An official website of the United States government
Abstract BackgroundMeasuring parathyroid hormone-related peptide (PTHrP) helps diagnose the humoral hypercalcemia of malignancy, but is often ordered for patients with low pretest probability, resulting in poor test utilization. Manual review of results to identify inappropriate PTHrP orders is a cumbersome process. MethodsUsing a dataset of 1330 patients from a single institute, we developed a machine learning (ML) model to predict abnormal PTHrP results. We then evaluated the performance of the model on two external datasets. Different strategies (model transporting, retraining, rebuilding, and fine-tuning) were investigated to improve model generalizability. Maximum mean discrepancy (MMD) was adopted to quantify the shift of data distributions across different datasets. ResultsThe model achieved an area under the receiver operating characteristic curve (AUROC) of 0.936, and a specificity of 0.842 at 0.900 sensitivity in the development cohort. Directly transporting this model to two external datasets resulted in a deterioration of AUROC to 0.838 and 0.737, with the latter having a larger MMD corresponding to a greater data shift compared to the original dataset. Model rebuilding using site-specific data improved AUROC to 0.891 and 0.837 on the two sites, respectively. When external data is insufficient for retraining, a fine-tuning strategy also improved model utility. ConclusionsML offers promise to improve PTHrP test utilization while relieving the burden of manual review. Transporting a ready-made model to external datasets may lead to performance deterioration due to data distribution shift. Model retraining or rebuilding could improve generalizability when there are enough data, and model fine-tuning may be favorable when site-specific data is limited.
more »
« less
A machine learning approach to robustly determine director fields and analyze defects in active nematics
A machine learning model for reliable director fields calculation from raw experimental images of active nematics. The model is accurate, robust to noise and generalizable, enhancing analysis such as the detection and tracking of topological defects.
more »
« less
- Award ID(s):
- 2011846
- PAR ID:
- 10503821
- Publisher / Repository:
- Royal Society of Chemistry
- Date Published:
- Journal Name:
- Soft Matter
- Volume:
- 20
- Issue:
- 8
- ISSN:
- 1744-683X
- Page Range / eLocation ID:
- 1869 to 1883
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Transfer learning is an effective technique for tuning a deep learning model when training data or computational resources are limited. Instead of training a new model from scratch, the parameters of an existing base model are adjusted for the new task. The accuracy of such a fine-tuned model depends on the suitability of the base model chosen. Model search automates the selection of such a base model by evaluating the suitability of candidate models for a specific task. This entails inference with each candidate model on task-specific data. With thousands of models available through model stores, the computational cost of model search is a major bottleneck for efficient transfer learning. In this work, we presentAlsatian, a novel model search system. Based on the observation that many candidate models overlap to a significant extent and following a careful bottleneck analysis, we propose optimization techniques that are applicable to many model search frameworks. These optimizations include: (i) splitting models into individual blocks that can be shared across models, (ii) caching of intermediate inference results and model blocks, and (iii) selecting a beneficial search order for models to maximize sharing of cached results. In our evaluation on state-of-the-art deep learning models from computer vision and natural language processing, we show thatAlsatianoutperforms baselines by up to 14x.more » « less
-
Summary Some rhizomatous grass and sedge species form tussocks that impact ecosystem structure and function. Despite their importance, tussock development and size controls are poorly understood due to the decadal to centennial timescales over which tussocks form.We explored mechanisms regulating tussock development and size in a ubiquitous arctic tussock sedge (Eriophorum vaginatum) using field observations and a mass balance model coupled with a tiller population model. Model–data fusion was used to quantify parameter and prediction uncertainty, determine model sensitivity, and test hypotheses on the factors regulating tussock size.The model accurately captured the dynamics of tussock development, characteristics, and size observed in the field. Tussock growth approached maximal size within several decades, which was determined by feedbacks between the mass balance of tussock root necromass and density‐dependent tillering. The model also predicted that maximal tussock size was primarily regulated by tiller root productivity and necromass bulk density and less so by tiller demography. These predictions were corroborated by field observations of tussock biomass and root characteristics.The study highlights the importance of belowground processes in regulating tussock development and size and enhances our understanding of the influence of tussocks on arctic ecosystem structure and function.more » « less
-
Abstract Minimum central pressure (Pmin) is an integrated measure of the tropical cyclone wind field and is known to be a useful indicator of storm damage potential. A simple model that predictsPminfrom routinely estimated quantities, including storm size, would be of great value. Here, we present a simple linear empirical model for predictingPminfrom maximum wind speed, a radius of 34-kt (1 kt ≈ 0.51 m s−1) winds (R34kt), storm center latitude, and the environmental pressure. An empirical model for the pressure deficit is first developed that takes as predictors specific combinations of these quantities that are derived directly from theory based on gradient wind balance and a modified Rankine-type wind profile known to capture storm structure inside ofR34kt. Model coefficients are estimated using data from the southwestern North Atlantic and eastern North Pacific from 2004 to 2022 using aircraft-based estimates ofPmin, extended best track data, and estimates of environmental pressure from Global Forecast System (GFS) analyses. The model has a near-zero conditional bias even for lowPmin, explaining 94.2% of the variance. Performance is superior to a variety of other model formulations, including a standard wind–pressure model that does not account for storm size or latitude (89.2% variance explained). Model performance is also strong when applied to high-latitude data and data near coastlines. Finally, the model is shown to perform comparably well in an operation-like setting based solely on routinely estimated variables, including the pressure of the outermost closed isobar. Case study applications to five impactful historical storms are discussed. Overall, the model offers a simple, fast, physically based prediction forPminfor practical use in operations and research. Significance StatementSea level pressure is lowest at the center of a hurricane and is routinely estimated in operational forecasting along with the maximum wind speed. While the latter is currently used to define hurricane intensity, the minimum pressure is also a viable measure of storm intensity that is known to better represent damage risk. A simple empirical model that predicts the minimum pressure from maximum wind speed and size, and based on the physics of the hurricane wind field, does not currently exist. This work develops such a model by using wind field physics to determine the important parameters and then uses a simple statistical model to make the final prediction. This model is quick and easy to use in weather forecasting and risk assessment applications.more » « less
-
Abstract BackgroundThis study posits that scaffolded team-based computational modeling and simulation projects can support model-based learning that can result in evidence of representational competence and regulatory skills. The study involved 116 students from a second-year thermodynamics undergraduate course organized into 24 teams, who worked on three two-week-long team-based computational modeling and simulation projects and reflected upon their experience. ResultsResults characterized different levels of engagement with computational model-based learning in the form of problem formulation and model planning, implementation and use of the computational model, evaluation, and interpretation of the outputs of the model, as well as reflection on the process. Results report on students’ levels of representational competence as related to the computational model, meaning-making of the underlying code of the computational model, graphical representations generated by the model, and explanations and interpretations of the output representations. Results also described regulatory skills as challenges and strategies related to programming skills, challenges and strategies related to meaning-making skills for understanding and connecting the science to the code and the results, and challenges and strategies related to process management mainly focused on project management skills. ConclusionCharacterizing dimensions of computational model-based reasoning provides insights that showcase students’ learning, benefits, and challenges when engaging in team-based computational modeling and simulation projects. This study also contributes to evidence-based scaffolding strategies that can support undergraduate students' engagement in the context of computational modeling and simulation.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1039/d3sm01253k
