queries/prompts, questions from medical examination banks, or queries based on pre-existing clinical cases. Unfortunately, these existing studies completely ignore an in-the-wild evaluation of the effectiveness of LLMs in answering everyday health concerns and queries typically asked by general users, which corresponds to the more prevalent use case for LLMs. To address this research gap, this paper presents the findings from a university-level competition that leveraged a novel, crowdsourced approach for evaluating the effectiveness of LLMs in answering everyday health queries. Over the course of a week, a total of 34 participants prompted four publicly accessible LLMs with 212 real (or imagined) health concerns, and the LLM generated responses were evaluated by a team of nine board-certified physicians. At a high level, our findings indicate that on average, 76% of the 212 LLM responses were deemed to be accurate by physicians. Further, with the help of medical professionals, we investigated whether RAG versions of these LLMs (powered with a comprehensive medical knowledge base) can improve the quality of responses generated by LLMs. Finally, we also derive qualitative insights to explain our quantitative findings by conducting interviews with seven medical professionals who were shown all the prompts in our competition. This paper aims to provide a more grounded understanding of how LLMs perform in real-world everyday health communication.

cs.CY↗

arXiv2025

Have LLMs Reopened the Pandora's Box of AI-Generated Fake News?

With the rise of AI-generated content spewed at scale from large language models (LLMs), genuine concerns about the spread of fake news have intensified. The perceived ability of LLMs to produce convincing fake news at scale poses new challenges for both human and automated fake news detection systems. To address this gap, this paper presents the findings from a university-level competition that aimed to explore how LLMs can be used by humans to create fake news, and to assess the ability of human annotators and AI models to detect it. A total of 110 participants used LLMs to create 252 unique fake news stories, and 84 annotators participated in the detection tasks. Our findings indicate that LLMs are ~68% more effective at detecting real news than humans. However, for fake news detection, the performance of LLMs and humans remains comparable (~60% accuracy). Additionally, we examine the impact of visual elements (e.g., pictures) in news on the accuracy of detecting fake news stories. Finally, we also examine various strategies used by fake news creators to enhance the credibility of their AI-generated content. This work highlights the increasing complexity of detecting AI-generated fake news, particularly in collaborative human-AI settings.

cs.CL↗

arXiv2024

Watermarking Counterfactual Explanations

Counterfactual (CF) explanations for ML model predictions provide actionable recourse recommendations to individuals adversely impacted by predicted outcomes. However, despite being preferred by end-users, CF explanations have been shown to pose significant security risks in real-world applications; in particular, malicious adversaries can exploit CF explanations to perform query-efficient model extraction attacks on the underlying proprietary ML model. To address this security challenge, we propose CFMark, a novel model-agnostic watermarking framework for detecting unauthorized model extraction attacks relying on CF explanations. CFMark involves a novel bi-level optimization problem to embed an indistinguishable watermark into the generated CF explanation such that any future model extraction attacks using these watermarked CF explanations can be detected using a null hypothesis significance testing (NHST) scheme. At the same time, the embedded watermark does not compromise the quality of the CF explanations. We evaluate CFMark across diverse real-world datasets, CF explanation methods, and model extraction techniques. Our empirical results demonstrate CFMark's effectiveness, achieving an F-1 score of ~0.89 in identifying unauthorized model extraction attacks using watermarked CF explanations. Importantly, this watermarking incurs only a negligible degradation in the quality of generated CF explanations (i.e., ~1.3% degradation in validity and ~1.6% in proximity). Our work establishes a critical foundation for the secure deployment of CF explanations in real-world applications.

cs.LG↗

arXiv2024

Hey GPT, Can You be More Racist? Analysis from Crowdsourced Attempts to Elicit Biased Content from Generative AI

The widespread adoption of large language models (LLMs) and generative AI (GenAI) tools across diverse applications has amplified the importance of addressing societal biases inherent within these technologies. While the NLP community has extensively studied LLM bias, research investigating how non-expert users perceive and interact with biases from these systems remains limited. As these technologies become increasingly prevalent, understanding this question is crucial to inform model developers in their efforts to mitigate bias. To address this gap, this work presents the findings from a university-level competition, which challenged participants to design prompts for eliciting biased outputs from GenAI tools. We quantitatively and qualitatively analyze the competition submissions and identify a diverse set of biases in GenAI and strategies employed by participants to induce bias in GenAI. Our finding provides unique insights into how non-expert users perceive and interact with biases from GenAI tools.

cs.CL↗

arXiv2023

A Taxonomy of Rater Disagreements: Surveying Challenges & Opportunities from the Perspective of Annotating Online Toxicity

Toxicity is an increasingly common and severe issue in online spaces. Consequently, a rich line of machine learning research over the past decade has focused on computationally detecting and mitigating online toxicity. These efforts crucially rely on human-annotated datasets that identify toxic content of various kinds in social media texts. However, such annotations historically yield low inter-rater agreement, which was often dealt with by taking the majority vote or other such approaches to arrive at a single ground truth label. Recent research has pointed out the importance of accounting for the subjective nature of this task when building and utilizing these datasets, and this has triggered work on analyzing and better understanding rater disagreements, and how they could be effectively incorporated into the machine learning developmental pipeline. While these efforts are filling an important gap, there is a lack of a broader framework about the root causes of rater disagreement, and therefore, we situate this work within that broader landscape. In this survey paper, we analyze a broad set of literature on the reasons behind rater disagreements focusing on online toxicity, and propose a detailed taxonomy for the same. Further, we summarize and discuss the potential solutions targeting each reason for disagreement. We also discuss several open issues, which could promote the future development of online toxicity research.

cs.CL↗

arXiv2023

RoCourseNet: Distributionally Robust Training of a Prediction Aware Recourse Model

Counterfactual (CF) explanations for machine learning (ML) models are preferred by end-users, as they explain the predictions of ML models by providing a recourse (or contrastive) case to individuals who are adversely impacted by predicted outcomes. Existing CF explanation methods generate recourses under the assumption that the underlying target ML model remains stationary over time. However, due to commonly occurring distributional shifts in training data, ML models constantly get updated in practice, which might render previously generated recourses invalid and diminish end-users trust in our algorithmic framework. To address this problem, we propose RoCourseNet, a training framework that jointly optimizes predictions and recourses that are robust to future data shifts. This work contains four key contributions: (1) We formulate the robust recourse generation problem as a tri-level optimization problem which consists of two sub-problems: (i) a bi-level problem that finds the worst-case adversarial shift in the training data, and (ii) an outer minimization problem to generate robust recourses against this worst-case shift. (2) We leverage adversarial training to solve this tri-level optimization problem by: (i) proposing a novel virtual data shift (VDS) algorithm to find worst-case shifted ML models via explicitly considering the worst-case data shift in the training dataset, and (ii) a block-wise coordinate descent procedure to optimize for prediction and corresponding robust recourses. (3) We evaluate RoCourseNet's performance on three real-world datasets, and show that RoCourseNet consistently achieves more than 96% robust validity and outperforms state-of-the-art baselines by at least 10% in generating robust CF explanations. (4) Finally, we generalize the RoCourseNet framework to accommodate any parametric post-hoc methods for improving robust validity.

cs.LG↗

arXiv2023

CounterNet: End-to-End Training of Prediction Aware Counterfactual Explanations

This work presents CounterNet, a novel end-to-end learning framework which integrates Machine Learning (ML) model training and the generation of corresponding counterfactual (CF) explanations into a single end-to-end pipeline. Counterfactual explanations offer a contrastive case, i.e., they attempt to find the smallest modification to the feature values of an instance that changes the prediction of the ML model on that instance to a predefined output. Prior techniques for generating CF explanations suffer from two major limitations: (i) all of them are post-hoc methods designed for use with proprietary ML models -- as a result, their procedure for generating CF explanations is uninformed by the training of the ML model, which leads to misalignment between model predictions and explanations; and (ii) most of them rely on solving separate time-intensive optimization problems to find CF explanations for each input data point (which negatively impacts their runtime). This work makes a novel departure from the prevalent post-hoc paradigm (of generating CF explanations) by presenting CounterNet, an end-to-end learning framework which integrates predictive model training and the generation of counterfactual (CF) explanations into a single pipeline. Unlike post-hoc methods, CounterNet enables the optimization of the CF explanation generation only once together with the predictive model. We adopt a block-wise coordinate descent procedure which helps in effectively training CounterNet's network. Our extensive experiments on multiple real-world datasets show that CounterNet generates high-quality predictions, and consistently achieves 100% CF validity and low proximity scores (thereby achieving a well-balanced cost-invalidity trade-off) for any new input instance, and runs 3X faster than existing state-of-the-art baselines.

cs.LG↗

arXiv2020

Clinical trial of an AI-augmented intervention for HIV prevention in youth experiencing homelessness

Youth experiencing homelessness (YEH) are subject to substantially greater risk of HIV infection, compounded both by their lack of access to stable housing and the disproportionate representation of youth of marginalized racial, ethnic, and gender identity groups among YEH. A key goal for health equity is to improve adoption of protective behaviors in this population. One promising strategy for intervention is to recruit peer leaders from the population of YEH to promote behaviors such as condom usage and regular HIV testing to their social contacts. This raises a computational question: which youth should be selected as peer leaders to maximize the overall impact of the intervention? We developed an artificial intelligence system to optimize such social network interventions in a community health setting. We conducted a clinical trial enrolling 713 YEH at drop-in centers in a large US city. The clinical trial compared interventions planned with the algorithm to those where the highest-degree nodes in the youths' social network were recruited as peer leaders (the standard method in public health) and to an observation-only control group. Results from the clinical trial show that youth in the AI group experience statistically significant reductions in key risk behaviors for HIV transmission, while those in the other groups do not. This provides, to our knowledge, the first empirical validation of the usage of AI methods to optimize social network interventions for health. We conclude by discussing lessons learned over the course of the project which may inform future attempts to use AI in community-level interventions.

cs.SI↗

arXiv2020

Preliminary Results from a Peer-Led, Social Network Intervention, Augmented by Artificial Intelligence to Prevent HIV among Youth Experiencing Homelessness

Each year, there are nearly 4 million youth experiencing homelessness (YEH) in the United States with HIV prevalence ranging from 3 to 11.5%. Peer change agent (PCA) models for HIV prevention have been used successfully in many populations, but there have been notable failures. In recent years, network interventionists have suggested that these failures could be attributed to PCA selection procedures. The change agents themselves who are selected to do the PCA work can often be as important as the messages they convey. To address this concern, we tested a new PCA intervention for YEH, with three arms: (1) an arm using an artificial intelligence (AI) planning algorithm to select PCA, (2) a popularity arm--the standard PCA approach--operationalized as highest degree centrality (DC), and (3) an observation only comparison group (OBS). PCA models that promote HIV testing, HIV knowledge, and condom use are efficacious for YEH. Both the AI and DC arms showed improvements over time. AI-based PCA selection led to better outcomes and increased the speed of intervention effects. Specifically, the changes in behavior observed in the AI arm occurred by 1 month, but not until 3 months in the DC arm. Given the transient nature of YEH and the high risk for HIV infection, more rapid intervention effects are desirable.

cs.SI↗

arXiv2020

Exploring Algorithmic Fairness in Robust Graph Covering Problems

Fueled by algorithmic advances, AI algorithms are increasingly being deployed in settings subject to unanticipated challenges with complex social effects. Motivated by real-world deployment of AI driven, social-network based suicide prevention and landslide risk management interventions, this paper focuses on robust graph covering problems subject to group fairness constraints. We show that, in the absence of fairness constraints, state-of-the-art algorithms for the robust graph covering problem result in biased node coverage: they tend to discriminate individuals (nodes) based on membership in traditionally marginalized groups. To mitigate this issue, we propose a novel formulation of the robust graph covering problem with group fairness constraints and a tractable approximation scheme applicable to real-world instances. We provide a formal analysis of the price of group fairness (PoF) for this problem, where we show that uncertainty can lead to greater PoF. We demonstrate the effectiveness of our approach on several real-world social networks. Our method yields competitive node coverage while significantly improving group fairness relative to state-of-the-art methods.

math.OC↗

arXiv2019

Artificial Intelligence for Low-Resource Communities: Influence Maximization in an Uncertain World

The potential of Artificial Intelligence (AI) to tackle challenging problems that afflict society is enormous, particularly in the areas of healthcare, conservation and public safety and security. Many problems in these domains involve harnessing social networks of under-served communities to enable positive change, e.g., using social networks of homeless youth to raise awareness about Human Immunodeficiency Virus (HIV) and other STDs. Unfortunately, most of these real-world problems are characterized by uncertainties about social network structure and influence models, and previous research in AI fails to sufficiently address these uncertainties. This thesis addresses these shortcomings by advancing the state-of-the-art to a new generation of algorithms for interventions in social networks. In particular, this thesis describes the design and development of new influence maximization algorithms which can handle various uncertainties that commonly exist in real-world social networks. These algorithms utilize techniques from sequential planning problems and social network theory to develop new kinds of AI algorithms. Further, this thesis also demonstrates the real-world impact of these algorithms by describing their deployment in three pilot studies to spread awareness about HIV among actual homeless youth in Los Angeles. This represents one of the first-ever deployments of computer science based influence maximization algorithms in this domain. Our results show that our AI algorithms improved upon the state-of-the-art by 160% in the real-world. We discuss research and implementation challenges faced in deploying these algorithms, and lessons that can be gleaned for future deployment of such algorithms. The positive results from these deployments illustrate the enormous potential of AI in addressing societally relevant problems.

cs.AI↗

arXiv2019

Influence Maximization for Social Good: Use of Social Networks in Low Resource Communities

This thesis proposal makes the following technical contributions: (i) we provide a definition of the Dynamic Influence Maximization Under Uncertainty (or DIME) problem, which models the problem faced by homeless shelters accurately; (ii) we propose a novel Partially Observable Markov Decision Process (POMDP) model for solving the DIME problem; (iii) we design two scalable POMDP algorithms (PSINET and HEALER) for solving the DIME problem, since conventional POMDP solvers fail to scale up to sizes of interest; and (iv) we test our algorithms effectiveness in the real world by conducting a pilot study with actual homeless youth in Los Angeles. The success of this pilot (as explained later) shows the promise of using influence maximization for social good on a larger scale.

cs.SI↗

arXiv2017

Activating the "Breakfast Club": Modeling Influence Spread in Natural-World Social Networks

While reigning models of diffusion have privileged the structure of a given social network as the key to informational exchange, real human interactions do not appear to take place on a single graph of connections. Using data collected from a pilot study of the spread of HIV awareness in social networks of homeless youth, we show that health information did not diffuse in the field according to the processes outlined by dominant models. Since physical network diffusion scenarios often diverge from their more well-studied counterparts on digital networks, we propose an alternative Activation Jump Model (AJM) that describes information diffusion on physical networks from a multi-agent team perspective. Our model exhibits two main differentiating features from leading cascade and threshold models of influence spread: 1) The structural composition of a seed set team impacts each individual node's influencing behavior, and 2) an influencing node may spread information to non-neighbors. We show that the AJM significantly outperforms existing models in its fit to the observed node-level influence data on the youth networks. We then prove theoretical results, showing that the AJM exhibits many well-behaved properties shared by dominant models. Our results suggest that the AJM presents a flexible and more accurate model of network diffusion that may better inform influence maximization in the field.

cs.SI↗