71446 research outputs found
Sort by
Large Language Models for Automated Evaluation of Radiology Reports with Fine-Grained Scoring
No full textThe current gold standard for evaluating generated chest x-ray (CXR) reports is through radiologist annotations. However, this process can be extremely time-consuming, especially if there are large numbers of reports to evaluate. In this work, we present a Large Language Model (LLM)-based automated evaluation metric for generated CXR reports called FineRadScore. Given a candidate and a ground truth report, FineRadScore gives the minimum number of line by line corrections required to go from the candidate to the ground truth report. Additionally, FineRadScore assigns a severity rating for each correction and generates comments regarding why the correction was needed. We demonstrate that FineRadScore is able to generate the corrections in a way that aligns with radiologists and has an understanding of how clinically meaningful each error is. We also demonstrate that, when used to get a sense of the quality of the report as a whole, it aligns with radiologists at a similar level to current state of the art automated CXR evaluation metrics. Finally, we analyze FineRadScore's shortcomings to pave the way for future works.Computer Scienc
A Novel Algorithm for Calculating Explicit Sampling Probabilities
No full textReinforcement learning (RL) is a branch of machine learning that tackles sequential decision making problems via an agent-environment framework, with the objective of maximizing a scalar reward signal. Many online reinforcement learning (RL) algorithms lack explicit sampling probabilities, the probability that an action is selected given the observed history up to that point. These explicit sampling probabilities are necessary for calculating estimators used in off-policy evaluation. Our primary contribution is the development of a Monte Carlo integration (MCI) based algorithm that closely matches the performance of randomized least squares value iteration (RLSVI), an efficient Bayesian RL algorithm that does not offer explicit sampling probabilities. Moreover, we present an application of this algorithm in the context of the ADAPTS-HCT clinical trial, which uses a novel hierarchical RL algorithm in a dyadic patient-caregiver structure to improve medication adherence following hematopoietic stem cell transplantation.Computer Scienc
Structural and Seismologic Characterization of the Newport-Inglewood fault of Los Angeles, California, and the Seattle Fault Zone and basin edge of Seattle, Washington: Implications for urban seismic hazard assessment
No full textActive fault systems located in densely-populated urban settings, especially those set within sedimentary basins that are known to amplify earthquake ground motions, pose some of the greatest seismic hazards in the world. Understanding these faults’ 3D structure and subsurface geometries is a critical element of understanding these hazards, including the expected patterns of surface faulting and levels of strong seismic shaking during future earthquakes. In this work, we characterize two active, complex, and hazardous fault systems located in densely populated U.S. cities. We define the 3D geometry of the Newport-Inglewood fault in the Los Angeles basin, California, using an unprecedented dataset of surface trace maps, oil wells, geophysical surveys, and seismicity. In contrast to simplified representations of the fault in current hazard maps and fault models, we show that the system contains 300 distinct fault segments that define a broad (up to 8 km wide) zone of surface faulting that poses significant fault displacement hazards. In addition, we show that the complex network of fault segments contains several potential earthquake gates that may accommodate or limit ruptures, thus informing our assessment of future earthquake magnitudes.
Using an array of densely-spaced nodal seismometers we deployed as part of this study, we help define the location of the Seattle fault zone and basin edge by resolving seismological effects with high spatial resolution. We pinpoint where the SFZ scatters and diffracts earthquake waves, as well as where seismic wave velocities measurably change in the SFZ. We perform P wave polarization analysis and ambient noise cross-correlation processing to constrain shallow shear and surface wave velocities, which are critical values for seismic hazard and geotechnical assessments. These results help us understand the future basin-edge effects expected to contribute to strong seismic shaking in future earthquakes in Seattle.
The Newport-Inglewood fault (NIF)
The Newport-Inglewood fault (NIF) is a complex, active right-lateral strike-slip fault system that extends over 60 km in length across metropolitan Los Angeles, California. The NIF forms the western boundary of the central Los Angeles sedimentary basin, and in current hazard maps has a complex pattern of disconnected active surface fault traces that are distributed over a broad area (USGS QFaults). The southern portion of the NIF generated the 1933 M 6.4 Long Beach earthquake, the deadliest seismic event in southern California history. The California Division of Mines and Geology (1988) and other sources describe a future earthquake on the NIF as one of the greatest hazards to life and property in the United States.
Along the NIF are ten large anticlines, which include some of the largest and most productive oil fields in southern California. These oil fields have been intensively explored by petroleum operations for over a century, which now provides one of the densest and highest-resolution subsurface datasets for any active fault system in the world. Many other geologic and seismologic datasets help define the NIF, including: mapped surface fault traces, 31 2D seismic reflection profiles that cross the fault system, comprehensive regional seismicity catalogs that collectively include over 1 million earthquakes, and petroleum industry contour maps of subsurface rock units deformed by the fault system. These datasets span a remarkable depth range, from the Earth’s surface down to a depth of ~20 km. We integrate and analyze all of these datasets in a computer-aided design (CAD) modeling environment to build a comprehensive 3D representation of the fault system.
We find that the NIF is comprised of 300 distinct, individual faults. These faults fall generally into two classes: large, NW-trending strike-slip faults, many of which are over 10 km in length, and smaller linking faults, which have diverse orientations and physically connect the strike-slip faults. The complete NIF system extends over a broad zone of surface faulting, reaching over ~8 km in some areas. This fault zone width is largely maintained at depth. Faults do not coalesce into a narrow master system—rather, large throughgoing strike-slip faults extend to the base of the seismogenic crust (~15 km depth), interpenetrating each other, and forming a ~5-8 km wide zone of throughgoing strike-slip faults at the base of the seismogenic crust.
The 3D fault geometry of the NIF resolved in our model has critical implications for seismic hazard in Los Angeles. Ground surface rupture is defined by the extent of potentially active faults that may rupture the surface in an earthquake on a given fault system. Potential rupture magnitude is dictated by the total available surface area of viable fault rupture patches, and by the connectivity of distinct fault planes, which can enable ruptures to pass across multiple fault segments. Surface rupture on complex, multi-fault systems pose significant fault displacement hazards, particularly to densely-populated urban settings, such as metropolitan Los Angeles. Our results suggest that future earthquakes on the NIF may involve many more fault segments, spread over a much wider area, with the potential for surface ground rupture than included in current hazard assessments. Given the urban setting of the NIF, this presents significant fault displacement hazards to buildings and other critical civic infrastructure. The width of fault zones at depth also has important implications for the potential locations of future earthquakes, which in turn may influence the intensity of future earthquake ground motions. Furthermore, the presence of multiple active fault splays at depth implies that interseismic strain accumulation and slip are partitioned across more than a single fault, which has implications for determining fault slip rates and earthquake recurrence rates.
We apply a particular focus to the Long Beach segment of the NIF, where the main fault strand takes two left steps, forming two prominent surface uplifts, i.e. the Long Beach restraining bend system. It is thought that the 1933 M 6.4 Long Beach earthquake, which nucleated to the ~25 km southeast, arrested here. We assemble a particularly dense and diverse array of independent datasets across the Long Beach system, and construct a robustly-constrained 3D model of the dual restraining bend there. We resolve an interconnected fault model, defined by local south-dipping reverse faults that physically link large, throughgoing strike-slip faults, and translate regional strike-slip into local dip-slip and uplift. We also map in detail a faulted and folded Pliocene-age horizon, and then restore this horizon to its undeformed state. This restoration is constrained by the precisely mapped dip-slip offsets across the linking reverse faults. This horizon restoration allows us to measure with precision how strike-slip has been partitioned across distinct faults in the Long Beach fault segment of the NIF across geologic time. This result has useful applications for slip rate calculations and understanding how such calculations might spatially vary across a multi-fault system. Furthermore, these types of studies present useful references for comparison and cross-validation with numerical models of dynamic rupture simulations. Running such simulations on a robustly constrained, realistic fault model, and comparing it to geologically observed slip partitioning, will allow us to better assess how earthquakes may or may not tend to follow certain slip pathways. This, in turn, will help us to assess the sizes, locations, and faulting styles of future earthquakes on the NIF, which serve as the basis for defining earthquake rupture forecasts and hazards assessments.
The Seattle fault zone (SFZ)
The Seattle fault zone (SFZ) is an active, blind thrust fault system located in metropolitan Seattle, Washington. It is a multi-stranded thrust fault that forms the southern boundary of the Seattle sedimentary basin. The last known earthquake to have ruptured on the SFZ occurred in 923-924 AD. The SFZ doubles as the Seattle basin’s southern edge, and as such it is expected to focus, trap and amplify seismic waves in the next large earthquake on the system. An earthquake on the SFZ poses the greatest known risk to the city of Seattle (City of Seattle Office of Emergency Management, 2021).
To better understand the role the SFZ and Seattle basin edge will play in future earthquake ground motion amplification, we deploy 100 nodal seismometers in 4 densely-spaced (~250 m interstation distance) transects across the system, and record one month of seismic ambient noise in July-August, 2019. This deployment was the result of a months-long public outreach campaign. All seismometers were voluntarily hosted in Seattle resident backyards, and families and students participated in most sensor installments.
Two regional earthquakes (M 2.8 and M 3.0) and one teleseism (M 7.2) occurred while the nodal array was deployed. In analysis of the regional earthquake recordings, we observe that P waves are scattered as they propagate within the SFZ, but have a clean, linear arrival within the Seattle basin. We also observe a clear increase in S wave speed as waves travel from the Seattle basin into the SFZ. We resolve that S wave horizontal components of motion are significantly diffracted south of the SFZ frontal fault, which is a signature of basin edge effects on the 3D wavefield. Such diffraction requires a sufficiently sharp contrast between bedrock and basin, suggesting that the impedance contrast at the SFZ is sharp enough to scatter these waves. Overall, our analysis of these earthquake recordings reveals detectable elastic structure changes that occur within the SFZ system and Seattle basin, confirming the complex geologic structure of the SFZ near the surface.
We compute shallow shear wave velocities (Vs) at all stations, applying the method of Park and Ishii (2018), using the polarization of the recorded diffracted P wave arrival from the M 7.2 teleseism. We compare these to Vs values reported in prior studies. We compute site amplification at every seismic station location, using the maximum amplitude of squared velocity seismograms windowed around earthquake S wave arrivals. We observe great spatial variability in these amplifications, especially on the horizontal components, up to a factor of 12.
We also perform preliminary ambient noise seismology analysis on our dataset. The nodal recordings of the Seattle array were greatly contaminated by anthropogenic urban noise, specifically spurious signals generated by highways (I-90 and I-5) that cross and flank the array. These spurious phases are incompatible with interpreting the cross-correlations as Green’s functions. We experiment with multiple approaches to suppressing these spurious phases, including ML-based Gaussian mixture modeling, but find that these methods do not eliminate the undesirable signals. By applying one-bit normalization in both the spectral and time domains, we are able to suppress these spurious signals.
We produce cross-correlation functions for all station pairs and all 9 components of the Green’s tensor, and use these to perform Frequency-Time Analysis (FTAN) of the ambient noise field recorded by the nodal array. We find that the average seismic velocities in the upper kilometer range from 400 and 900 m/s, relatively high compared to other urban sedimentary basins (e.g., Tokyo and Los Angeles), which could be due to the relatively high compaction of glacial tills present in Seattle. We produce dispersion curves and observe that higher modes are excited in surface waves across the array; these travel with about twice the wavespeed of the fundamental mode. These preliminary results will be applied in future imaging studies.
Lastly, we characterize the urban hum that defines ambient noise in the city of Seattle. We produce full-month spectrograms, which resolve temporal patterns of the local ambient wavefield. All stations’ spectrograms show strong diurnal patterns, with noise power picking up daily at 5 AM PT and quieting every evening. Noise power levels on weekends are consistently lower than the five days of the work week. Noise power peaks within the 3-20 Hz range. These observations highlight the dominance of anthropogenic sources in the high frequency wavefield.
Overall, our results reveal nuanced interplay between anthropogenic noise, basin-edge effects, shallow site effects, and influences of fault zone structure on the radiated seismic wavefield. Our results have useful applications for understanding how the SFZ and Seattle basin edge scatter and diffract earthquake ground motions, which is an important consideration for hazard assessment for future potential SFZ earthquakes.Earth and Planetary Science
Queer Death Cult
No full textThis thesis is a fictional novel – or at least a portion of it – that aims to represent a
unique queer internet sub-culture often under-represented in modern transsexual art: that
of the online spaces harboured on the /lgbt/ 4chan board and affiliated Discord servers.
Engagement with such spaces cultivates a peculiar mentality within their users and
affects their internalisation of the transsexual experience. With this manuscript my aim is
to form an autoethnography of sorts, one that can archive this sub-culture from a
particular point in time and present it to an unfamiliar reader through an engaging text.
The writing is comprised of two distinct elements. The first is that of the general
manuscript: a paperback horror novel written by a fictitious author named ‘Alice’, her
work of autofiction centering around the damaging psychological influences 4chan /lgbt/
culture has on her and her online friends. The second: fabricated marginalia around the
text at the hands of an imagined editor whom ‘Alice’ has sent this draft of her manuscript
to. As the editor’s notes progress they evolve from standard editing notation towards a
one-sided dialogue aimed at ‘Alice’, guiding her to a less fractured worldview from the
perspective of a trans person who had previously challenged and overcome her own
internalised transphobia.Extension Studie
A γδ T cell–IL-3 axis controls allergic responses through sensory neurons
No full textIn naïve individuals, sensory neurons directly detect and respond to allergens, leading both to the sensation of itch and the activation of local innate immune cells, which initiate the allergic immune response. In the setting of chronic allergic inflammation, immune factors prime sensory neurons, causing pathologic itch. While these bidirectional neuroimmune circuits drive responses to allergens, whether immune cells regulate the set-point for neuronal activation by allergens in the naïve state is unknown. Here we describe a gd T cell-IL-3 signaling axis that controls the allergen responsiveness of cutaneous sensory neurons. We define a poorly characterized epidermal gd T cell subset, termed GD3 cells, that produces its hallmark cytokine IL-3 to promote allergic itch and the initiation of the allergic immune response. Mechanistically, IL-3 acts on Il3ra-expressing sensory neurons in a JAK2-dependent manner to lower their threshold for allergen activation without independently eliciting itch. This gd T cell-IL-3 signaling axis further acts via STAT5 to promote neuropeptide production and the initiation of allergic immunity. These results reveal an endogenous immune rheostat that sits upstream of and governs sensory neuronal responses to allergens upon first exposure. This pathway may explain individual differences in allergic susceptibility and opens novel therapeutic avenues for treating allergic diseases.Accepted Manuscrip
How Pain Came to Matter: The Global Political Economy of the United States Opioid Crisis
No full textAuthor's Origina
MEDUSA for Identifying Death Regulatory Genes in Chemo-genetic Profiling Data
No full textThis publication requires a 6 month embargo period.Accepted Manuscrip
Quantum States and Spectra of Small Cylindrical and Toroidal Lattices
No full textCaelan Brooks is affiliated with Harvard University as a Physics PhD candidate, Kunal K. Das is unaffiliated with Harvard.We examine the spectrum and quantum states of small lattices with cylindrical and toroidal topology subject to a scalar gauge potential that introduces a position dependent phase in the inter-site coupling. Equivalency of gauges assumed in infinite lattices is generally lost due to the periodic boundary conditions, and conditions that restore it are identified. We trace the impact of various system parameters including gauge choice, boundary conditions and inter-site coupling strengths, and an additional axial field. We find gauge dependent appearance of avoided crossings and persistent degeneracies, and we show their impact on the associated eigenstates. Smaller lattices develop prominent gaps in spectral lines associated with edge states, which are suppressed in the thermodynamic limit. Toroidal lattices have counterparts of most of the features observed in cylindrical lattices, but notably they display a transition from localization to delocalization determined by the relation between the field parameter and the number of lattice sites.PhysicsProo
