Search for: All records

Sparse regression and feature extraction are the cornerstones of knowledge discovery from massive data. Their goal is to discover interpretable and predictive models that provide simple relationships among scientific variables. While the statistical tools for model discovery are well established in the context of linear regression, their generalization to nonlinear regression in material modeling is highly problem‐specific and insufficiently understood. Here we explore the potential of neural networks for automatic model discovery and induce sparsity by a hybrid approach that combines two strategies: regularization and physical constraints. We integrate the concept of Lp regularization for subset selection with constitutive neural networks that leverage our domain knowledge in kinematics and thermodynamics. We train our networks with both, synthetic and real data, and perform several thousand discovery runs to infer common guidelines and trends: L2 regularization or ridge regression is unsuitable for model discovery; L1 regularization or lasso promotes sparsity, but induces strong bias that may aggressively change the results; only L0 regularization allows us to transparently fine‐tune the trade‐off between interpretability and predictability, simplicity and accuracy, and bias and variance. With these insights, we demonstrate that Lp regularized constitutive neural networks can simultaneously discover both, interpretable models and physically meaningful parameters. We anticipate that our findings will generalize to alternative discovery techniques such as sparse and symbolic regression, and to other domains such as biology, chemistry, or medicine. Our ability to automatically discover material models from data could have tremendous applications in generative material design and open new opportunities to manipulate matter, alter properties of existing materials, and discover new materials with user‐defined properties. 

The lack of sex-specific cardiovascular disease criteria contributes to the underdiagnosis of women compared to that of men. For more than half a century, the Framingham Risk Score has been the gold standard to estimate an individual’s risk of developing cardiovascular disease based on the age, sex, cholesterol levels, blood pressure, diabetes status, and the smoking status. Now, machine learning can offer a much more nuanced insight into predicting the risk of cardiovascular diseases. The UK Biobank is a large database that includes traditional risk factors and tests related to the cardiovascular system: magnetic resonance imaging, pulse wave analysis, electrocardiograms, and carotid ultrasounds. Here, we leverage 20,542 datasets from the UK Biobank to build more accurate cardiovascular risk models than the Framingham Risk Score and quantify the underdiagnosis of women compared to that of men. Strikingly, for a first-degree atrioventricular block and dilated cardiomyopathy, two conditions with non-sex-specific diagnostic criteria, our study shows that women are under-diagnosed 2× and 1.4× more than men. Similarly, our results demonstrate the need for sex-specific criteria in essential primary hypertension and hypertrophic cardiomyopathy. Our feature importance analysis reveals that out of the top 10 features across three sexes and four disease categories, traditional Framingham factors made up between 40% and 50%; electrocardiogram, 30%–33%; pulse wave analysis, 13%–23%; and magnetic resonance imaging and carotid ultrasound, 0%–10%. Improving the Framingham Risk Score by leveraging big data and machine learning allows us to incorporate a wider range of biomedical data and prediction features, enhance personalization and accuracy, and continuously integrate new data and knowledge, with the ultimate goal to improve accurate prediction, early detection, and early intervention in cardiovascular disease management. Our analysis pipeline and trained classifiers are freely available at https://github.com/LivingMatterLab/CardiovascularDiseaseClassification.