33760 research outputs found
Sort by
TRACTABLE DESIGN OF DYNAMICAL POLICIES FOR POPULATION GAMES
No full textFor large populations of agents, designing incentives that can guide the agents intomaking better choices and improve their outcome is challenging. Especially when the
choices of the agents affect the evolution of an exogenous system that we wish to control.
The interaction between the two dynamical systems makes it difficult to design incentives
that can change their choices of strategy while satisfying requirements, such as constraints
on the trajectory of the exogenous system. Evolutionary game theory is a natural frame-
work for studying the strategic choices of agents and their learning that determines their
strategy revisions. When a large number of agents noncooperatively select strategies from
a common set, the framework of population games allows for amenable design and analysis.
We investigate the following three problems related to incentive design for a largenumber of agents whose choices influence the evolution of an exogenous system:
Problem 1: We consider populations of agents whose learning rule combines char-acteristics of several learning rules that were previously studied individually. We refer to
such previously studied rules as canonical rules and as hybrid rules the ones that combine
characteristics of the canonical rules. We show that the conic combination of all sensible
δ-passive canonical rules will result in a δ-passive hybrid rule.
Problem 2: We consider agents during an epidemic outbreak scenario and design
incentives to guide them into using strategies that mitigate the transmission of the disease.
We solved for an optimal social state that minimizes the transmission rate given a long-term
budget constraint on the incentive. Then we determined a dynamic payoff mechanism that
leads the population’s strategy distribution to converge to an optimal social state. Our
convergence proof uses a Lyapunov function, which can be used to bound the peak number
of infected agents during the outbreak. The Lyapunov function can be leveraged during
design to select a payoff mechanism that meets a requirement on the peak number of
infected agents.
Problem 3: (a) We generalize the design concept from the second problem to obtain
a more general incentive design method for populations coupled to an exogenous system,
with explicit constraints for the type of exogenous system that can be controlled with the
proposed method, and assuming only that the agents’ learning rule satisfies positive cor-
relation, Nash stationarity, and δ-passivity. We show that the result can be applied to
previous work on epidemic population games, and also to other exogenous systems such
as the Leslie-Gower model for the interaction between populations of hosts and parasites.
Like in the second problem, we proved convergence to a desired equilibrium using a Lya-
punov function, that can be used to bound the transient behavior of the system. But,
in this problem we also obtained a bound on the instantaneous cost of implementing thisincentive. (b) We remove the assumption that the agents’ learning rule satisfies δ-passivity
and consider the agents’ learning rule to satisfy only positive correlation and Nash station-
arity. We design a dynamic incentive and a memoryless incentive that guide a population
to a desirable equilibrium. For the memoryless incentive, we prove convergence using a
Lyapunov function, that can also be used to obtain anytime bound on the states of the
exogenous system and of the population
OPTIMIZING COMMUNICATION IN PARALLEL DEEP LEARNING ON EXASCALE-CLASS MACHINES
No full textDeep learning has made significant advancements across various fields, driven by increasingly larger neural networks and massive datasets. However, these improvements come at the cost of high computational demands, necessitating the use of thousands of GPUs operating in parallel for extreme scale model training. At such scales, the overheads associated with inter-GPU communication become a major bottleneck, severely limiting efficient hardware resource utilization.
This dissertation addresses communication challenges in large-scale parallel training. It develops hybrid parallel algorithms designed to reduce communication overhead, along with asynchronous, message-driven communication methods that enable better overlap of computation and communication. A performance modeling framework is presented to identify communication-minimizing configurations for given workloads. Finally, scalable implementations of latency-optimal collective communication are developed to support efficient training at scale. These contributions improve the performance and scalability of distributed deep learning systems. By tackling these critical communication challenges, this work contributes to more efficient deep learning training at scale, enabling faster model convergence and better resource utilization across large GPU clusters
Essays in Public Economics
No full textThis dissertation investigates behavioral responses to taxation across three key domains: labor supply, migration, and tax evasion. Chapter 1 introduces the three papers that make up this dissertation, and places them in the context of the field of public economics. Chapter 2 examines the labor supply effects of the California Earned Income Tax Credit (CalEITC), a state-level supplement to the federal EITC targeted at very-low-income workers with children. Using a triple-difference approach and American Community Survey data, I find that the CalEITC led to a shift from full-time to part-time work among single mothers, particularly those in larger households, consistent with a discrete model of labor supply. Chapter 3, coauthored with Karen Conway and Jonathan Rork, evaluates how state tax preferences for the elderly affect interstate migration. Leveraging tax return data and variation in state policy over two decades, we find that while overall income taxes slightly affect elderly migration, elderly-specific tax breaks do not, suggesting limited behavioral response to these targeted incentives. Chapter 4, coauthored with Daniel Reck, explores how assumptions about income misreporting affect estimates of top income inequality. By creating a formal structure within which to express the work of several competing approaches to estimating the changes in income inequality over the past 60 years, we can separate out economic and methodological drivers of the differences between these approaches. We also review the current state of the economic literature, concluding that relying on audit data to measure misreporting likely understate inequality growth by failing to account for the rise of pass-through income among the very-wealthy
Provider-to-Provider Communication in Blood and Marrow Transplant: Best Practices in Shared Patient Care to Improve Equity, Access and Outcomes
No full textProvider-to-provider communication (P2PC) across transitions in patient care has gone largely unstudied. No established guidelines, models, or scales measure P2PC directly. P2PC is especially important between hematology-oncology providers and blood and marrow transplant (BMT) physicians for patients with blood cancers like leukemia. When the burden of P2PC falls on the patient, it can cost time, money, and lives. While access to BMT is limited by factors beyond physician control, less than half of patients who may need BMT receive the procedure. Barriers exist across race/ethnicity, geography, and socioeconomic status. However, strong P2PC and improved care coordination may alleviate barriers to care.
In Study 1, I conducted a scoping review and identified only 10 studies measuring P2PC in BMT, with only one directly aiming to assess P2PC. Studies generally supported P2PC as a tactic to overcome barriers to BMT. In Study 2, I combined national datasets to explore quantitative associations between social vulnerability, hematology-oncology provider and BMT physician density, and the unmet need of allogeneic BMT across all U.S. counties (n=3,141). Physician density and social vulnerability were significant predictors of unmet need. I identified Texas, North Carolina, Florida, Nevada, and Georgia as the top states most at-risk for allogeneic BMT access challenges. In Study 3, I used a qualitative case study approach in the Northeastern U.S. (n=15), South Central U.S. (n=15), and key informant interviews across the U.S. (n=12) to identify six themes highlighting P2PC’s ability to facilitate shared care and decision-making, peer education and referral, overcome barriers to BMT, and improve outcomes. I found that P2PC is more challenging when patients and healthcare systems have fewer resources, and that tools and measures can help optimize P2PC and care coordination. Additionally, I identified barriers and facilitators to P2PC in BMT across roles and resources, processes, and relationship building, proposing best practices.
Results solidify the need for a P2PC measure, laying the groundwork for its development. Research assessed emerging and existing models that may apply to P2PC for future theory development. Work can guide education, future research, and interventions in BMT and inform expansion to other fields of medicine
Optimizing Ribozyme Reporters Using RNase J1 in E.coli Based Cell-Free Systems
No full textSynthetic ribozyme reporters are RNA regulatory elements that can be modified to control the expression of downstream reporter proteins such as green fluorescent protein (GFP) in response to the binding of a target ligand. Ligand binding can activate or inhibit ribozyme self-cleavage, regulating the release of a mRNA segment that encodes for a fluorescent protein. RNase J1, a 5’-to-3’ endonuclease from Gram-positive bacteria, enables improved degradation of this protein in E. coli, unlike the native RNase E, which inefficiently degrades the transcript due to a 5’-hydroxyl group. We introduced RNase J1 into E. coli-based cell-free lysates and tested its effect on reporter expression. In a one-pot solution, RNase J1 decreased GFP expression in cell free reactions with the ribozyme reporters irrespective of ligand presence, however it did not decrease the expression of Pbab GFP, the control reporter tested. However, when testing a two-step reaction with titrated crude lysate containing RNaseJ1, we achieved expected performance in which RNaseJ1 decreased the fluorescent signal with the addition of a ligand.
Further testing and optimization are needed to confirm that the decrease in fluorescent signal is both significant and reproducible, followed by additional optimization to increase the magnitude of that decrease
Reporte de salud de 2025 sobre la bahÃa de Chesapeake y su cuenca hidrográfica
No full textEste informe proporciona una evaluación transparente, oportuna, y geográficamente detallada de la bahÃa de Chesapeake y su cuenca. Desde 2016, UMCES ha involucrado a las partes interesadas de toda la cuenca para transformar el informe en una evaluación de la salud de la cuenca de Chesapeake. Los indicadores de la bahÃa evalúan la salud del ecosistema acuático, mientras que los indicadores de la cuenca hidrográfica cubren las condiciones ecológicas, sociales, y económicas. Este es el sexto año que se evalúa la cuenca, y se ha añadido un nuevo indicador ecológico: el estrés térmico.
En general, la cuenca hidrográfica de la bahÃa de Chesapeake obtuvo una puntuación de C+ (57%). Se incluyeron cinco indicadores ecológicos, cuatro indicadores económicos, y tres indicadores sociales.
La nota global de la bahÃa de Chesapeake fue una C (50%), en 2025, un ligero descenso respecto del año anterior, pero aun asà una tendencia a largo plazo de mejora significativa.https://ian.umces.edu/site/assets/files/32679/reporte-de-salud-de-2025-sobre-la-bahia-de-chesapeake-y-su-cuenca-hidrografica.pd
Leveraging additional VIIRS information to improve wildfire tracking in the western US
No full textSee paper publication. This dataset is the list of VIIRS known active fire and candidate fire pixels for large wildfires in the western US, 2020. Column descriptions:
longitude.
latitude.
fire_mask: value of VIIRS level-2 classification, 0-9. <7 indicates candidate fires and 7-9 indicate low, nominal, and high confidence known active fires.
confidence: "x" = candidate, "l" = low, "n" = nominal, "h" = high.
acq_date: string date of observation in format yyyy-mm-dd.
acq_time: string time of observation in format HH:MM UTC.
acq_datetime: python datetime object for date + time, UTC.
j: column index (or x-position) of the pixel in the swath, used for view zenith angle and pixel size.
vza: view zenith angle.
sza: solar zenith angle.
daynight: D or N from VIIRS L2 product.
i750: corresponding colocated i-index in the M bands.
j750: corresponding colocated j-index in the M bands.
frp: fire radiative power calculated here in this study.
frp_old: original fire radiative power calculated in the VIIRS product for known fire pixels.
dist_m13b: distance, in degrees, to the nearest known active fire pixel whose calculated background M13 radiance was used in our FRP calculation for candidate fires.
geometry: shapely geometry column of lat/lon point in parquet file.
satellite: SNPP or NOAA20.
fireid: FEDS fire ID.
startdate: FEDS fire start date.
enddate: FEDS fire end date.
name: common fire name pulled from MTBS for the largest 20 fires.Recent record-breaking fire activity in the western US poses clear threats to humans, ecosystems,
and climate. Larger and faster fires increase the challenges for fire managers and further motivate
the need for improved tracking of extreme fire behavior. There are also known limitations to our
current ability to monitor fires from space. These include infrequent coverage from moderate
resolution (≤ 1 km) sensors, smoke and cloud obscuration, omission of small or low-intensity
fires, and atmospheric attenuation of fire radiative power (FRP). These effects diminish our
ability to quantify fire behavior and emissions, including persistent burning behind the flaming
fire front, particularly in ecosystems with high fuel loads. In this study, we examined the Visible
Infrared Imaging Radiometer Suite (VIIRS) imagery and data products to assess the utility of
candidate fire pixels in addition to the low/nominal/high confidence 375-m fire detections
already included in the active fire product. We found that these candidate pixels added 45% more
daytime detections and 12% more nighttime detections for large fires in the western US 2020 fire
season. Candidate fires were highly consistent with areas of flaming and smoldering fire activity
identified by near-coincident airborne data as well as patterns of known active or candidate fires
in adjacent VIIRS overpasses, without significantly increasing false detections (commission
errors). The candidate fire detections helped fill data gaps due to cloud obscuration during large
fires that generated pyrocumulonimbus (pyroCb) clouds. Including this additional information
also impacted estimates of fire activity, increasing fire persistence by 20% and FRP by 7% across
our sample. Although the contribution from candidate fire detections to total FRP was relatively
small, including these additional pixels could provide a more consistent estimate of fire
emissions for smoke models and air quality forecasts by filling gaps in active fire information
and improving the representation of smoldering fire activity. These results demonstrate the
potential to augment the standard VIIRS product with candidate fire information for known large
fire events to improve fire tracking and downstream products. Such approaches to leverage
additional VIIRS information may be suitable for other biomass burning regions where global
fire detection algorithms provide incomplete information for specific fire types and observing
conditions.This work was supported by funding from the NASA Earth Information System (EIS) Fire
Project and the NASA Earth Science Technology Office (ESTO) FireTech and NASA Wildland
FireSense Programs. TL acknowledges support from the NOAA Climate and Global Change
Postdoctoral Fellowship Program, administered by UCAR’s Cooperative Programs for the
Advancement of Earth System Science (CPAESS) under the NOAA Science Collaboration
Program award NA21OAR4310383. JTR acknowledges support from the US DOE Office of
Science RUBISCO Science Focus Area and NASA's Modeling Analysis and Prediction program
(grant no. 80NSSC21K1362). YC and JTR acknowledge support from the US DOE LLNL-
LDRD program (grant no. DE-AC52-07NA27344 and project no. 22-ERD-008, “Multiscale
Wildfire Simulation Framework and Remote Sensing”).https://doi.org/10.1016/j.rse.2025.11515
The Impact of the atpB Gene on Metabolic and Viral Replication Cycles in E.Coli
No full textBacteriophage has shown to be a promising treatment for bacterial infections as antibiotic resistance becomes more prevalent. The ability of bacteriophage to effectively infect host cells, however, requires the use of the host cell’s metabolic pathways for energy. Previous research has found that certain genetic alterations within these host cell pathways can disrupt bacterial growth and replication efficiency. In our research, we decided to specifically investigate how bacteriophage replication is affected by deletions of the atpB and atpE genes, which code for proton transport proteins involved in E coli’s ATP biosynthesis pathway.
To determine the effects of removing these genes, we first performed streak plates and growth curves with our parent and knockout strains of E. coli to evaluate their impact on bacterial growth. We then ran lysis curves with the parent, ΔatpB, and ΔatpE strains of E. coli using T2 and T4 phage. Finally, we completed a series of plaque assays with the parent and ΔatpB strains and a two time point phage titer experiment to quantify how bacteriophage replication was affected by the knockout gene.
We discovered that bacteriophage replication and overall growth of the E. coli was hindered by the ΔatpB and ΔatpE knockout genes. The streak plates and growth curves showed that the knockout strain of E. coli grew significantly slower than the parent strain. The lysis curves revealed that the atpB and atpE knockout strains exhibited far less phage-driven lysis than the parent strain. Additionally, the plaque assays and two time point experiments with the knockout strains showed no plaques forming, while the parent strain had plaque formation. While the atpB gene is nonessential for E. coli growth, it greatly affects the efficiency and ability for the bacteria to grow. This information points to the fact that bacteriophage cannot be used as an effective treatment in bacterial infections that have mutations affecting the atp genes. The results from our experiment also point to the need for more research to be done on effectively knocking out the atp genes as a potential treatment for controlling bacterial infections. We concluded that effective phage replication is dependent on sufficient ATP availability in the host cells. This finding could be used to help improve the efficiency of phage therapy
Charting the Next Course: The Comprehensive Conservation & Management Plan for Maryland's Coastal Bays (2025–2035)
No full textThe 2025 CCMP is the third in a series of original and revised CCMPs designed to guide the Maryland Coastal Bays Program.https://ian.umces.edu/site/assets/files/32814/charting-the-next-course-the-comprehensive-conservation-management-plan-for-marylands-coastal-bays-2025-2035.pd
Colorimetric Detection of miRNA with DNA-Functionalized Gold Nanoparticles
No full textEarly detection of cancer is important for improving patient prognosis. miRNAs are good biomarkers for cancer screening because they are found in peripheral blood and their dysregulation is associated with cancer development. However, current miRNA detection assays, like RT-PCR, require expensive equipment. We are developing an assay that can be heated and quantified in a portable sensor. The assay involves a rolling circle transcription reaction that activates a Cas12a enzyme in the presence of a miRNA target, resulting in the cleavage of DNA-conjugated gold nanoparticles. This produces a colorimetric output that is quantified using a 3D printed portable detector. The presence of miRNA successfully initiates ligation, rolling circle transcription, and the activation of Cas12a. Successful aggregation of AuNPs occurs in the presence of the linker sequence, while absence of a linker prevents aggregation. However, the Cas12a activation assay must be combined with the AuNP complex to ensure proper aggregation when linker is intact, and lack of aggregation when Cas12a cleaves the linker