MD-SOAR Maryland Shared Open Access Repository
Not a member yet
34521 research outputs found
Sort by
What the Peers Say!
No full textDr. Ramana Vinjamuri is a leading innovator in brain-computer interfaces, neuroprosthetics, and human–machine interaction, with a distinguished academic journey spanning Kakatiya University, Villanova University, the University of Pittsburgh, Johns Hopkins University, and Stevens Institute of Technology, where he received the Harvey N. Davis Distinguished Teaching Award in 2018. Currently an Associate Professor of Computer Science and Electrical Engineering at UMBC, he also serves as the Center Director of the NSF Industry–University Cooperative Research Center (IUCRC) on BRAIN at UMBC, Visiting Scientist at the National Institute on Drug Abuse (NIDA), and Visiting Professor at the Indian Institute of Technology, Hyderabad and MAHE, Manipal, India. Dr. Vinjamuri is the recipient of the prestigious NSF CAREER Award, NSF IUCRC Planning Grant, and the Mary E. Switzer Merit Fellowship, and was recently recognized as one of Technical.ly’s 2025 RealLIST Innovators in Maryland. His pioneering work integrates neural engineering, artificial intelligence, and rehabilitation technologies to create transformative solutions for human health and well-being. I am deeply honored to have the opportunity to interview him and gain insights from his remarkable contributions.https://dtmst.e-geoinfo.com/index.php/dtmst/article/view/
XRISM/Xtend Transient Search (XTS) detected a flux increase of XRISM J1236-6432
No full textAuthors: T. Kiyomoto (Saitama U.), K. Fukushima, Y. Kanemaru, S. Ogawa (JAXA), M. Audard (U. de Geneve), E. Behar (Technion), T. Hakamata (Osaka U.), S. Inoue (Kyoto U.), Y. Ishihara (Chuo U.), C. Kang (Ehime U.), T. Kohmura (TUS), H. Kuramoto (Osaka U.), J. Kurashima (U. of Miyazaki), Y. Maeda (JAXA), H. Matsumoto (Osaka U.), T. Matsushima (U. of Miyazaki), A. Miyamoto (Osaka U.), M. Mizumoto (UTEF), K. Mori (U. of Miyazaki), Y. Motogami (Saitama U.), N. Nagashima (Chuo U.), T. Narita (Kyoto U.), M. Nobukawa (NUE), H. Noda (Tohoku U.), K. Pottschmidt (UMBC, NASA GSFC, CRESST), Y. Sakamoto (Tohoku U.), M. Shidatsu (Ehime U.), H. Sugai (Chuo U.), H. Takahashi (Hiroshima U.), T. Takagi (Ehime U.), S. Takatuska (Osaka U.), R. Takemoto (U. of Miyazaki), Y. Terada (Saitama U.), Y. Terashima (Ehime U.), Y. Tsuboi (Chuo U.), H. Uchida (Kyoto U.), T. Yoneyama (Chuo U.), M. Yoshimoto (Ehime U.), J. SOKOLOSKI (Columbia U.)XRISM/Xtend Transient Search (XTS) detected a gradual increase in the flux of the X-ray source XRISM J1236-6432 from 2025-6-12 to 2025-6-13 TT. The source position is determined to be (R.A., Dec.) = (188.884, -64.528), with a systematic error of ∼ 40 arcsec. A plausible counterpart is the 4XMM J123530.3-643125, which is located ∼ 20 arcsec apart from the position of XRISM J1236-6432. All statistical uncertainties in this report are provided as a 90% confidence level unless stated otherwise.
In the duration of ∼ 80 ks, the flux increased several times. The average flux is calculated as 5 × 10⁻¹³ erg s⁻¹ cm⁻² (0.4 – 10.0 keV), using an absorbed power-law model. This result is one order of magnitude greater than the catalog flux of the 4XMM J123530.3-643125. The photon index and column density are derived as 1 and < 3 × 10²⁰ cm⁻², respectively.
We derived the above systematic error for the flux by comparing our derived values for the sources detected with XTS in several observations with those for the corresponding X-ray counterparts. We estimated the systematic error for the source position from the separations between the detected sources with the corresponding counterparts in the same field of view.https://www.astronomerstelegram.org/?read=1723
Large Language Models (LLMs) as Traffic Control Systems at Urban Intersections: A New Paradigm
No full textThis study introduces a novel approach for traffic control systems by using Large Language Models (LLMs) as traffic controllers. The study utilizes their logical reasoning, scene understanding, and decision-making capabilities to optimize throughput and provide feedback based on traffic conditions in real time. LLMs centralize traditionally disconnected traffic control processes and can integrate traffic data from diverse sources to provide context-aware decisions. LLMs can also deliver tailored outputs using various means such as wireless signals and visuals to drivers, infrastructures, and autonomous vehicles. To evaluate LLMs’ ability as traffic controllers, this study proposed a four-stage methodology. The methodology includes data creation and environment initialization, prompt engineering, conflict identification, and fine-tuning. We simulated multi-lane four-leg intersection scenarios and generated detailed datasets to enable conflict detection using LLMs and Python simulation as a ground truth. We used chain-of-thought prompts to lead LLMs in understanding the context, detecting conflicts, resolving them using traffic rules, and delivering context-sensitive traffic management solutions. We evaluated the performance of GPT-4o-mini, Gemini, and Llama as traffic controllers. Results showed that the fine-tuned GPT-mini achieved 83% accuracy and an F1-score of 0.84. The GPT-4o-mini model exhibited a promising performance in generating actionable traffic management insights, with high ROUGE-L scores across conflict identification of 0.95, decision making of 0.91, priority assignment of 0.94, and waiting time optimization of 0.92. This methodology confirmed LLMs’ benefits as a traffic controller in real-world applications. We demonstrated that LLMs can offer precise recommendations to drivers in real time including yielding, slowing, or stopping based on vehicle dynamics. This study demonstrates LLMs’ transformative potential for traffic control, enhancing efficiency and safety at intersections.https://www.mdpi.com/2624-8921/7/1/1
Summertime Diurnal Variability of Formaldehyde Over the Contiguous United States: Constraints From Pandonia Global Network
No full textThe diurnal cycle of Formaldehyde (HCHO) provide critical insights into tropospheric photochemistry. Here we use tropospheric HCHO retrievals from the Pandonia Global Network (PGN), combined with NASA's GEOS Composition Forecast (GEOS-CF) model to understand the diurnal variation of summertime HCHO across the contiguous US. While PGN HCHO tropospheric column (HCHOₜᵣₒₚ) shows a weak diurnal cycle in most regions, a distinct midday peak is found at the urban sites of Southern US (mainly in Houston), likely driven by highly reactive VOC emissions. For the vertical profile of HCHO mixing ratio within the Planetary Boundary Layer (PBL), PGN shows a significant decrease from 0.5 to 2 km while GEOS-CF exhibits an excessively well-mixed vertical shape. This discrepancy in HCHO profile leads to an overestimated HCHOₜᵣₒₚ in GEOS-CF in Northeast Coastal US and Southeast US. These findings offer valuable insights for interpreting geostationary satellite observations and understanding model biases in surface ozone (O₃).Tianlang Zhao and Jingqiu Mao acknowledge NASA grants 80NSSC19M0154 and 80NSSC21K0428, and UAF Troth Yeddha’ PhD fellowship, for supporting this work. Jennifer Kaiser acknowledges the NASA Grant80NSSC21K0944. We thank the helpful discussions with Jeffrey Geddes from Boston University, Luke Valin from USEPA, William R Simpson from University of Alaska Fairbanks, Katherine Travis, James Crawford, Laura Judd from NASA’s Langley Research Center, Xiaoming Jin from Rutgers, The State University of New Jersey. We also thank all other helpful discussions in the HAQAST TEMPO for Ozone Tiger Team meetings. We acknowledge the Alaska EPCOR for the travel grant. We thank the Pandonia Global Network, and all the Pandora PIs, for maintaining the network and providing data used in this study.https://onlinelibrary.wiley.com/doi/abs/10.1029/2025GL11603
Assurance of Machine Learning for Human-Robot Interaction
No full textThe incredible advancements in artificial intelligence over the past decade have enabled technologies that once lived in research labs to now interact with users from all walks of life. As these agents evolve digitally and expand their physical presence through robotics, the risks associated with human interaction grow—necessitating stronger assurances. These risks stem from the inherent difficulty of deploying machine learning models, which must sense and interpret dynamic environments and human behavior, compared to more predictable, classical software systems. This thesis explores how deep learning can enhance human-robot interaction (HRI) by enabling general, flexible representations that support robust and unconstrained language grounding. Through the development of a neural object representation system, I demonstrate improved performance over prior category-based methods on a challenging, crowd-sourced dataset. Building on this, I introduce joint language-vision modeling, which further enhances generalization and usability, and extends the system to operate directly on speech—broadening accessibility for diverse user populations. However, the generalization power of deep learning introduces new challenges, especially in safety-critical scenarios involving physically embodied robots. To address this, I propose a data-centric threat model for adversarial attacks on vision systems, exposing the limitations of existing defenses. Extending this analysis to human-sensing systems, I identify disparities in adversarial robustness, particularly for users with diverse speech characteristics. Through a comprehensive case study, I show that while robustness training often entails performance trade-offs, rejectionbased defenses—augmented through sampling—can achieve a better balance between robustness, performance, and equity. Finally, I revisit concept-based learning through the lens of assurance, introducing end-to-end differentiable neurosymbolic reasoning to align neural perception with symbolic tasks in both vision and speech. These methods improve interpretability, robustness, and fairness, while enabling alignment verification. Collectively, this work reflects a broader methodology: advancing capabilities, quantifying emerging risks, and designing mitigations that inform new paradigms for assured AI. This cycle—of innovation, analysis, and refinement—serves as a foundation for developing safe, equitable, and assured AI systems
Quantifying the Impacts of Dynamic Lapse Regimes on Snow Simulations over Complex Terrains
No full textAccurate characterization of gridded meteorological distributions in complex terrain—especially relationships between meteorological fields and altitude—is essential for simulating snowpack dynamics. This is challenging due to sparse long-term observations and strong spatial variability in near-surface meteorology. This study evaluated elevation-based dynamic lapse rate corrections—which account for local, hourly variations in temperature, pressure, humidity, and longwave radiation—against the commonly assumed static lapse rate of -6.5°C km⁻¹ in a 1-km resolution land surface model. We examined snow water equivalent (SWE), snow depth (SD), snow cover (SC), and 2-m air temperature (T2) across western U.S. coastal mountains. Both correction methods improved T2 and snow simulations relative to those without correction, with similarly broad improvements over the simulation period (June 2006 through December 2020). Dynamic correction led to marginally broader snow improvements in some regions relative to no correction—for example, in high-elevation Upper Cascade leeward regions, SWE root mean square error improved across 91% of the area with dynamic correction, compared to 89% with static correction. While overall improvement differences between methods were small for the full simulation period, dynamic correction demonstrated clearer advantages during an extreme snow drought year. SC improvements were less pronounced, but dynamic correction better detected snow during the drought year, while static correction performed slightly better during a surplus year. Although future climate scenarios were not modeled, the increasing prevalence of snow droughts highlights the value of this approach for improving snowpack simulations under a warming climate.https://journals.ametsoc.org/view/journals/hydr/aop/JHM-D-25-0021.1/JHM-D-25-0021.1.xm
Stratospheric Circulation in the Southern Hemisphere: links to tropical winds, ozone and the Hunga Eruption -Part 2: Evidence from a Global Chemistry-Climate Model
No full textThe Southern Hemisphere (SH) stratospheric circulation can be categorized based upon the development of a low-latitude jet (LLJ) in the upper stratosphere during winter months. We analyze the dynamics of the LLJ based on a large ensemble of chemistry-climate model simulations, supported by reanalysis data. The LLJ is associated with westerly wind anomalies in the middle and upper stratosphere during mid-winter, together with weak planetary wave activity and a slower Brewer-Dobson circulation. Circulation effects extend into the tropical upper stratosphere, where the LLJ impacts the tropical semi-annual oscillation (SAO). Additionally, the LLJ is linked to a persistent, strong polar vortex in the lower stratosphere during October–December. This cold, strong vortex is associated with decreased ozone amounts in winter and spring; ozone radiative feedbacks reinforce the cold vortex after sunlight returns in October. The 2022 Hunga eruption coincided with an anomalously strong LLJ year, and ensemble simulations of Hunga impacts show that the eruption pushed the SH winter circulation towards LLJ behavior, although the ensemble-mean forced Hunga signal is small and embedded within a large amount of stochastic variability. These results advance our understanding of how LLJ dynamics connect to the large-scale stratospheric circulation and ozone depletion, with implications for predicting polar climate and composition.NCAR's Community Earth System Model project is supported primarily by the National Science Foundation (NSF) under Cooperative Agreement No. 1852977. Computing and data storage 47 resources, including the Derecho supercomputer (doi:10.5065/qx9a-pg09), were provided by the Computational and Information Systems Laboratory (CISL) at NCAR. XW’s model simulations were supported through NSF NCAR Strategic Capability (NSC) projects and by the help of Garth D’Attilo. WY’s work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344. YZ acknowledges the National Oceanic and Atmospheric Administration (grant nos. 03-01-07-001, 60 NA17OAR4320101, and NA22OAR4320151). VLH acknowledges support from NASA grants 80NSSC22K1074, 80NSSC22K0017, 80NSSC23K0848, and 80NSSC24K0269 and NSF grant AGS2411429.https://www.authorea.com/users/770423/articles/1339325-stratospheric-circulation-in-the-southern-hemisphere-links-to-tropical-winds-ozone-and-the-hunga-eruption-part-2-evidence-from-a-global-chemistry-climate-model?commit=13040505d85a59c90ce8d01060b3dca70828c38
Improving Object Classification Accuracy from Electromagnetic Data Using Attention Mechanisms
No full text26th International Conference on Electromagnetics in Advanced Applications (ICEAA 2025), September 8-12, 2025, Palermo, ItalyObject classification using electromagnetic waves iscrucial in various applications, including remote sensing, security screening, and biomedical imaging. However, accurately classifying arbitrarily oriented objects from electromagnetic scatteringdata remains a significant challenge. In this work, we propose an attention-based machine-learning framework designed to improve the robustness and accuracy of electromagnetic object classification. Our model leverages an attention mechanism to focus on the most informative scattering features dynamically, enabling enhanced feature extraction and improved generalization across different object orientations. We demonstrate the effectiveness of attention-based models in enhancing object classification robustness using a numerical dataset, showing that the proposed method outperforms conventional machine learning models regarding classification accuracy.https://userpages.cs.umbc.edu/simsek/cps/2025_ICEAA_Attention.pd
Unprecedentedly bright X-ray flaring in Cygnus X-1 observed by INTEGRAL
No full textWe study three extraordinarily bright X-ray flares originating from Cyg X-1 seen on 2023 July 10 detected with INTEGRAL. The flares had a duration on the order of only ten minutes each, and within seconds reached a 1-100 keV peak luminosity of 1.1-2.6 × 10³⁸ erg/s⁻¹. The associated INTEGRAL/IBIS count rate was about ~10x higher than usual for the hard state. To our knowledge, this is the first time that such strong flaring has been seen in Cyg X-1, despite the more than 21 years of INTEGRAL monitoring, with almost ~20 Ms of exposure, and the similarly deep monitoring with RXTE/PCA that lasted from 1997 to 2012. The flares were seen in all three X-ray and γ-ray instruments of INTEGRAL. Radio monitoring by the AMI Large Array with observations 6 h before and 40 h after the X-ray flares did not detect a corresponding increase in radio flux. The shape of the X-ray spectrum shows only marginal change during the flares, i.e., photon index and cut-off energy are largely preserved. The overall flaring behavior points toward a sudden and brief release of energy, either due to the ejection of material in an unstable jet or due to the interaction of the jet with the ambient clumpy stellar wind.We especially acknowledge the crucial contribution of Katja Pottschmidt – not only to this paper but the field of Black-hole timing in general. Without her support, mentorship, and scientific insight this work would not have been possible. Her untimely passing is felt sorely. This work has been partially funded by the Bundesministerium für Wirtschaft und Klimaschutz under Deutsches Zentrum für Luft- und Raumfahrt grant 50 OR 1909. This research is supported by the DFG research unit FOR 5195 ‘Relativistic Jets in Active Galaxies’ (project number 443220636, grant number WI 1860/20-1). TB & JR acknowledge partial funding from the French Space Agency (CNES). The material is based upon work supported by NASA under award number 80GSFC24M0006. MP acknowledges support by the Spanish Ministry of Science trough Grant PID2022-136828NB-C43, and by the Generalitat Valenciana through grant CIPROM/2022/49. The research is based on observations with INTEGRAL, an ESA project with instruments and science data center funded by ESA member states (especially the PI countries: Denmark, France, Germany, Italy, Switzerland, Spain) and with the participation of Russia and the USA. This research has made use ISIS 1.6.2-51 (Houck & Denicola 2000) and of ISIS functions (ISISscripts) provided by ECAP/Remeis observatory and MIT (https://www.sternwarte.uni-erlangen.de/isis/).http://arxiv.org/abs/2508.2087
Facilitating Online Healthcare Support Group Formation Using Topic Modeling
No full textPatients increasingly seek peer support in online health forums; however, the large-scale and inconsistent engagement patterns of these forums often fail to meet patients’ support needs effectively. Smaller, personalized support groups could address these challenges by tailoring interactions to users’ shared experiences and demographics. This study introduces the Group-specific Dirichlet Multinomial Regression (gDMR) model, a structured framework for automating and personalizing support group formation using user-generated content, demographic, and interaction data. By incorporating group-specific parameters and node embeddings, gDMR captures nuanced demographic and behavioral patterns, extending traditional Dirichlet Multinomial Regression (DMR). Experiments demonstrate gDMR’s ability to form groups that are more semantically coherent and contextually relevant than those produced by baseline models. This scalable model reduces manual effort in group personalization, fostering inclusive engagement and enhancing patient-centered care. Findings highlight gDMR’s potential as a framework for digital healthcare platforms, from online health communities to healthcare systems utilizing electronic health records (EHRs), advancing health informatics through support group formation.https://ebooks.iospress.nl/doi/10.3233/SHTI25099