Grant Rotskoff*

I will describe a theoretical framework for controlling the response and dynamics of nonequilibrium systems using time-dependent external couplings. These external "control protocols" satisfy a number of distinct variational principles, which become natural machine learning objective functions. The variational principles not only constrain, for example, the finite-time energetic efficiency of a given transformation, but also provide a natural objective for optimization. Finally, I will show how to solve these problems computationally for some model systems of interest, including models of experimental active matter and self-assembling actin networks.