While neural networks can help to improve the accuracy of fluid flow simulations, new research shows how their accuracy is limited unless the right approach is taken. By embedding fluid properties into neural networks, simulation accuracy can improve by orders of magnitude.
