Kaushik Dutta

Negar Maleki, Balaji Padamanabhan, Kaushik Dutta

Graph representations for real-world social networks in the past have missed two important elements: the multiplexity of connections as well as representing time. To this end, in this paper, we present a new dynamic heterogeneous graph representation for social networks which includes time in every single component of the graph, i.e., nodes and edges, each of different types that captures heterogeneity. We illustrate the power of this representation by presenting four time-dependent queries and deep learning problems that cannot easily be handled in conventional homogeneous graph representations commonly used. As a proof of concept we present a detailed representation of a new social media platform (Steemit), which we use to illustrate both the dynamic querying capability as well as prediction tasks using graph neural networks (GNNs). The results illustrate the power of the dynamic heterogeneous graph representation to model social networks. Given that this is a relatively understudied area we also illustrate opportunities for future work in query optimization as well as new dynamic prediction tasks on heterogeneous graph structures.

Negar Maleki, Balaji Padmanabhan, Kaushik Dutta

As large language models continue to advance in Artificial Intelligence (AI), text generation systems have been shown to suffer from a problematic phenomenon termed often as "hallucination." However, with AI's increasing presence across various domains including medicine, concerns have arisen regarding the use of the term itself. In this study, we conducted a systematic review to identify papers defining "AI hallucination" across fourteen databases. We present and analyze definitions obtained across all databases, categorize them based on their applications, and extract key points within each category. Our results highlight a lack of consistency in how the term is used, but also help identify several alternative terms in the literature. We discuss implications of these and call for a more unified effort to bring consistency to an important contemporary AI issue that can affect multiple domains significantly.

Sonal Prabhune, Balaji Padmanabhan, Kaushik Dutta

We investigate the existence and persistence of a specific type of gender bias in some of the popular LLMs and contribute a new benchmark dataset, RealWorldQuestioning (released on HuggingFace ), developed from real-world questions across four key domains in business and health contexts: education, jobs, personal financial management, and general health. We define and study entropy bias, which we define as a discrepancy in the amount of information generated by an LLM in response to real questions users have asked. We tested this using four different LLMs and evaluated the generated responses both qualitatively and quantitatively by using ChatGPT-4o (as "LLM-as-judge"). Our analyses (metric-based comparisons and "LLM-as-judge" evaluation) suggest that there is no significant bias in LLM responses for men and women at a category level. However, at a finer granularity (the individual question level), there are substantial differences in LLM responses for men and women in the majority of cases, which "cancel" each other out often due to some responses being better for males and vice versa. This is still a concern since typical users of these tools often ask a specific question (only) as opposed to several varied ones in each of these common yet important areas of life. We suggest a simple debiasing approach that iteratively merges the responses for the two genders to produce a final result. Our approach demonstrates that a simple, prompt-based debiasing strategy can effectively debias LLM outputs, thus producing responses with higher information content than both gendered variants in 78% of the cases, and consistently achieving a balanced integration in the remaining cases.

Sonal Prabhune, Balaji Padmanabhan, Kaushik Dutta

Large Language Models (LLMs) are increasingly deployed in business-critical domains such as finance, education, healthcare, and customer support, where users expect consistent and reliable recommendations. Yet LLMs often exhibit variability when prompts are phrased with minor differences, even when semantically equivalent. Such inconsistency undermines trust, complicates compliance, and disrupts user experience. While personalization is desirable in certain contexts, many enterprise scenarios-such as HR onboarding, customer support, or policy disclosure-require invariant information delivery regardless of phrasing or prior conversational history. Existing approaches, including retrieval-augmented generation (RAG) and temperature tuning, improve factuality or reduce stochasticity but cannot guarantee stability across equivalent prompts. In this paper, we propose a reinforcement learning framework based on Group Relative Policy Optimization (GRPO) to directly optimize for consistency. Unlike prior applications of GRPO, which have been limited to reasoning and code generation, we adapt GRPO to enforce stability of information content across groups of semantically equivalent prompts. We introduce entropy-based helpfulness and stability rewards, treating prompt variants as groups and resetting conversational context to isolate phrasing effects. Experiments on investment and job recommendation tasks show that our GRPO-trained model reduces variability more effectively than fine-tuning or decoding-based baselines. To our knowledge, this is a novel application of GRPO for aligning LLMs toward information consistency, reframing variability not as an acceptable feature of generative diversity but as a correctable flaw in enterprise deployments.

Sneha Dasgupta, Rishav Mukherjee, Kaushik Dutta et al.

Automatic evaluation of the retinal fundus image is emerging as one of the most important tools for early detection and treatment of progressive eye diseases like Glaucoma. Glaucoma results to a progressive degeneration of vision and is characterized by the deformation of the shape of optic cup and the degeneration of the blood vessels resulting in the formation of a notch along the neuroretinal rim. In this paper, we propose a deep learning-based pipeline for automatic segmentation of optic disc (OD) and optic cup (OC) regions from Digital Fundus Images (DFIs), thereby extracting distinct features necessary for prediction of Glaucoma. This methodology has utilized focal notch analysis of neuroretinal rim along with cup-to-disc ratio values as classifying parameters to enhance the accuracy of Computer-aided design (CAD) systems in analyzing glaucoma. Support Vector-based Machine Learning algorithm is used for classification, which classifies DFIs as Glaucomatous or Normal based on the extracted features. The proposed pipeline was evaluated on the freely available DRISHTI-GS dataset with a resultant accuracy of 93.33% for detecting Glaucoma from DFIs.

Shalinda Adikari, Kaushik Dutta

As organizations increasingly rely on professionally oriented networks such as LinkedIn (the largest such social network) for building business connections, there is increasing value in having one's profile noticed within the network. As this value increases, so does the temptation to misuse the network for unethical purposes. Fake profiles have an adverse effect on the trustworthiness of the network as a whole, and can represent significant costs in time and effort in building a connection based on fake information. Unfortunately, fake profiles are difficult to identify. Approaches have been proposed for some social networks; however, these generally rely on data that are not publicly available for LinkedIn profiles. In this research, we identify the minimal set of profile data necessary for identifying fake profiles in LinkedIn, and propose an appropriate data mining approach for fake profile identification. We demonstrate that, even with limited profile data, our approach can identify fake profiles with 87% accuracy and 94% True Negative Rate, which is comparable to the results obtained based on larger data sets and more expansive profile information. Further, when compared to approaches using similar amounts and types of data, our method provides an improvement of approximately 14% accuracy.