Using very long baseline interferometry, the Event Horizon Telescope (EHT) collaboration has resolved the shadows of two supermassive black holes. Model comparison is traditionally performed in image space, where imaging algorithms introduce uncertainties in the recovered structure. Here, we develop a deep learning framework to perform parameter inference in visibility space, directly using the data measured by the interferometer without introducing potential errors and biases from image recon- struction. First, we train and validate our framework on synthetic data derived from general relativistic magnetohydrodynamics (GRMHD) simulations that vary in magnetic field state, spin, and 𝑅high. Applying these models to the real data obtained during the 2017 EHT campaign, and only considering total intensity, we do not derive meaningful constraints on either of these parameters. At present, our method is limited both by theoretical uncertainties in the GRMHD simulations and variation between snapshots of the same underlying physical model. However, we demonstrate that spin and 𝑅high could be recovered using this framework through continuous monitoring of our sources, which mitigates variations due to turbulence. In future work, we anticipate that including spectral or polarimetric information will greatly improve the performance of this framework.
The novel coronavirus (COVID-19) pandemic has had a devastating global impact, profoundly affecting daily life, healthcare systems, and public health infrastructure. Despite the availability of treatments and vaccines, hospitalizations and deaths continue. Real-time surveillance of infection trends supports resource allocation and mitigation strategies, but reliable forecasting remains a challenge. While deep learning has advanced time-series forecasting, its effectiveness relies on large datasets, a significant obstacle given the pandemic’s evolving nature. Most models use national or state-level data, limiting both dataset size and the granularity of insights. To address this, we propose the Fine-Grained Infection Forecast Network (FIGI-Net), a stacked bidirectional LSTM structure designed to leverage county-level data to produce daily forecasts up to two weeks in advance. FIGI-Net outperforms existing models, accurately predicting sudden changes such as new outbreaks or peaks, a capability many state-of-the-art models lack. This approach could enhance public health responses and outbreak preparedness.
Google Trends reports how frequently specific queries are searched on Google over time. It is widely used in research and industry to gain early insights into public interest. However, its data generation mechanism introduces missing values, sampling variability, noise, and trends. These issues arise from privacy thresholds mapping low search volumes to zeros, daily sampling variations causing discrepancies across historical downloads, and algorithm updates altering volume magnitudes over time. Data quality has recently deteriorated, with more zeros and noise, even for previously stable queries. We propose a comprehensive statistical methodology to preprocess Google Trends search information using hierarchical clustering, smoothing splines, and detrending. We validate our approach by forecasting U.S. influenza hospitalizations with a univariate ARIMAX model. Compared to omitting exogenous variables, our results show that raw Google Trends data degrades modeling performance, while preprocessed signals enhance forecast accuracy by 58% nationally and 24% at the state level.
The global shipping network, which moves over 80% of the world’s goods, is not only a vital backbone of the global economy but also one of the most polluting industries. Studying how this network operates is crucial for improving its efficiency and sustainability. While the transport of solid goods like packaged products and raw materials has been extensively researched, far less is known about the competitive trade of crude oil and petroleum, despite these commodities accounting for nearly 30% of the market. Using 4 years of high-resolution data on oil tanker movements, we employ sequential motif mining and dynamic mode decomposition to uncover global spatio-temporal patterns in the movement of individual ships. Across all ship classes, we demonstrate that maximizing the proportion of time ships spend carrying cargo –a metric of efficiency– is achieved through strategic diversification of routes and the effective use of intra-regional ports for trips without cargo. Moreover, we uncover a globally stable travel structure in the fleet, with pronounced seasonal variations linked to annual and semi-annual regional climate patterns and economic cycles. Our findings highlight the importance of integrating high-resolution data with innovative analysis methods not only to improve our understanding of the underlying dynamics of shipping patterns, but to design and evaluate strategies aimed at reducing their environmental impact.
Accurate, real-time forecasts of influenza hospitalizations would facilitate prospective resource allocation and public health preparedness. State-of-the-art machine learning methods are a promising approach to produce such forecasts, but they require extensive historical data to be properly trained. Unfortunately, data on influenza hospitalizations, for the 50 states in the United States, are only available since the beginning of 2020. In addition, the data are far from perfect as they were under-reported for several months before health systems began consistently submitting their data. To address these issues, we propose a transfer learning approach. We extend the currently available two-season dataset for state-level influenza hospitalizations by an additional ten seasons. Our method leverages influenza-like illness (ILI) data to infer historical estimates of influenza hospitalizations. This data augmentation enables the implementation of advanced machine learning techniques, multi-horizon training, and an ensemble of models to improve hospitalization forecasts. We evaluated the performance of our machine learning approaches by prospectively producing forecasts for future weeks and submitting them in real time to the Centers for Disease Control and Prevention FluSight challenges during two seasons: 2022-2023 and 2023-2024. Our methodology demonstrated good accuracy and reliability, achieving a fourth place finish (among 20 participating teams) in the 2022-23 and a second place finish (among 20 participating teams) in the 2023-24 CDC FluSight challenges. Our findings highlight the utility of data augmentation and knowledge transfer in the application of machine learning models to public health surveillance where only limited historical data is available.
Background: The novel coronavirus disease (COVID-19) sparked significant health concerns worldwide, prompting policy makers and health care experts to implement nonpharmaceutical public health interventions, such as stay-at-home orders and mask mandates, to slow the spread of the virus. While these interventions proved essential in controlling transmission, they also caused substantial economic and societal costs and should therefore be used strategically, particularly when disease activity is on the rise. In this context, geosocial media posts (posts with an explicit georeference) have been shown to provide a promising tool for anticipating moments of potential health care crises. However, previous studies on the early warning capabilities of geosocial media data have largely been constrained by coarse spatial resolutions or short temporal scopes, with limited understanding of how local political beliefs may influence these capabilities. Objective: This study aimed to assess how the epidemiological early warning capabilities of geosocial media posts for COVID-19 vary over time and across US counties with differing political beliefs. Methods: We classified US counties into 3 political clusters, democrat, republican, and swing counties, based on voting data from the last 6 federal election cycles. In these clusters, we analyzed the early warning capabilities of geosocial media posts across 6 consecutive COVID-19 waves (February 2020-April 2022). We specifically examined the temporal lag between geosocial media signals and surges in COVID-19 cases, measuring both the number of days by which the geosocial media signals preceded the surges in COVID-19 cases (temporal lag) and the correlation between their respective time series. Results: The early warning capabilities of geosocial media data differed across political clusters and COVID-19 waves. On average, geosocial media posts preceded COVID-19 cases by 21 days in republican counties compared with 14.6 days in democrat counties and 24.2 days in swing counties. In general, geosocial media posts were preceding COVID-19 cases in 5 out of 6 waves across all political clusters. However, we observed a decrease over time in the number of days that posts preceded COVID-19 cases, particularly in democrat and republican counties. Furthermore, a decline in signal strength and the impact of trending topics presented challenges for the reliability of the early warning signals. Conclusions: This study provides valuable insights into the strengths and limitations of geosocial media data as an epidemiological early warning tool, particularly highlighting how they can change across county-level political clusters. Thus, these findings indicate that future geosocial media based epidemiological early warning systems might benefit from accounting for political beliefs. In addition, the impact of declining geosocial media signal strength over time and the role of trending topics for signal reliability in early warning systems need to be assessed in future research.
Irritability and Social Media Use in US Adults Roy H. Perlis, MD, MSc; Ata Uslu, MS; Jonathan Schulman, MS; Faith M. Gunning, PhD; Mauricio Santillana, PhD; Matthew A. Baum, PhD; James N. Druckman, PhD; Katherine Ognyanova, PhD; David Lazer, PhD JAMA Netw Open. 2025;8(1):e2452807. doi:10.1001/jamanetworkopen.2024.52807
Abstract
ImportanceEfforts to understand the complex association between social media use and mental health have focused on depression, with little investigation of other forms of negative affect, such as irritability and anxiety. ObjectiveTo characterize the association between self-reported use of individual social media platforms and irritability among US adults. Design, Setting, and ParticipantsThis survey study analyzed data from 2 waves of the COVID States Project, a nonprobability web-based survey conducted between November 2, 2023, and January 8, 2024, and applied multiple linear regression models to estimate associations with irritability. Survey respondents were aged 18 years and older. ExposureSelf-reported social media use. Main Outcomes and MeasuresThe primary outcome was score on the Brief Irritability Test (range, 5-30), with higher scores indicating greater irritability. ResultsAcross the 2 survey waves, there were 42 597 unique participants, with mean (SD) age 46.0 (17.0) years; 24 919 (58.5%) identified as women, 17 222 (40.4%) as men, and 456 (1.1%) as nonbinary. In the full sample, 1216 (2.9%) identified as Asian American, 5939 (13.9%) as Black, 5322 (12.5%) as Hispanic, 624 (1.5%) as Native American, 515 (1.2%) as Pacific Islander, 28 354 (66.6%) as White, and 627 (1.5%) as other (ie, selecting the other option prompted the opportunity to provide a free-text self-description). In total, 33 325 (78.2%) of the survey respondents reported daily use of at least 1 social media platform, including 6037 (14.2%) using once a day, 16 678 (39.2%) using multiple times a day, and 10 610 (24.9%) using most of the day. Frequent use of social media was associated with significantly greater irritability in univariate regression models (for more than once a day vs never, 1.43 points [95% CI, 1.22-1.63 points]; for most of the day vs never, 3.37 points [95% CI, 3.15-3.60 points]) and adjusted models (for more than once a day, 0.38 points [95% CI, 0.18-0.58 points]; for most of the day, 1.55 points [95% CI, 1.32-1.78 points]). These associations persisted after incorporating measures of political engagement. Conclusions and RelevanceIn this survey study of 42 597 US adults, irritability represented another correlate of social media use that merits further characterization, in light of known associations with depression and suicidality.