Machine learning has gained widespread popularity in fields such as research, finance, behaviorism, marketing, economics, and communications, where large volumes of training data are available and where an underlying model or system may not be readily accessible or easily parameterized. On the other hand, breakthroughs into scientific applications where underlying physical models exist and are well understood are harder to find. This is presumably because the physics-based solution, while imperfect, exploits the intrinsic relation between inputs and outputs and can often outperform a purely data-driven solution. Furthermore, adoption of machine learning may encounter resistance because these new algorithms are often opaque, and it is not always clear under what circumstances the methods will fail.

You do not have access to this content, please speak to your institutional administrator if you feel you should have access.