Theorie-Seminar

Dean Lee (MSU)
Parametric Matrix Models

I give an introduction to a new machine learning approach called parametric matrix models, which are based on the matrix equations of quantum physics rather than the biology of neurons. Rather than fitting output functions according to some specified form, PMMs learn the underlying equations that produce the desired output, similar
to how physics problems are solved. I discuss the connection to eigenvector continuation and reduced basis methods, show the proof of the universal function approximation theorem, and then go through several applications to scientific computing as well as more general machine learning applications related to image recognition.