MD-SOAR Maryland Shared Open Access Repository
Not a member yet
34521 research outputs found
Sort by
DUNE: A Machine Learning Deep UNet++ based Ensemble Approach to Monthly, Seasonal and Annual Climate Forecasting
No full textCapitalizing on the recent availability of ERA5 monthly averaged, long-term data records of mean atmospheric and climate fields derived from high-resolution reanalysis, deep- learning architectures provide an alternative to physics-based daily numerical weather predictions for subseasonal-to-seasonal (S2S) and annual forecasts. A novel Deep UNet++-based Ensemble (DUNE) neural architecture is introduced, incorporating encoder-decoder structures with residual blocks. When initialized with data from the prior month, season, or year, this architecture delivers an AI-based global forecasts for monthly, seasonal, and annual means of T2m and SST. ERA5 monthly mean data are utilized as inputs for T2m over land, SST over oceans, and climatological monthly solar radiation at the top of the atmosphere, covering 40 years of data to train the model. Validation forecasts are conducted for another two years, followed by five years of forecast evaluations to capture natural annual variability. Rigorous testing was performed using a cross-validation approach with multiple validation and testing periods. The DUNE-trained inference weights enable forecasts to be generated within seconds. Performance metrics such as RMSE, ACC, and HSS are analyzed globally and across specific regions. DUNE AI’s forecasts outperform persistence, climatology, and multiple linear regression across all domains. DUNE forecasts demonstrate comparable statistical accuracy to NOAA’s operational monthly outlooks for the U.S., but at significantly higher spatial resolutions. RMSE seasonal comparisons with NOAA’s NMME and ECMWF’s SEAS5 show that DUNE outperforms both in most seasons and captures major anomalies with finer spatial detail.We want to express our gratitude to the NASA/ESTO Firesense Program Manager, Haris Riris, and his staff for their support on our grant number 80NSSC22K1405, which has allowed graduate student support to pursue this critical research project. Their belief in the significance of this AI/ML study in advancing knowledge in seasonal and annual prediction will be instrumental in addressing the impact and risk of wildfires on climate change. We would also like to acknowledge the support from the NASA HPC office in making the GSFC/NCCS computing facility available to this grant, without which these breakthrough findings would not have been possible. Moreover, we would like to thank the NCCS staff for their professional management in providing reliable system support and access to their advanced machine learning system. That support has been vital in facilitating our data aggregation, training, and predictions of the extensive number of experiments needed to test and evaluate this unique DUNE AI/ML Earth system forecasting model. Finally, we wish to recognize Prof. Karuna Joshi and the support provided by the UMBC NSF-funded Center for Accelerated Real Time Analytics (CARTA) for providing their Computing and Laboratory resources to support this machine learning collaborative research study.https://journals.ametsoc.org/view/journals/aies/aop/AIES-D-24-0073.1/AIES-D-24-0073.1.xm
Decentralised Resource Sharing in TinyML: Wireless Bilayer Gossip Parallel SGD for Collaborative Learning
No full textWith the growing computational capabilities of microcontroller units (MCUs), edge devices can now support machine learning models. However, deploying decentralised federated learning (DFL) on such devices presents key challenges, including intermittent connectivity, limited communication range, and dynamic network topologies. This paper proposes a novel framework, bilayer Gossip Decentralised Parallel Stochastic Gradient Descent (GD PSGD), designed to address these issues in resource-constrained environments. The framework incorporates a hierarchical communication structure using Distributed Kmeans (DKmeans) clustering for geographic grouping and a gossip protocol for efficient model aggregation across two layers: intra-cluster and inter-cluster. We evaluate the framework's performance against the Centralised Federated Learning (CFL) baseline using the MCUNet model on the CIFAR-10 dataset under IID and Non-IID conditions. Results demonstrate that the proposed method achieves comparable accuracy to CFL on IID datasets, requiring only 1.8 additional rounds for convergence. On Non-IID datasets, the accuracy loss remains under 8\% for moderate data imbalance. These findings highlight the framework's potential to support scalable and privacy-preserving learning on edge devices with minimal performance trade-offs.http://arxiv.org/abs/2501.0481
Learning-Based Thrust Regulation of Solid-Fuel Ramjet in Flight Conditions
No full textThis paper investigates the performance of a learning-based control system for regulating the thrust generated by a solid fuel ramjet engine in realistic flight scenarios. An integrated simulation framework is developed that combines a longitudinal missile dynamics model, a missile autopilot, a quasi-static engine dynamics model, and a learning controller for thrust regulation. The missile autopilot is based on the classical three-loop topology. The learning controller is an adaptive PID controller whose gains are recursively optimized using the retrospective cost adaptive control algorithm. First, harmonic acceleration commands are used to simulate variable flight conditions that affect the thrust generated by the engine model. Next, an interception scenario is simulated by integrating a guidance law in the loop. Numerical results indicate that the learning controller can regulate the generated thrust despite wide variations in operating conditions.https://arc.aiaa.org/doi/10.2514/6.2025-280
Signal Processing of Images for Convective Boundary Layer Height Estimation from Radar (SPICER) and multi-instrument verification
No full textThe study of the planetary boundary layer (PBL) is one of the main topics of the atmospheric community. The current study presents a new algorithm for PBL height determination using a publicly available but unexplored data source, the Weather Service Radar (WSR-88D). The diurnal evolution of the PBL is also known as Convective Boundary Layer (CBL), key in the study of convection and precipitation. This paper presents the Signal Processing of Images for Convective Boundary Layer Height Estimation (SPICER) algorithm that can automatically detect the CBL Height (CBLH) for all of the 159 radar locations across the United States during clear days. The present work is the first step to applying SPICER to a network of Next Generation Radars (NEXRAD) with continuous countrywide coverage. With the possible combination with the Automated Surface Observing System network (ASOS), a source of ceilometer profile data, a validated dataset of CBLH estimates can be expected soon. The algorithm treats averaged differential reflectivity vs range as an image and applies filtering plus Canny edge detection to estimate the CBLH. In addition, another algorithm is presented to automate the detection of the mixing layer height (MLH), a proxy for CBLH from Raman Lidar and a 915 MHz wind profiler. A comparison of CBLH estimates vs widely used methods in meteorology (Radiosondes, Raman Lidar, ceilometer, 915 MHz wind profiler, and Doppler Lidar-based derived Value-Added Product (VAP) ) is performed to validate the NEXRAD detected CBLH using SPICER. The SPICER algorithm shows over 0.9 correlation with radiosonde measurements.We would like to acknowledge John Banghoff, David Stensrud, and Matthew Kumjian from Penn State University for their initial work on PBL estimation using NEXRAD and a valuable discussion. Demoz was partially supported by a grant from NOAA Center for Atmospheric Science—Meteorology (NCASM II), which is funded by the U.S. Department of Commerce, National Oceanic and Atmospheric Administration, Educational Partnership Program under Agreement No. NA22SEC4810015https://ieeexplore.ieee.org/abstract/document/10839448
Eyeballing Combinatorial Problems: A Case Study of Using Multimodal Large Language Models to Solve Traveling Salesman Problems
No full textMultimodal Large Language Models (MLLMs) have demonstrated proficiency in processing di-verse modalities, including text, images, and audio. These models leverage extensive pre-existing knowledge, enabling them to address complex problems with minimal to no specific training examples, as evidenced in few-shot and zero-shot in-context learning scenarios. This paper investigates the use of MLLMs' visual capabilities to 'eyeball' solutions for the Traveling Salesman Problem (TSP) by analyzing images of point distributions on a two-dimensional plane. Our experiments aimed to validate the hypothesis that MLLMs can effectively 'eyeball' viable TSP routes. The results from zero-shot, few-shot, self-ensemble, and self-refine zero-shot evaluations show promising outcomes. We anticipate that these findings will inspire further exploration into MLLMs' visual reasoning abilities to tackle other combinatorial problems.https://arxiv.org/abs/2406.0686
Topological X-states in a quantum impurity model
No full textTopological qubits are inherently resistant to noise and errors. However, experimental demonstrations have been elusive as their realization and control is highly complex. In the present work, we demonstrate the emergence of topological X-states in the long-time response of a locally perturbed quantum impurity model. The emergence of the double-qubit state is heralded by the lack of decay of the response function as well as the out-of-time order correlator signifying the trapping of excitations, and hence information in local edge modes.This work was supported by Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq), Brazil, through grant No. 200267/2023-0. M.V.S.B. acknowledges the support of CNPq, under Grant No. 304120/2022- 7. E.M. acknowledges the support of CNPq, under Grant No. 309584/2021-3. The work was also financed (M.V.S.B. and E.M.), in part, by the São Paulo Research Foundation (FAPESP), Brazil, Process Number 2022/15453-0. S.D. and M.F.C. acknowledge support from the John Templeton Foundation under Grant No. 62422.http://arxiv.org/abs/2501.1391
Nursing Facility Characteristics Are Differentially Associated With Family Satisfaction and Regulatory Star Ratings
No full textResearch suggests that nursing facility structural characteristics are important contributors toward residents’ quality of care. We use 2021 data from 220 Maryland nursing facilities to examine associations between two different quality-of-care metrics: family satisfaction and Care Compare five-star quality ratings. We used descriptive statistics to explore differences in quality metrics across facility ownership (for-profit vs. non-profit), geographic location (urban vs. rural), and resident census (1–60, 61–120, and 121+). Relationships were examined across overall ratings, as well as across subdomains of the two frameworks (e.g., staffing). Family members of residents in non-profit, rural, and low-census facilities rated facilities higher. Non-profit and low-resident census facilities were more likely to be rated four or five stars, while no significant association was observed across geographic location, or interactions across structural factors. Findings emphasize the need for comprehensive quality-of-care frameworks that explore quality care across stakeholders and types of facilities.https://journals.sagepub.com/doi/10.1177/0733464824131138
Dynamical drivers of free-tropospheric ozone increases over equatorial Southeast Asia
No full textSAGE III/ISS Science Team Meeting at GA Tech, Sept 12, 202
Understanding the Challenges of Maker Entrepreneurship
No full textACM Human-Comuter InteractionThe maker movement embodies a resurgence in DIY creation, merging physical craftsmanship and arts with digital technology support. However, mere technological skills and creativity are insufficient for economically and psychologically sustainable practice. By illuminating and smoothing the path from ``maker" to ``maker entrepreneur," we can help broaden the viability of making as a livelihood. Our research centers on makers who design, produce, and sell physical goods. In this work, we explore the transition to entrepreneurship for these makers and how technology can facilitate this transition online and offline. We present results from interviews with 20 USA-based maker entrepreneurs {(i.e., lamps, stickers)}, six creative service entrepreneurs {(i.e., photographers, fabrication)}, and seven support personnel (i.e., art curator, incubator director). Our findings reveal that many maker entrepreneurs 1) are makers first and entrepreneurs second; 2) struggle with business logistics and learn business skills as they go; and 3) are motivated by non-monetary values. We discuss training and technology-based design implications and opportunities for addressing challenges in developing economically sustainable businesses around making.We thank all the research participants for their participation in our study. We thank Rachel Liao and Kathy Song who participated in in this project. Quentin Romero Lauro and Courtney Kreitzer who participated in a summer REU related to this research. This material is based upon work supported by the National Science Foundation under Grant No. 2222719.http://arxiv.org/abs/2501.1376
EmoXpt: Analyzing Emotional Variances in Human Comments and LLM-Generated Responses
No full text2025 IEEE 15th Annual Computing and Communication Workshop and Conference (CCWC), 6 - 8 January 2025, University of Nevada, Las Vegas, USA.The widespread adoption of generative AI has generated diverse opinions, with individuals expressing both support and criticism of its applications. This study investigates the emotional dynamics surrounding generative AI by analyzing human tweets referencing terms such as ChatGPT, OpenAI, Copilot, and LLMs. To further understand the emotional intelligence of ChatGPT, we examine its responses to selected tweets, highlighting differences in sentiment between human comments and LLM-generated responses. We introduce EmoXpt, a sentiment analysis framework designed to assess both human perspectives on generative AI and the sentiment embedded in ChatGPT's responses. Unlike prior studies that focus exclusively on human sentiment, EmoXpt uniquely evaluates the emotional expression of ChatGPT. Experimental results demonstrate that LLM-generated responses are notably more efficient, cohesive, and consistently positive than human responses.http://arxiv.org/abs/2501.0659