BABijaya Adhikari

Bijaya Adhikari

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Bijaya Adhikari has a total of 372 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Data-Driven Near-Optimal Regional Epidemic Surveillance

    by Rares-Mihail Neagu, Jeffrey Keithley, Xin Zan, Sriram Pemmaraju, Bijaya Adhikari

    In the early stages of an epidemic outbreak, when the incidence data is limited, policy makers actively try to establish the scale and scope of the outbreak. Two of the most pertinent questions they seek to answer are i) has the outbreak started in a particular region (e.g. a country) and ii) how many sub-regions (e.g. a state or a county) have been affected thus far? We study these problems as a data-driven knowledge extraction task, i.e., given mobility and epidemiological data, we aim to extract actionable knowledge about which regions should receive additional surveillance resources so that an outbreak is detected early and its spatial footprint is accurately characterized. To this end, we formulate two discrete-lattice optimization objectives for epidemic surveillance, 1) maximizing the probability of detecting at least one infected sub-region and 2) maximizing the expected number of infected sub-regions detected. Both objectives turn out to be NP-hard. However, we prove that they are both submodular, which paves the way for efficient epidemic surveillance from limited data. We use this structure to derive curvature bounds for the objective. We also relate the bound with the parameters of the metapopulation disease dynamics. We propose a scalable algorithm, Lattice-Adapted Modified Greedy (LAMGreedy), that uses the curvature, and marginal-gain machinery to establish constant factor approximation guarantee. Extensive experiments on real-world mobility networks for COVID-19 and Influenza across three U.S. states show that LAMGreedy consistently outperforms natural and the state-of-the-art baselines under multiple cost models. The empirical results validate the prevalence and curvature estimates, and demonstrate that the proposed tools extract high-quality surveillance knowledge from mobility and epidemic data under realistic budget constraints.

  • Open Access

    epiDAMIK 2026: Workshop on Data-driven Decision Making for Public and Population Health

    by Alexander Rodríguez, Bijaya Adhikari, Ajitesh Srivastava, Kai Wang, Marie-Laure Charpignon, Serina Chang, Jiaming Cui, Wei Jin, Ananya Joshi, Anil Vullikanti, et al.

  • Open Access

    Spectral Gap-Driven Coarsening for Dynamic Graph Neural Networks

    by Hieu Vu, Rares-Mihail Neagu, Bijaya Adhikari

  • Open Access

    Linking Attentional Modulation to Auditory Attention Decoding: Using Colocated Stimuli With a Fixed Temporal Structure.

    by Jusung Ham, Ian Pope, Jinhee Kim, Hwan Shim, Bijaya Adhikari, Yu-Hsiang Wu, Kyogu Lee, Barbara G Shinn-Cunningham, Octav Chipara, Inyong Choi

    Neural representations of task-relevant sounds are emphasized when they are attended to, compared with when they are ignored. Classic markers of this modulation, such as amplitude change of event-related potentials (ERPs) and attentional modulation indices (AMIs) derived from envelope-tracking analyses, provide robust quantitative measures of top-down attentional strength. However, it remains unclear how well modern auditory attention decoding (AAD) algorithms applied to short electroencephalography (EEG) segments reflect these established neural signatures.Here, we used a two-stream, colocated listening paradigm with fixed and highly regular temporal structure, enabling precise isolation of ERPs and reliable computation of AMI. Participants attended to one of two simultaneous speech streams and detected occasional pitch deviants, while a 64-channel EEG was recorded. We compared three decoding pipelines-a forward linear model-based decoder, a backward linear model-based decoder, and a convolutional neural network (CNN) decoder-in their ability to classify the attended stream from single 4-s trials, a window short enough to reveal performance differences while still supporting above-chance decoding. Importantly, we examined how decoding outcomes relate to classical attentional modulation, including ERP peak amplitudes and AMI.All models achieved significant AAD performance, with the CNN decoder yielding the highest accuracy. Decoding success of all models aligned with known attentional modulation of ERPs, while the forward model decoder exhibited stronger alignment to the N1 peak-related AMI. These findings demonstrate how fixed temporal structure and colocation provide a testbed linking attention decoding to underlying neural mechanisms.


  • Open Access

    Semantic Knowledge Guided Diffusion Model for Biomedical Ontology Definition Generation

    by Shailesh Dahal, Ratri Mukherjee, Nicholas Mathews, Bijaya Adhikari, Kishlay Jha

    Biomedical ontologies such as the Unified Medical Language System (UMLS) play a crucial role in supporting practical applications such as predicting gene–disease associations, drug–drug interactions, and biomedical question answering. However, automatically maintaining and updating them remains a critical barrier as authoring high-quality textual definitions is both time-consuming and labor-intensive. As a result, UMLS ontology updates significantly lag behind the pace of scientific discovery, leaving many concepts without definitions or with outdated descriptions. Despite the pressing need for automated generation of UMLS concept definitions, this problem has largely remained unaddressed in prior research. To address this, we propose a novel diffusion-based approach for generating high-quality definitions of biomedical concepts. A key innovation in our approach is the adaptive semantic modulation mechanism, which dynamically adjusts the influence of semantic priors throughout the denoising process, enabling the model to flexibly balance local lexical fidelity with global semantic alignment. Extensive experiments on the largest available biomedical ontology dataset demonstrate that the proposed approach both significantly and consistently outperforms strong baseline algorithms across multiple evaluation metrics.

  • Open Access

    Contact Observations from an Intensive Care Unit

    by Hieu Vu, Roger Struble, Philip M. Polgreen, Bijaya Adhikari, Ted Herman

    Location and interaction data for workers in a hospital unit are useful for epidemiological research. This article describes a one-week study measuring contacts between healthcare professionals in a medical intensive care unit. Measurements capture the duration of contact, defined as being within six feet (1.8 ± 0.1 meters) distance between instrumented persons or between persons and selected locations throughout the unit. Within each patient room, measurements distinguish between three places: the sink, the computer, and the vitals monitor. Data from the study, approximately 15 million records, are processed into different formats that facilitate analysis.


  • Open Access

    Summarizing Clinical Notes using LLMs for ICU Bounceback and Length-of-Stay Prediction

    by Akash Choudhuri, Philip Polgreen, Alberto Segre, Bijaya Adhikari

    Recent advances in the Large Language Models (LLMs) provide a promising avenue for retrieving relevant information from clinical notes for accurate risk estimation of adverse patient outcomes. In this empirical study, we quantify the gain in predictive performance obtained by prompting LLMs to study the clinical notes and summarize potential risks for downstream tasks. Specifically, we prompt LLMs to generate a summary of progress notes and state potential complications that may arise. We then learn representations of the generated notes in sequential order and estimate the risks of patients in the ICU getting readmitted in ICU after discharge (ICU bouncebacks) and predict the overall length of stay in the ICU. Our analysis in the real-world MIMIC III dataset shows performance gains of 7.17% in terms of AUC-ROC and 14.16% in terms of AUPRC for the ICU bounceback task and 2.84% in terms of F-1 score and 7.12% in terms of AUPRC for the ICU LOS Prediction task. This demonstrates that the LLM-infused models outperform the approaches that only directly rely on clinical notes and other EHR data.


  • Open Access

    Domain Knowledge Augmented Contrastive Learning on Dynamic Hypergraphs for Improved Health Risk Prediction

    by Akash Choudhuri, Hieu Vu, Kishlay Jha, Bijaya Adhikari

    Accurate health risk prediction is crucial for making informed clinical decisions and assessing the appropriate allocation of medical resources. While recent deep learning based approaches have shown great promise in risk prediction, they primarily focus on modeling the sequential information in Electronic Health Records (EHRs) and fail to leverage the rich mobility interactions among health entities. As a result, the existing approaches yield unsatisfactory performance in downstream risk prediction tasks, especially tasks such as Clostridioides difficile Infection (CDI) incidence prediction that are primarily spread through mobility interactions. To address this issue, we propose a new approach that leverages Hypergraphs to explicitly model mobility interactions to improve predictive performance in health risk prediction tasks. Unlike regular graphs that are limited to modeling pairwise relationships, hypergraphs can effectively characterize the complex high-order semantic relationships between health entities. Moreover, we introduce a new contrastive learning strategy that exploits the domain knowledge to generate semantically meaningful positive (homologous) and negative (heterologous) pairs needed for contrastive learning. This unique contrastive pair augmentation strategy boosts the power of contrastive learning by generating feature representations that are both robust and well-aligned with the domain knowledge. Experiments on two real-world datasets demonstrate the advantage of our approach in both short-term and long-term risk prediction tasks, such as Clostridioides difficile infection incidence prediction and MICU transfer prediction. Our framework obtains gains in performance up to 29.49 % for PHOP, 30.64 % for MIMIC-IV for MICU transfer prediction, 13.17 % for PHOP, and 4.45 % for MIMIC-IV for CDI Incidence Prediction.