Well they memorize points and lines (or tanh) between different parts of the space right? So it depends on whether a useful generalization can be extracted from

Well they memorize points and lines (or tanh) between different parts of the space right? So it depends on whether a useful generalization can be extracted from the line estimation and how dense the points on the landscape are no?