Our goal is to develop materials and physical systems that learn on their own desired properties or precise responses. In contrast to designed matter which requires fabrication, here a generic material is trained through externally applied fields and therefore relinquishes the need to manipulate the microstructure. As a result, our approach could be highly scalable, allowing to control systems with an enormous number of degrees of freedom. We are developing training (learning) algorithms that enable a wide variety of responses, including global material properties, as well as spatially varying operating both in the quasi-static and dynamical regimes. Our work could also enable adaptive materials whose function can be altered at will, as well as intelligent materials that are able to perform computations which could be useful for neuromorphic computing.
